<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Fenbushi Capital | 分布式资本 on Medium]]></title>
        <description><![CDATA[Stories by Fenbushi Capital | 分布式资本 on Medium]]></description>
        <link>https://medium.com/@fenbushicapital?source=rss-7c9ec2c710ab------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*lJdoDrRhZ43OhVjbVeRa4A.png</url>
            <title>Stories by Fenbushi Capital | 分布式资本 on Medium</title>
            <link>https://medium.com/@fenbushicapital?source=rss-7c9ec2c710ab------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 23:42:41 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@fenbushicapital/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[An Introduction to Indistinguishability Obfuscation]]></title>
            <link>https://fenbushicapital.medium.com/an-introduction-to-indistinguishability-obfuscation-e00bf07b95fe?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/e00bf07b95fe</guid>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[cryptography]]></category>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Thu, 18 Dec 2025 20:02:38 GMT</pubDate>
            <atom:updated>2025-12-18T20:02:38.318Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u6bdtYNNhf838Y9i-gs5Yw.jpeg" /></figure><p>Special thanks to Sora Suegami for feedback and source material.</p><p>Indistinguishability obfuscation (iO) has been buzzing through Ethereum circles as a “holy grail” of cryptography, a crown once worn by ZK and FHE. The concept has a turbulent history, from early impossibility results to fragile breakthroughs to today’s “Diamond iO” benchmarks. At its core, iO is the simple idea that code can be made runnable but unreadable. Of course, it’s still wildly impractical. If it ever matures, iO could transform blockchains: powering private governance, confidential markets, and tamper-proof infrastructure. This blog post walks through the history, the benchmarks, and what iO might mean on-chain.</p><h3>iO: A Book and a Maze</h3><p>Imagine a world-class chef. Her genius lies not just in the dishes she makes, but in how she makes them: the subtle variations, the order of steps, the tricks that she’s refined over her lifetime. She wants you to be able to cook her food at home, but doesn’t want you to learn her secrets.</p><p>So she gives you a magic recipe book. You open it, follow the instructions step by step, and out comes a perfect replica of the signature dish. However, the recipe is written in magic ink. You’re able to follow the instructions and use her secrets to cook the meal as if you were the chef herself, but you never learn what the secret to making it her way actually was.</p><p>This highlights the <em>obfuscation </em>of iO: a cryptographic primitive that allows software to be published in a way that anyone can run it, but no one can reverse-engineer. Due to its power and what can be derived from it, iO is a “holy grail” of cryptography and web3: developers can publish obfuscated programs that users can interact with freely, that any node can run, but that yields no insight to the code that is actually being run, <em>even</em> <em>if there are secret keys hidden in the code</em>.</p><p>Now think of two mazes: one is a straight line from start to finish, the other is a labyrinth of twists and turns. Both mazes have the same entrance and the same exit. iO applies a magic fog to the mazes: you’re able to walk from start to finish, but you have no idea which maze you’ve traversed.</p><p>This is the <em>indistinguishability </em>trait of iO: two different programs that compute the same function become indistinguishable once obfuscated. If the entrance is your input and the exit is the output of the program, you will get to the correct output without being able to deduce anything meaningful from the intermediate steps.</p><p><strong>The definition of iO: </strong>Given two different programs that computes the same function (in other words, both programs will give the same output for any input), once obfuscated, you can’t tell which program your obfuscation corresponds to. An <em>obfuscator</em> takes in a program and produces an obfuscated binary can then be run on any computer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ET7JHhPfoh5xExw5xDw3PQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*45OxQvS7qANENj9QksjwsA.png" /><figcaption>iO’s ideal functionality</figcaption></figure><p>iO unlocks a direct path to both privacy and verifiability: confidential auctions, private voting, or contracts can be naively built for correctness, and once obfuscated, be private as well. Any validator or node can run the program, but won’t be able to see the inner workings of it: no TEEs or special hardware needed.</p><p>The road to iO wasn’t straightforward. In 2001, a seminal paper — <a href="https://www.iacr.org/archive/crypto2001/21390001.pdf">On the (Im)possibility of Obfuscating Programs</a> — showed that perfect “black-box” obfuscation is impossible. Some mazes can never be fogged without leaking their structure. For a while after this paper was published, it seemed like the dream of meaningful obfuscation was dead.</p><p>In 2013 however, cryptographers introduced indistinguishability obfuscation (iO) — a weaker and far more nuanced approach. It didn’t try to fog all mazes; it only required that if two mazes have the same entrance and exit, then once obfuscated, they look the same.</p><p>The realization of iO’s usefulness in 2013 opened the floodgates. Even though early candidate constructions were impractical, iO revealed itself as a foundational primitive. Complex cryptographic systems can be built almost trivially by mixing iO with other primitives. In fact, several primitives themselves can be made from iO as well, and it has since been found that much of modern cryptography can be built from iO.</p><h4><strong>A Locksmith for Cryptography</strong></h4><p>Cryptography has equips defenders with a growing keyring of specialized tools, or cryptographic primitives. One key unlocks digital signatures, another zero-knowledge proofs, another multiparty computation, yet another functional encryption. Each key is tailored to its lock, and defenders must carefully choose and wield them against attackers.</p><p>iO brings a primitive that can act as a locksmith: much (although <a href="https://eprint.iacr.org/2015/341.pdf">not all</a>) of today’s cryptography can be made from iO. With iO and basic assumptions, much of modern cryptography can be derived: public key cryptography, signatures, NIZKs, functional encryption, attribute-based encryption, even FHE can be constructed from iO. If quick and efficient iO can be achieved, the hope is that anything that can be made from iO (such as FHE) would also be quick and efficient.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GPp-oepTvPjStYdq-DQn5A.png" /><figcaption>The use cases of different cryptographic primitives</figcaption></figure><h3>A History</h3><h4>iO Exists (2000–2001)</h4><p>At the turn of the millennium, researchers asked a natural question: could software be made <em>unreadable yet runnable</em>? In 2000, early work laid out formal definitions of what an “obfuscator” should achieve. However, in 2001, Barak, Goldwasser, Wigderson, and others reached a result which proved that <em>virtual black-box (VBB) obfuscation is impossible</em>. Certain programs, they showed, can never be obfuscated without leaking secrets. iO was defined as a by-product of the impossibility, but it wasn’t seen as powerful or interesting in its own right.</p><p>The impossibility result was a shock — and for more than a decade, obfuscation seemed like a dead end.</p><h4><strong>iO Winter (2001–2013)</strong></h4><p>After the impossibility result, the field went quiet. Program obfuscation was seen as a curiosity, not a research frontier. For twelve years, cryptographers pursued other directions: lattices, homomorphic encryption, multiparty computation. Obfuscation was dismissed as theoretically alluring, but fundamentally out of reach.</p><h4>A New Hope (2013)</h4><p>More than a decade after the impossibility result, Garg, Gentry, Sahai, and Waters introduced <a href="https://eprint.iacr.org/2013/451.pdf">the first candidate construction of indistinguishability obfuscation (iO)</a>. Their breakthrough used a then-new algebraic tool called <em>multilinear maps</em> and a related primitive, <em>functional encryption</em>.</p><p>The key idea was to show an iO candidate construction from multilinear maps. The paper revived the field of program obfuscation after a decade of impossibility results and demonstrated that iO could be extremely powerful if efficiently realized.</p><h4>iO Breakthroughs and Attacks (2013–202)</h4><p>Indistinguishability obfuscation (iO) had leapt from dream to reality. Researchers quickly realized it was not just another primitive: it was a useful for almost all cryptographic constructs. From iO, one could construct almost everything: fully homomorphic encryption, zero-knowledge proofs, secure multiparty computation, even primitives once thought unrelated. For a time, iO looked like a single foundation stone upon which modern cryptography could be rebuilt.</p><p>But the excitement came with fragility. These first constructions relied on powerful but delicate new algebraic objects called multilinear maps. Between 2013 and 2017, a wave of attacks exposed cracks in these foundations. By the end of this era, the cryptographic community faced a paradox: iO was unimaginably powerful in theory, but its concrete instantiations were flawed.</p><p>The fragilities were uncovered and exploited. <a href="https://eprint.iacr.org/2014/906.pdf">The first zeroing attacks</a> were unveiled in 2014, breaking leading iO candidates. Other works catalogued the growing list of broken assumptions. The field had reached a crossroads: if iO was real, it could not rest on multilinear maps.</p><p>Broadly, the exotic assumptions that constructions were using were unreliable. Until iO could be built from standard cryptographic assumptions, it was clear that achieving robustness would be an additional hurdle to overcome. Researchers began searching for <a href="https://eprint.iacr.org/2019/643.pdf">solid assumptions from which to build iO</a>, which culminated in a breakthrough in 2020.</p><h3>Towards Solid Ground (2020–Today)</h3><p>A major breakthrough arrived in 2020, when Jain, Lin, and Sahai proposed <a href="https://eprint.iacr.org/2020/1003.pdf">the first iO construction from standard, broadly trusted assumptions</a>: lattice problems (LWE), subgroup-based assumptions (SXDH), constant-depth pseudorandom generators, and learning with parity noise (LPN). This was the turning point: iO no longer lived only on shaky, uncertain assumptions.</p><p>In 2025, another milestone arrived: <a href="https://eprint.iacr.org/2025/236">Diamond iO</a>, which further streamlined the foundations by avoiding functional encryption bootstrapping. These advances remain far from practical — current iO schemes are astronomically inefficient. But now for the first time, cryptographers can seriously imagine iO built on bedrock, and can get a taste for the first obfuscated programs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W23fCid1164kGtJTYOlWIw.png" /><figcaption>Important papers and achievements in iO</figcaption></figure><h3>The iO Chasm: Why it’s not here today</h3><p>The biggest (practical) bottleneck to cryptography is typically computational complexity. If an insecure computation, say a multiplication, takes one clock cycle to compute on a modern CPU, how many would it take to compute the same operation in FHE? In a TEE? Using ZK? Early FHE schemes used to take minutes to do these operations that a computer would do in nanoseconds. Now, advancements have brought this “minutes per operation” overhead to milliseconds per operation.</p><p>ZK proofs had a similar lifecycle: once it was proved that ZK proofs could be generated for any (NP) computation, the race was on to make the proving and verifying practical for a personal computer to perform. Proving still is a computationally intense task: a zkP2P proof can take on the order of 5 minutes to generate on a mobile device, but this kind of performance is good enough for on-chain nodes. Despite ZK still being too slow for consumer applications, it is good enough for on-chain use.</p><p>iO has two stages to benchmark: <strong>obfuscation</strong> and <strong>evaluation</strong>. Obfuscation takes in a program, which may include a secret like a cryptographic key, resulting in an executable that looks unintelligible but preserves the functionality of the program. Evaluation simply takes the obfuscated program and evaluates it on given inputs. The evaluation will leave the evaluator with the same result as the unobfuscated program, without allowing the evaluator to learn anything about how the original program was implemented.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zREhpZMUKcRvF-EkbISkFg.png" /><figcaption>The stages of obfuscation</figcaption></figure><p>In <a href="https://mediatum.ub.tum.de/doc/1246288/1246288.pdf">a 2022 paper</a>, researchers from the Technical University of Munich set out to benchmark Garg et al.’s construction from 2013. Take a simple 2-bit multiplication circuit:</p><ul><li>Can multiply two numbers between 0 and 3</li><li>4 bits of inputs (2-bits per operand)</li><li>4 output bits</li><li>1–8 AND gates per output bit (4 to 32 gates total)</li></ul><p>Any program that is used today can use upwards of billions to trillions of multiplications, all with 64-bit operands (not 2). The researchers found that to obfuscate the 2-bit gate above would take:</p><ul><li>10²⁷ years to obfuscate (the universe has only existed for 10¹⁰, or ten billion years)</li><li>20 zettabytes (200 trillion gigabytes)</li><li>10⁸ years to evaluate (if the universe came with an obfuscated 2-bit multiplication gate built-in and we started evaluating multiplications at the beginning of the universe, we would have only performed 100 2-bit multiplications)</li></ul><h3>The Diamond iO Benchmark</h3><p>One of the most promising benchmarks available for iO is <a href="https://machina-io.com/posts/hello_world_first.html">that of Machina IO’s</a>, benchmarking the <a href="https://eprint.iacr.org/2025/236">Diamond iO construction</a>. The results are both a win in bringing iO from theory to reality and a stark indicator of how much is still left to be done. While Diamond iO <em>is</em> able to obfuscate a watered-down multiplication circuit in less than a human lifetime (whereas the previous benchmark took more time than the universe has been around), obfuscating a full program would still take much more time than the universe has been around. The benchmark is a lower bound: a few alterations from the theoretical construction were made, including using a simpler FHE scheme and omitting the implementation of a hash function (which remains too complicated to implement).</p><p>The Diamond iO benchmark consists of a simple aggregator program: it performs additions and multiplications under encryption on an input and reveals the output. The multiplications are done under FHE encryption; the evaluator is able to perform the FHE encryption and decryption without learning the secret key. The multiplication circuit in this model is public: the obfuscation in this case is useful not to keep the operation secret, but to keep implementation details such as the secret key used private from the evaluator.</p><p><strong>The cost of obfuscation.</strong> Creating an obfuscated program with Diamond iO is extremely expensive. Even a toy example that adds or multiplies a handful of times requires tens of minutes to hours of computation, consumes tens to hundreds of gigabytes of memory, and produces obfuscated binaries tens of gigabytes in size. This step is done once per program, but its scale illustrates why iO is currently impractical.</p><p><strong>The cost of evaluation.</strong> Running the obfuscated program is also slow and resource-hungry compared to normal cryptographic tools. Evaluation times are measured in minutes, not milliseconds, and still demand tens or hundreds of gigabytes of RAM. For context, a zero-knowledge proof or a homomorphic encryption operation can often be carried out on commodity hardware in seconds or less, while the iO benchmark shows evaluation bottlenecks that are orders of magnitude slower.</p><p>The following tables compare the speed from <a href="https://machina-io.com/posts/hello_world_first.html">Machina IO’s benchmarks</a> to other private computations. As multiple constructions and benchmarks exist for FHE and ZK computations with varying tradeoffs in efficiency, we provide a general number in the table and give sources to FHE and ZK resources below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gP_VhelG8i9hkLtOa2yOIQ.png" /><figcaption>Approximate execution speed for different cryptographic primitives</figcaption></figure><p>The first benchmark was measured by the number of multiplications done on the input: no multiplications, 5 multiplications, 10 multiplications.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EjyVOjh1xAVxWoBVJw986Q.png" /></figure><p>If you think this is a very simple benchmark, it is. For comparison, here are some common on-chain cryptographic operations and their multiplicative depth.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LWlZh6mCw2CgSNcnRLEDyw.png" /></figure><h3>On-Chain iO</h3><p>Smart contracts running obfuscated circuits provide a new bounty of use cases. By being able to embed secrets on-chain, we achieve public functionalities with the following features:</p><ul><li>A trust model that is independent from trusted hardware (TEEs)</li><li>The accountability of a public ledger in the use of secrets (who accesses the obfuscated functionality)</li><li>The programmability of smart contracts (whitelists on who can use the functionality, etc)</li></ul><p>In the case of encryption and decryption, a trusted setup can be performed in which a set of encryption and decryption keys are generated. The decryption keys can then be embedded in a program which is then obfuscated and put on-chain, and the encryption key can be given to relevant parties. Now, anyone who has the encryption keys can post encrypted values to the blockchain which can then be processed and decrypted by the obfuscated program/smart contract on demand. Implementation details such as the processing ciphertexts using secret keys can now be kept private from the public nodes running the computation.</p><p>The difference between this and a signature or commitment scheme is the ability to process the encrypted value: the encrypted values that are posted can be processed by the smart contract with only the output being revealed, with inputs and intermediate values remaining hidden.</p><p>In the context of a proof of reserve, this can be powerful: treasuries can post encrypted data corresponding to proof of reserves on-chain. This data cannot be read by anyone; it can only be used by the obfuscated program. Any node running the obfuscated program however, cannot see this encrypted data in the clear, even when running the obfuscated program: the node can only compute and decrypt the output. As long as treasuries post updated data, proofs of reserve would now be able to be computed on demand (anyone doubting the veracity of the data can request a proof that the encrypted data is correct).</p><p>Taking this a step further, this provides a model for oblivious on-chain state and advanced permissioned functionalities. Encrypted data can be posted on-chain, unreadable and unusable by anyone other than the obfuscated functionality: this functionality could be a proof of reserve, as in the case above, the computation of specific proofs, or even something as simple as reading the private on-chain balance of an address.</p><p>The secret embedded in a program need not be restricted to that of a secret encryption key. An on-chain program can have an API token embedded in it, which can then be used to generate API requests. As long as the secret token that proves the program’s identity cannot be copied or learned by any node running it, iO can give us a framework for privileged on-chain computation.</p><h3>Standing at the Edge of the Maze</h3><p>Indistinguishability obfuscation sits at a strange frontier: it exists, but only just. For cryptographers, it has already re-drawn the map of what is possible. For builders in blockchain and Web3, it dangles the promise of governance without coercion, markets without exploitation, bridges without betrayal. At the same time, it raises sharp new questions: what does verification mean when logic itself is hidden? How do we keep a dark forest from turning into a black box?</p><p>iO is not ready for production blockchains today. But it has shifted the Overton window: we now know that software can be made runnable yet unreadable. If history is any guide, the gap between theory and practice will narrow — perhaps suddenly, as it did for zero-knowledge proofs and homomorphic encryption. When it does, iO may prove not just a master key for cryptography, but a cornerstone for the next generation of decentralized systems.</p><h3>Resources</h3><h4>iO</h4><ul><li>On the Impossibility of Obfuscating Programs, Barak et al, 2001 <a href="https://www.iacr.org/archive/crypto2001/21390001.pdf">https://www.iacr.org/archive/crypto2001/21390001.pdf</a></li><li>Candidate Indistinguishability Obfuscation and Functional Encryption for all circuits, Garg et al, 2013 <a href="https://eprint.iacr.org/2013/451.pdf">https://eprint.iacr.org/2013/451.pdf</a></li><li>Cryptanalysis on the Multilinear Map over the Integers and its Related Problems, Cheon et al, 2014 <a href="https://eprint.iacr.org/2014/906.pdf">https://eprint.iacr.org/2014/906.pdf</a></li><li>Limits on the Power of Indistinguishability Obfuscation and Functional Encryption, Asharov, Segev, 2015 <a href="https://eprint.iacr.org/2015/341.pdf">https://eprint.iacr.org/2015/341.pdf</a></li><li>Implementing BP-Obfuscation Using Graph-Induced Encoding, Halevi, Halevi, Shoup, 2017 <a href="https://eprint.iacr.org/2017/104.pdf">https://eprint.iacr.org/2017/104.pdf</a></li><li>Indistinguishability Obfuscation without Multilinear Maps: New Paradigms via Low Degree Weak Pseudorandomness and Security Amplification, Ananth et al., 2019 <a href="https://eprint.iacr.org/2019/643.pdf">https://eprint.iacr.org/2019/643.pdf</a></li><li>iO: State of the Art <a href="https://sites.google.com/view/iostate-of-the-art/home">https://sites.google.com/view/iostate-of-the-art/home</a></li><li>Indistinguishability Obfuscation from Well-Founded Assumptions, Jain, Lin, and Sahai, 2020 <a href="https://eprint.iacr.org/2020/1003.pdf">https://eprint.iacr.org/2020/1003.pdf</a></li><li>Idea: Benchmarking Indistinguishability Obfuscation — A candidate implementation, Banescu et al, 2022 <a href="https://mediatum.ub.tum.de/doc/1246288/1246288.pdf">https://mediatum.ub.tum.de/doc/1246288/1246288.pdf</a></li><li>Diamond iO: A Straightforward Construction of Indistinguishability Obfuscation from Lattices, Suegami, Bottazzi, and Park, 2025 <a href="https://eprint.iacr.org/2025/236">https://eprint.iacr.org/2025/236</a></li><li>Hello, World: The First Signs of Practical iO, Suegami, Bottazzi, and Park, 2025 <a href="https://machina-io.com/posts/hello_world_first.html">https://machina-io.com/posts/hello_world_first.html</a></li></ul><h4>ZK</h4><ul><li>Evaluating the Efficiency of zk-SNARK, zk-STARK, and Bulletproof in Real-World Scenarios: A Benchmark Study, El-Hajj and Roelink, 2024 <a href="https://www.researchgate.net/publication/382138687_Evaluating_the_Efficiency_of_zk-SNARK_zk-STARK_and_Bulletproof_in_Real-World_Scenarios_A_Benchmark_Study">https://www.researchgate.net/publication/382138687_Evaluating_the_Efficiency_of_zk-SNARK_zk-STARK_and_Bulletproof_in_Real-World_Scenarios_A_Benchmark_Study</a></li><li>Benchmarking ZK-Friendly Hash Functions and SNARK Proving Systems for EVM-compatible Blockchains, Guo et al, 2024 <a href="https://arxiv.org/pdf/2409.01976">https://arxiv.org/pdf/2409.01976</a></li><li>Benchmarking ZKP Development Frameworks <a href="https://ethresear.ch/t/benchmarking-zkp-development-frameworks-the-pantheon-of-zkp/14943">https://ethresear.ch/t/benchmarking-zkp-development-frameworks-the-pantheon-of-zkp/14943</a></li><li>ZKBench <a href="https://zkbench.dev/">https://zkbench.dev/</a></li></ul><h4>FHE</h4><ul><li>Making FHE Faster for ML: Beating our Previous Paper Benchmarks with Concrete ML, Zama <a href="https://www.zama.ai/post/making-fhe-faster-for-ml-beating-our-previous-paper-benchmarks-with-concrete-ml">https://www.zama.ai/post/making-fhe-faster-for-ml-beating-our-previous-paper-benchmarks-with-concrete-ml</a></li><li>Cross-Platform Benchmarking of the FHE Libraries: Novel Insights into SEAL and OpenFHE, Faneela et al, 2025 <a href="https://eprint.iacr.org/2025/473.pdf">https://eprint.iacr.org/2025/473.pdf</a></li><li>FHEBench, Jaing and Ju, 2022 <a href="https://arxiv.org/pdf/2203.00728">https://arxiv.org/pdf/2203.00728</a></li><li>SEAL FHE <a href="https://github.com/microsoft/SEAL">https://github.com/microsoft/SEAL</a></li><li>OpenFHE <a href="https://openfhe.org/">https://openfhe.org/</a></li><li>HELib <a href="https://github.com/homenc/HElib">https://github.com/homenc/HElib</a></li><li>Zama <a href="https://www.zama.ai/">https://www.zama.ai/</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e00bf07b95fe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Landscape of Multi-Party Fully Homomorphic Encryption]]></title>
            <link>https://fenbushicapital.medium.com/the-landscape-of-multi-party-fully-homomorphic-encryption-441c1ea16179?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/441c1ea16179</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Tue, 29 Jul 2025 20:54:22 GMT</pubDate>
            <atom:updated>2025-07-29T20:54:22.784Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cFbkIqlWZYa2VxuL.png" /></figure><p>Authored by <a href="https://www.rileynwong.com/">Riley Wong</a>, <a href="https://emergentresearch.net/">Emergent Research</a>, July 2025.</p><p>A PDF version of this report can be found <a href="https://www.emergentresearch.net/landscape-of-mpfhe">here</a>.</p><h3>Abstract</h3><p>Multi-party fully homomorphic encryption (MP-FHE) combines the strengths of multi-party computation (MPC) with fully homomorphic encryption (FHE), making it possible to collaboratively compute on encrypted data without revealing private inputs nor requiring a trusted third party. Benefits of MP-FHE include distributed trust, privacy, and quantum-resistance, and can be used currently for applications such as privacy-preserving machine learning, distributed data analytics, encrypted decision making, and more. MP-FHE is still considered a relatively new field in cryptography and would benefit from further optimizations given its computational intensity and coordination between parties, as well as further developments in library and developer support. New paradigms for what can be possible with MP-FHE in the broader world of programmable cryptography have yet to be fully explored.</p><h3>Acknowledgements</h3><p>Special thanks to Pia Park, Janabel Xia, Michele Orrù, Violet Vienhage, Aayush Gupta, and Yuki Yuminaga for their insightful comments, discussions, and feedback, and to <a href="https://x.com/fenbushi">Fenbushi Research</a> for their grant support! ❤</p><h3>Introduction</h3><p>Fully homomorphic encryption (FHE) is a cryptographic scheme that makes it possible to compute on encrypted data. Multiparty computation (MPC) is a cryptographic protocol that allows multiple parties to jointly compute while keeping their data private, and often needs more than one round of communication. Multi-party fully homomorphic encryption (MP-FHE) combines these techniques for multiple parties to jointly compute encrypted data with minimal interaction.</p><p>A main benefit of MP-FHE over just FHE includes distributing decryption keys, which can be beneficial for security purposes as it reduces single points of failure and can decentralize trust. A main benefit of MP-FHE over just MPC means that the data existing in an encrypted state allows for fewer rounds of interaction and can be beneficial for scaling up for more parties. In short, MP-FHE benefits from the non-interactive and scalable features of FHE as well as the distributed security of MPC. MP-FHE offers strengths in providing distributed or decentralized architecture; trust and security benefits; and opportunities for collective governance mechanisms.</p><p>General practical use cases for MP-FHE may include scenarios that involve private data being distributed or computed among multiple parties, such as: private set intersections; privacy-preserving machine learning without revealing private data; privacy-preserving similarity calculations for private encodings; privacy-preserving data analysis for jointly computing aggregate statistics.</p><p>Examples of real world use cases for MP-FHE may include: privacy-preserving voting for citizen elections; training machine learning models while keeping individual health records private; privacy-preserving workplace abuse reporting; or private smart contracts and decentralized financial transactions.</p><p>In its current state, there exist a small handful of implementations, namely through the Lattigo and OpenFHE libraries. MP-FHE is still relatively nascent as a technology and undergoing active research; can be computationally expensive in its present state; and is considered quantum-safe when using lattice-based schemes. Novel use cases and applications for MP-FHE have yet to be fully explored.</p><h3>Secure Multi-Party Computation (SMPC)</h3><p>Secure multi-party computation (SMPC, or MPC) is a cryptographic technique that allows <em>multiple parties</em> to <em>jointly compute</em> a function over their <em>private inputs</em>. The parties can see the result of the computation without ever needing to reveal their data to one another.</p><p>Mathematically, we may frame MPC as a problem involving <em>n</em> parties P₁, P₂, … , Pₙ, where each party Pᵢ holds a private input xᵢ, with the goal for all parties to compute a function <em>f </em>(x₁, x₂, …, xₙ) = y.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/922/0*wS77LfZY6Eps_7aC.png" /><figcaption>Figure 1: Multi-party computation</figcaption></figure><p>The main flow of secret sharing-based multi-party computation generally goes as follows:</p><ol><li>Offline phase:</li></ol><p>a. Preprocessing: Multiplication triplets, message authentication codes, and random values used for masking inputs can be generated ahead of time to be used later, independent of inputs.</p><p>2. Online phase:</p><p>a. Distributed input shares: Each party splits their input data into <em>secret-shares</em> and sends them to the other parties.</p><ul><li>In math terms: each input xᵢ is split into shares [xᵢ]₁, [xᵢ]₂, …, [xᵢ]ₙ. These shares are distributed to the other parties.</li></ul><p>b. Distributed computation: Each party runs a partial computation based on the partial secret-shares they’ve received.</p><ul><li>In math terms: Given that each party Pᵢ holds [xᵢ]₁, [xᵢ]₂, …, [xᵢ]ₙ, each party can evaluate <em>f</em> for their set of received distributed shares: <em>f </em>([xᵢ]ᵢ, [x₂]ᵢ, …, [xₙ]ᵢ) = [y]ᵢ.</li><li>The result of this partial computation [y]ᵢ is shared with other parties.</li></ul><p>c. Output reconstruction: The result of the computation is constructed by combining each partial computation’s result.</p><ul><li>In math terms: <em>Reconstruct</em>([y]₁, [y]₂, …, [y]ₙ) = y = <em>f </em>(x₁, x₂, …, xₙ).</li></ul><p>No single party learns the private inputs of the computation, nor the intermediate values; only the result.</p><p>Currently, the main approaches to multi-party computation are:</p><ol><li>Secret sharing (e.g. Shamir’s secret sharing, SPDZ)</li><li>Garbled circuits</li><li>Multi-party Fully Homomorphic Encryption (See: <a href="#4023">Multi-Party Fully Homomorphic Encryption (MP-FHE)</a>)</li></ol><h3>Shamir’s Secret Sharing (SSS)</h3><p>In secret-sharing based protocols, each party splits their private input into separate shares and distributes these shares to the other parties. Then, each party computes on these collected shares, without knowledge of the original inputs. At the end, parties combine the resulting computation to reconstruct the full output.</p><p>Mathematically, a secret share can be thought of as a point on a polynomial curve of degree <em>n</em>. With <em>n</em>+1 points on the curve, we can reconstruct the polynomial curve using <a href="https://practicaldsc.org/wn25/guides/machine-learning/interpolation/#:~:text=of%20polynomials%20shortly.-,Point%20Representation,=%201%20+%202%20x%202%20.">polynomial interpolation</a>, such as Lagrange; otherwise, it would be mathematically impossible to reconstruct.</p><p><a href="https://dl.acm.org/doi/pdf/10.1145/359168.359176">Shamir’s secret sharing</a> is a foundational tool used in MPC, though it is not by itself a full MPC protocol, and may need additional protocols for securely evaluating functions, handling malicious adversaries, or verifying results from tampering or collusion.</p><h3>SPDZ (Smart, Pastro, Damgard, Zakarias)</h3><p>Named after the paper authors, <a href="https://eprint.iacr.org/2011/535">SPDZ</a> (pronounced “speedz”) is one of the most common protocols for multi-party computation, using secret sharing as a foundation for securely computing functions over shared values, even when malicious adversaries are present (i.e. the protocol remains secure even if some parties try to cheat).</p><p>It operates in two distinct phases: offline, which is computationally heavy and input-independent; and online, which is fast and input-dependent.</p><p>The general flow of SPDZ goes as follows:</p><ol><li>Offline phase / preprocessing: This step happens independent of the computation inputs and occurs before any parties input their private data.</li></ol><p>a. The system generates <a href="https://medium.com/partisia-blockchain/beavers-trick-e275e79839cc"><strong>multiplication triples</strong></a>, sets of values (a, b, c) such that c = a*b. These triples will be used later to quickly compute secure multiplications. Each triple is secret-shared among the multiple parties.</p><p>b. Parties also generate random values for masking inputs as well as <a href="https://textbook.cs161.org/crypto/macs.html"><strong>message authentication codes (MACs)</strong></a><strong> </strong>to ensure integrity and detect cheating.</p><p>2. Online phase:</p><p>a. Distributed input shares: each party uses random values from the preprocessing phase to mask their inputs, then splits their private inputs into random shares and distributes them to all parties. No single party can reconstruct the original inputs.</p><p>b. Distributed computation: Each party evaluates their shares using addition and multiplication, using the multiplication triples for multiplication operations and the MACs to authenticate operations for integrity.</p><p>c. Output reconstruction: Parties reconstruct the computation result by combining output shares and verify MAC checks.</p><h3>Garbled Circuits</h3><p><a href="https://eprint.iacr.org/2012/265.pdf"><strong>Garbled circuits</strong></a> are a separate cryptographic protocol from secret sharing that also allows two or more parties to jointly compute a function on private inputs without revealing these inputs to one another.</p><p>In a 2-party garbled circuit, one party, the <strong>garbler</strong>, represents the computation as an encrypted boolean circuit.</p><p>A <strong>boolean circuit</strong> is a circuit made up of logic gates (AND, OR, NOT, etc.) that take bits (0 or 1) as input and outputs the corresponding bit for that gate, and can represent computations using these binary operations. The boolean circuit can furthermore be represented as a truth table of 0s and 1s.</p><p>To encrypt the boolean circuit, the garbler replaces the truth table values with labels, represented by randomly generated strings, and encrypts the output entries in the table using these generated labels. The garbler then permutes, or “garbles,” the rows of the encrypted table. This garbled truth table represents an encrypted boolean circuit, which the garbler sends to the second party, along with the labels that represent her inputs.</p><p>This second party, the <strong>evaluator</strong>, then takes the input labels from the garbler, as well as the input labels of their own inputs using <strong>oblivious transfer</strong>, and evaluates the encrypted circuit.</p><p>Thus, the inputs and intermediate values are not revealed to either party. This process can be extended to multiparty scenarios using secret sharing, where instead of a single garbler, each of the multiple parties garbles a portion of the circuit.</p><p>(See <a href="#3281">Appendix: Garbled Circuits </a>for a slightly more detailed explanation.)</p><h3>Oblivious Transfer</h3><p>In garbled circuits, the evaluator needs to give the garbler their inputs into the circuit, without revealing what those values are. <a href="https://eprint.iacr.org/2005/187.pdf"><strong>Oblivious transfer</strong></a> is a fundamental building block of garbled circuits and allows this process of delivering encrypted input bits to happen privately and securely.</p><p>In “1-out-of-2” oblivious transfer, the sender has two messages, and the receiver picks one of the two to learn without revealing their choice. The sender learns nothing about the receiver’s choice, and the receiver learns nothing about the other message. This process makes it possible to securely evaluate an encrypted boolean circuit without revealing private inputs.</p><h3>Threshold Decryption</h3><p><a href="https://csrc.nist.gov/projects/threshold-cryptography"><strong>Threshold decryption</strong></a> is a technique where no single party can decrypt a message, but multiple parties can collaborate to decrypt it together. While MPC frameworks could require all parties for decryption, threshold decryption is significantly more common given its benefits around availability, fault tolerance, or usability. For example, if some parties go offline, get disconnected, crash, have high latency or downtime, or are malicious or adversarial, the MPC decryption can still succeed.</p><h3>Table: Existing MPC Libraries</h3><p>Some notable open-source libraries for MPC include:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/942/0*4tZLdCFlD9Mfsa7Y.png" /><figcaption>(see clickable links for each library below)</figcaption></figure><p>Library Links: <a href="https://nigelsmart.github.io/SCALE/"><strong>SCALE-MAMBA</strong></a>, <a href="https://github.com/emp-toolkit"><strong>EMP-toolkit</strong></a>, <a href="https://github.com/data61/MP-SPDZ"><strong>MP-SPDZ</strong></a>, <a href="https://github.com/lschoe/mpyc"><strong>MPyC</strong></a>, <a href="https://github.com/encryptogroup/MOTION"><strong>MOTION</strong></a>, <a href="https://github.com/Crypto-TII/FANNG-MPC"><strong>FANNG-MPC</strong></a></p><h3>Existing Applications and Challenges</h3><p>Secure MPC has already been <a href="https://mpc.cs.berkeley.edu/blog/deployment-dilemma.html">adopted</a> at places like Signal and Coinbase, namely in their key recovery protocols. SMPC was also used by the Internet Security Research Group (ISRG) in their <a href="https://mpc.cs.berkeley.edu/posts/Privacy-Preserving-COVID-19-Exposure-Notifications/">COVID-19</a> exposure notification system for analyzing privacy-preserving aggregate statistics from mobile devices.</p><p>Currently, MPC protocols still require heavy computational and communication overhead, as they often require multiple rounds of interaction between parties, and/or assume parties are online and synchronous in order to communicate between rounds. MPC protocols would also benefit from improvements in scaling as the number of parties increases; further research into security guarantees against malicious parties and/or collusion; and further developments in usability and developer support and tooling.</p><h3>Fully Homomorphic Encryption (FHE)</h3><p><a href="https://dl.acm.org/doi/abs/10.1145/1536414.1536440">Fully homomorphic encryption</a> is a cryptographic technique that allows for computations <em>on encrypted data</em>, without ever decrypting it. This means that data can stay encrypted throughout the entire process, from rest, transfer, and computation, resulting in an encrypted result that can only be decrypted by the original party.</p><p>This makes it possible to provide private computations as a service. For example, a machine learning model may be trained and give predictions on fully encrypted data, resulting in encrypted outputs; I can give my private data to this model and benefit from its predictions without my private data ever being revealed or decrypted by the model.</p><p>As an example use case, I may want to use a translation service such as Google Translate. With an FHE translation service, I can run the translation model on encrypted text, keeping my encrypted input text private from the translation service providers.</p><p>At the moment, FHE is still very computationally expensive, and would benefit from both mathematical as well as hardware optimizations.</p><p>How do homomorphic encryption (HE) and <em>fully</em> homomorphic encryption (FHE) differ? Given the way HE works, noise builds up in the computations such that, after enough operations, the noise eventually makes the output undecryptable, adding a hard computational limit. FHE involves a <a href="https://www.cs.cmu.edu/~odonnell/hits09/gentry-homomorphic-encryption.pdf"><strong>bootstrapping</strong></a> step that resets this noise in the process, making it possible to continue operations indefinitely.</p><p>Many FHE schemes rely on lattice-based constructions called <strong>Learning With Errors (LWE)</strong> and <strong>Ring Learning With Errors (RLWE)</strong>, or a torus-based construction called <strong>Torus Learning With Errors (TLWE)</strong>.</p><h3>Learning with Errors (LWE)</h3><p>The task of solving a system of linear equations is not necessarily difficult. However, by adding small random errors with a specific Gaussian distribution, solving a system of linear equations with the added errors becomes an NP-hard problem. <a href="https://dl.acm.org/doi/10.1145/1060590.1060603">Introduced</a> by Oded Regev in 2005, this is referred to as the <a href="https://cims.nyu.edu/~regev/papers/lwesurvey.pdf">Learning with Errors problem</a>.</p><p>What makes LWE a lattice-based construction? In 2-dimensions, a lattice can be thought of like a repeating grid. In higher dimensions, a lattice consists of linear combinations of <em>vectors</em>. Learning with errors can be thought of as points on a lattice-grid that are <em>a little off-grid due to the added noise</em>. This makes LWE analogous to the <strong>Closest Vector Problem (CVP).</strong></p><h3>Closest Vector Problem</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/756/0*oJmj_lkxiDQ-Gy5f.png" /><figcaption>Figure 2: Closest vector problem. b1 and b2 are the basis vectors for lattice L. Random point P is highlighted in orange, with the closest lattice point highlighted in green.</figcaption></figure><p>Given a lattice <em>L</em> and a random point <em>P</em>, which lattice point is the closest? In high dimensions, finding the solution for this requires an exponentially large search space.</p><p>In LWE, you can think of the public key as a bunch of noisy points near the lattice points. You would need the secret key in order to find the nearest lattice point, and the distance between these points is the noise or error in LWE. Only someone who knows the original lattice grid, or the secret key, would be capable of removing the noise, or decrypt the message. Otherwise, it would be computationally infeasible, an NP-hard problem. This mathematical hardness is what makes LWE-based cryptography schemes assumed to be computationally secure and quantum-proof.</p><h3>Bootstrapping and Noise Management</h3><p>With LWE, the added random noise is what ensures security. However, this noise grows with each operation. After enough operations, the noise becomes too large to distinguish the decrypted message. Without a way to manage the noise, we hit a limit on the number of operations we can compute, leaving us with only <strong>somewhat</strong><em> </em>or <strong>leveled homomorphic encryption</strong>.</p><p>What allows for homomorphic encryption to be <em>fully</em> homomorphic? <strong>Bootstrapping</strong> is a trick that allows us to “reset” the amount of noise in this process, making LWE-based encryption somewhat homomorphic, i.e. we can only run a limited number of operations, to <em>fully</em> homomorphic, i.e. we can continue to securely run unlimited operations.</p><p>The general idea of bootstrapping is that, by <em>homomorphically</em> decrypting the FHE ciphertext using an encryption of the original secret key, and re-encrypting it with a new secret key through a process called <strong>key switching</strong>, the noise is “reset” to a smaller value without ever revealing the underlying message. The specific technical details of bootstrapping <a href="https://www.zama.ai/post/what-is-bootstrapping-homomorphic-encryption">depends</a> on the FHE scheme used.</p><h3>Ring Learning with Errors (RLWE)</h3><p>Building off of LWE, <a href="https://link.springer.com/chapter/10.1007/978-3-642-13190-5_1">ring-LWE</a> operates similarly in that it’s also lattice-based, involves recovering a secret from a set of noisy equations, and is NP-hard and assumed to be quantum-resistant.</p><p>Instead of using vectors and matrices like in LWE, RLWE operates over a <strong>polynomial ring</strong>, a set of polynomials with defined addition and multiplication operations. The polynomial ring offers an algebraic structure such that computations are faster and more compact without sacrificing security or mathematical hardness, using polynomial multiplication and Fast Fourier Transforms (FFTs) instead of the matrix-vector operations of LWE.</p><p>While LWE has been foundational early in the field of FHE, RLWE is often what actually gets used in FHE implementations currently.</p><h3>Torus Learning with Errors (TLWE)</h3><p><a href="https://eprint.iacr.org/2018/421">Torus-LWE</a> also builds off of the LWE problem, and is optimized specifically for boolean circuits, bit-level data (i.e. 0’s and 1’s), and logic gates, making it useful for applications such as encrypted decision trees, privacy-preserving rule-checking, bit-wise search or matching, or FHE for embedded systems, whereas RLWE would be more useful for applications such as privacy-preserving machine learning or privacy-preserving data analytics.</p><p>In Torus-LWE, the mathematical torus construction refers to the set of real numbers mod 1, and is therefore particularly applicable for bit-level data represented as 0’s and 1’s. While it includes fast, single-bit bootstrapping, making it especially efficient for bit-wise data operations, it can also be much less expressive compared to ring-based constructions.</p><h3>Main FHE Schemes</h3><h3>Arithmetic</h3><h3>2011: BGV (Brakerski, Gentry, Vaikuntanathan)</h3><p>Named after the paper authors, <a href="https://eprint.iacr.org/2011/277">BGV</a> relies on lattice-based Ring-LWE for performing encrypted addition and multiplication and is one of the first practical FHE schemes, acting as a foundation for many FHE systems used currently.</p><h3>2014: BFV (Brakerski, Fan-Vercauteren)</h3><p>Developed shortly after and named after the paper authors, <a href="https://eprint.iacr.org/2012/144">BFV</a> relies on lattice-based Ring-LWE to perform exact arithmetic for addition and multiplication on integers. BFV is seen as simpler and more usable over its BGV predecessor, with better noise management and more efficient multiplication.</p><p>BFV works for discrete values, such as counting, voting, or small integers. BFV works less well for real number calculations, and requires bootstrapping to reduce noise as computation depth grows.</p><h3>2017: CKKS (Cheon, Kim, Kim, Song)</h3><p>Named after the paper authors, <a href="https://eprint.iacr.org/2016/421">CKKS</a> works well for approximate arithmetic on real or complex numbers, such as for use cases like encrypted machine learning, signal processing, or statistics. CKKS also relies on Ring-LWE, including a step that encodes real or complex numbers, and trades off exactness for efficiency for real numbers. CKKS works less well for use cases where precision is a priority.</p><h3>Boolean</h3><h3>2014: FHEW (Fast Homomorphic Encryption over the Weak Ring)</h3><p>Introduced by Ducas &amp; Micciancio, <a href="https://link.springer.com/chapter/10.1007/978-3-662-46800-5_24">FHEW</a> is optimized for binary data, is specialized for fast bootstrapping, and is meant to be useful for boolean circuits and bit-level operations. FHEW relies on Torus-LWE, with the W referring to the “weak ring” in the torus representation (as opposed to the polynomial rings in Ring-LWE schemes such as BGV and BFV).</p><p>The mathematical torus construction makes arithmetic more efficient and bootstrapping faster compared to those based on polynomial rings.</p><h3>2016: TFHE (Torus FHE)</h3><p>Paper authored by Chillotti, Gama, Georgieva, and Izabachène, <a href="https://eprint.iacr.org/2018/421">TFHE</a> builds off of FHEW, operates over the torus and implements logic gates using gate bootstrapping. Similar to FHEW, TFHE also works well for bit-level computations, such as for logic circuits, decision trees, and Boolean functions, operating on binary data with a fast bootstrapping mechanism that allows for unlimited computation depth. It works less well for large-scale arithmetic computations, such as matrix multiplications.</p><p>Named after the paper authors, fast fully homomorphic encryption schemes over the torus such as FHEW and TFHE are sometimes also referred to as CGGI (Chillotti‑Gama‑Georgieva‑Izabachène).</p><h3>Table: Fully Homomorphic Encryption Schemes</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/914/0*RMYHipAkfdOuMFK1.png" /><figcaption>Figure 3: Overview and evolution tree of FHE schemes.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/954/0*PNU_3b-X75E96cF7.png" /></figure><h3>Table: Existing FHE Libraries</h3><p>The <a href="https://github.com/jonaschn/awesome-he">Awesome-HE</a> Github repo compiles a large list of homomorphic encryption libraries as well. Some notable open-source libraries for FHE include:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/986/0*AG5ON_iPE_JLtENX.png" /><figcaption>(see clickable links for each library below)</figcaption></figure><p>Library Links: <a href="https://github.com/HomEnc/HElib"><strong>HElib</strong></a>, <a href="https://github.com/microsoft/SEAL"><strong>SEAL</strong></a>, <a href="https://gitlab.com/palisade/palisade-release"><strong>PALISADE</strong></a>, <a href="https://github.com/tuneinsight/lattigo"><strong>Lattigo</strong></a>, <a href="https://github.com/zama-ai/fhevm"><strong>TFHE-rs</strong></a>, <a href="https://github.com/openfheorg/openfhe-development"><strong>OpenFHE</strong></a></p><p>Note that, as of spring 2025, only Lattigo and OpenFHE provide robust multi-party support for FHE.</p><h3>Existing Applications and Challenges</h3><p>Production systems already using <em>somewhat</em> homomorphic encryption include Apple’s <a href="https://www.swift.org/blog/announcing-swift-homomorphic-encryption/">Live Caller ID Lookup</a>, which uses HE to send an encrypted query to a server about whether a number is a spam caller or not, without the server ever seeing the specific phone number in the query; and Microsoft’s <a href="https://www.microsoft.com/en-us/research/blog/password-monitor-safeguarding-passwords-in-microsoft-edge/">Password Monitor</a>, which checks if a user’s password exists a list of breached passwords without the server ever seeing the specific password in the query.</p><p>Potential applications that FHE may be particularly fitting for currently include healthcare and genomics data, finance, government and census data, and privacy-preserving machine learning inference, to name a few.</p><p>While FHE libraries do exist that are ready for use, as of spring 2025, fully homomorphic encryption systems are still currently rare to see deployed in production. FHE is still extremely computationally heavy, often orders of magnitude slower than their plaintext operations counterparts. Existing bootstrapping methods for resetting noise are still computationally expensive as well. For many use cases, FHE is still not quite computationally feasible to easily or accessibly use compared to other privacy-preserving approaches.</p><p>For these reasons, <strong>trusted execution environments (TEEs)</strong>, such as <strong>secure enclaves</strong>, are often a more practical alternative to FHE approaches for secure computation, as they are currently often faster, cheaper, and easier to deploy, as of spring 2025. Unlike FHE, TEEs require trusted third party vendors and can be vulnerable to hardware exploits, such as side-channel attacks.</p><p>Additional advances and further research in performance optimizations, hardware acceleration, developer tooling, and usability would make it more possible for FHE to be adopted more widely.</p><h3>Multi-Party Fully Homomorphic Encryption (MP-FHE)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/760/0*yTCYplVZQQToVGsH.png" /><figcaption>Figure 4: Multi-party fully homomorphic encryption.</figcaption></figure><p>In multi-party fully homomorphic encryption (MP-FHE), multiple parties can <em>jointly compute</em> on <em>encrypted data</em>. No single party holds the decryption key, and no party reveals their private inputs. To reveal the computation output, enough parties must agree to cooperate to decrypt the result.</p><p>MP-FHE makes it possible to combine the power of FHE, computing on encrypted data, with the multi-party control over a secret key, making it possible to run <em>collaborative, privacy-preserving computations</em> non-interactively and without needing a trusted third party. Some benefits of MP-FHE include, notably, collective key ownership and privacy-preserving computation; removes the need for trusted setup; and, as with FHE, is assumed to be quantum-safe.</p><p>Potential tradeoffs of MP-FHE are that, currently, it requires high computational overhead, as FHE is already slow and resource intensive and now adds additional complexity for multi-party support; may require multiple rounds of communication between parties during distributed key generation and threshold decryption steps; is still relatively new and may lack mature, robust, and well-supported open-source libraries or developer tooling; and distributed bootstrapping techniques are still considered relatively complex and slow and requires additional research advances for optimization.</p><h3>Putting It Together</h3><p>At a high level, we combine the pieces of generating distributed secret key shares from MPC with computing on encrypted data from FHE.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/962/0*vvHTK4iK9YL5j9sP.png" /><figcaption>Figure 5: General flow of multi-party fully homomorphic encryption.</figcaption></figure><p>A general high level flow of an MP-FHE computation may look like:</p><ol><li>Joint key generation: Multiple parties jointly generate a <em>shared public key</em> and <em>distributed secret key shares</em>. The secret key has been secret-shared using a protocol such as Shamir’s Secret Sharing or distributed key generation, and as a result, no single party knows the full decryption key.</li><li>Encryption: Each party encrypts their private data using the shared public key. These encrypted inputs are fed into a computation using FHE.</li><li>Homomorphic evaluation: Either a designated evaluator or all of the parties can run the computation on the encrypted inputs using FHE schemes such as BGV/BFV, CKKS, or TFHE, resulting in an encrypted output. This FHE step may also be outsourced to a third party that can run FHE computations, as all the inputs are encrypted.</li><li>Collective decryption: Each party can partially decrypt the output using their secret key shares. Similar to Shamir’s secret sharing, i.e. using Lagrange interpolation, multiple parties must combine their partial decryption shares in order to construct the decrypted output. The final output is only fully decrypted when a minimum threshold of parties participate in this process.</li></ol><h3>Key Switching</h3><p>Key switching is a fundamental technique used in FHE transforms a ciphertext encrypted under one secret key into an equivalent ciphertext under a different secret key, without ever decrypting the data.</p><p>While also used in FHE schemes, typically as a part returning the ciphertext to the original encryption key after bootstrapping, key switching may be worth noting for MP-FHE schemes.</p><p>In MP-FHE, where a secret key may be distributed among multiple parties, key switching is a collaborative process, where parties jointly generate a collective switching key that allows the ciphertext to be re-encrypted under a different shared decryption key. Additionally, if any party leaves or joins, key switching is needed to move existing ciphertexts from the old collective secret key over to a new collective secret key.</p><h3>Table: Existing MP-FHE Libraries</h3><p>Notable open-source libraries for MP-FHE include:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/946/0*5nOPe9FtK9O3F52K.png" /><figcaption>(see clickable links for each library below)</figcaption></figure><p>Library Links: <a href="https://github.com/tuneinsight/lattigo"><strong>Lattigo</strong></a>, <a href="https://github.com/openfheorg/openfhe-development"><strong>OpenFHE</strong></a></p><p>As of spring 2025, both Lattigo and OpenFHE are actively maintained.</p><h3>Existing Applications and Challenges</h3><p>As of spring 2025, existing MP-FHE systems are typically still considered experimental or research-grade.</p><p>Some potential use cases for MP-FHE include private set intersection (PSI) applications, privacy-preserving abuse reporting, or privacy-preserving <a href="https://proteus.photos/">content provenance</a> databases. Other useful potential domains include: joint data analytics for healthcare, financial, government data, and other domains involving personally identifiable information (PII); aggregate genomic and biomedical data analysis; privacy-preserving abuse detection and content moderation; and privacy-preserving collective decision making, voting, and governance.</p><p>Similar to FHE, MP-FHE faces challenges around performance, and would benefit from optimizations in bootstrapping and key switching methods, advances in hardware acceleration, and scaling; and on the multi-party side, further research in security guarantees against malicious parties. Existing MP-FHE schemes combine properties of MPC and FHE, and novel frameworks that are inherently unique to MP-FHE are yet to be fully explored.</p><h3>Bibliography</h3><p>Al Badawi, A., &amp; Polyakov, Y. (2023). <a href="https://eprint.iacr.org/2023/149">Demystifying bootstrapping in fully homomorphic encryption</a>. <em>Cryptology ePrint Archive</em>.</p><p>Apple and Google. (2021). <a href="https://covid19-static.cdn-apple.com/applications/covid19/current/static/contact-tracing/pdf/ENPA_White_Paper.pdf">Exposure Notification Privacy-preserving Analytics (ENPA) white paper</a>.</p><p>Bartusek, J., Bergamaschi, T., Khoury, S., Mutreja, S., &amp; Paradise, O. (2024, June). <a href="https://dl.acm.org/doi/abs/10.1145/3662158.3662815">On the Communication Complexity of Secure Multi-Party Computation With Aborts.</a> In <em>Proceedings of the 43rd ACM Symposium on Principles of Distributed Computing</em> (pp. 480–491).</p><p>Bellare, M., Hoang, V. T., &amp; Rogaway, P. (2012, October). <a href="https://eprint.iacr.org/2012/265.pdf">Foundations of garbled circuits.</a> In <em>Proceedings of the 2012 ACM conference on Computer and communications security</em> (pp. 784–796).</p><p>Brakerski, Z., Gentry, C., &amp; Vaikuntanathan, V. (2014). (Leveled) fully homomorphic encryption without bootstrapping. <em>ACM Transactions on Computation Theory (TOCT)</em>, 6(3), 1–36.</p><p>Chen, E., Lawrence, B., &amp; Zhang, Y. X. (Eds.). (2025). <a href="https://github.com/0xPARC/0xparc-intro-book"><em>Programmable Cryptography: Four Easy Pieces</em></a>. 0xPARC.</p><p>Cheon, J. H., Kim, A., Kim, M., &amp; Song, Y. (2017). <a href="https://eprint.iacr.org/2016/421">Homomorphic encryption for arithmetic of approximate numbers</a>. In <em>Advances in cryptology–ASIACRYPT 2017: 23rd international conference on the theory and applications of cryptology and information security, Hong kong, China, December 3–7, 2017, proceedings, part i 23</em> (pp. 409–437). Springer International Publishing.</p><p>Corrigan-Gibbs, H., &amp; Boneh, D. (2017). <a href="https://www.usenix.org/conference/nsdi17/technical-sessions/presentation/corrigan-gibbs">Prio: Private, robust, and scalable computation of aggregate statistics</a>. In <em>14th USENIX symposium on networked systems design and implementation (NSDI 17)</em> (pp. 259–282).</p><p>Damgård, I., Pastro, V., Smart, N., &amp; Zakarias, S. (2012, August). <a href="https://eprint.iacr.org/2011/535">Multiparty computation from somewhat homomorphic encryption.</a> In <em>Annual Cryptology Conference</em> (pp. 643–662). Berlin, Heidelberg: Springer Berlin Heidelberg.</p><p>Ducas, L., &amp; Micciancio, D. (2015, April). <a href="https://link.springer.com/chapter/10.1007/978-3-662-46800-5_24">FHEW: bootstrapping homomorphic encryption in less than a second</a>. In <em>Annual international conference on the theory and applications of cryptographic techniques</em> (pp. 617–640). Berlin, Heidelberg: Springer Berlin Heidelberg.</p><p>Fan, J., &amp; Vercauteren, F. (2012). <a href="https://eprint.iacr.org/2012/144">Somewhat practical fully homomorphic encryption.</a> <em>Cryptology ePrint Archive</em>.</p><p>Gamiz, I., Regueiro, C., Lage, O., Jacob, E., &amp; Astorga, J. (2025). <a href="https://link.springer.com/article/10.1007/s10207-024-00939-4">Challenges and future research directions in secure multi-party computation for resource-constrained devices and large-scale computations</a>. <em>International Journal of Information Security</em>, <em>24</em>(1), 1–29.</p><p>Gentry, C. (2009, May). <a href="https://dl.acm.org/doi/abs/10.1145/1536414.1536440">Fully homomorphic encryption using ideal lattices.</a> In <em>Proceedings of the forty-first annual ACM symposium on Theory of computing</em> (pp. 169–178).</p><p>Gong, Y., Chang, X., Mišić, J., Mišić, V. B., Wang, J., &amp; Zhu, H. (2024). <a href="https://link.springer.com/article/10.1186/s42400-023-00187-4">Practical solutions in fully homomorphic encryption: a survey analyzing existing acceleration methods</a>. <em>Cybersecurity</em>, <em>7</em>(1), 5.</p><p>I. Chillotti, N. Gama, M. Georgieva and M. Izabachene, <a href="https://eprint.iacr.org/2018/421">TFHE: fast fully homomorphic encryption over the torus</a>, J. Cryptology, 33 (2020) no. 1 34–91.</p><p>Jadoul, R., van Leeuwen, B., &amp; Zajonc, O. (2025). <a href="https://eprint.iacr.org/2025/965">Multiparty FHE Redefined: A Framework for Unlimited Participants</a>. <em>Cryptology ePrint Archive.</em></p><p>Jin, S., Gu, Z., Li, G., Chen, D., Koç, C. K., Cheung, R. C., &amp; Dai, W. (2024). <a href="https://eprint.iacr.org/2024/1629">Efficient Key-Switching for Word-Type FHE and GPU Acceleration</a>. <em>Cryptology ePrint Archive</em>.</p><p>Joye, M. (2021). <a href="https://eprint.iacr.org/2021/1402">Guide to fully homomorphic encryption over the [discretized] torus</a>. <em>Cryptology ePrint Archive.</em></p><p>Kun, J. (2024, May 4). <a href="https://www.jeremykun.com/2024/05/04/fhe-overview/">A High-Level Technical Overview of Fully Homomorphic Encryption.</a></p><p>Kun, J. (2024, October 29). <a href="https://www.jeremykun.com/fhe-in-production/">Fully Homomorphic Encryption in Production Systems.</a></p><p>Li, M., Yang, Y., Chen, G., Yan, M., &amp; Zhang, Y. (2024, July). <a href="https://dl.acm.org/doi/abs/10.1145/3634737.3644993">SoK: Understanding Design Choices and Pitfalls of Trusted Execution Environments</a>. In <em>Proceedings of the 19th ACM Asia Conference on Computer and Communications Security</em> (pp. 1600–1616).</p><p>Lyubashevsky, V., Peikert, C., &amp; Regev, O. (2010). <a href="https://link.springer.com/chapter/10.1007/978-3-642-13190-5_1">On ideal lattices and learning with errors over rings</a>. In <em>Advances in Cryptology–EUROCRYPT 2010: 29th Annual International Conference on the Theory and Applications of Cryptographic Techniques, French Riviera, May 30–June 3, 2010. Proceedings 29</em> (pp. 1–23). Springer Berlin Heidelberg.</p><p>Mouchet, C., Bossuat, J. P., Troncoso-Pastoriza, J. R., &amp; Hubaux, J. P. (2020). <a href="https://infoscience.epfl.ch/entities/publication/79285058-0d3c-464e-a805-e7bf10ed10a0">Lattigo: A Multiparty Homomorphic Encryption Library in Go</a>. In <em>Proceedings of the 8th Workshop on Encrypted Computing and Applied Homomorphic Cryptography</em> (pp. 64–70).</p><p>Mouchet, C., Troncoso-Pastoriza, J., Bossuat, J. P., &amp; Hubaux, J. P. (2021). <a href="https://infoscience.epfl.ch/entities/publication/8e86d137-68bf-4be1-8f7f-cef26dcb5d7b">Multiparty Homomorphic Encryption from Ring-Learning-with-Errors</a>. <em>Proceedings on Privacy Enhancing Technologies</em>, <em>2021</em>(4), 291–311.</p><p>Rabin, M. O. (2005). <a href="https://eprint.iacr.org/2005/187.pdf">How to exchange secrets with oblivious transfer.</a> <em>Cryptology ePrint Archive</em>.</p><p>Regev, O. (2009). <a href="https://dl.acm.org/doi/abs/10.1145/1568318.1568324">On lattices, learning with errors, random linear codes, and cryptography.</a> <em>Journal of the ACM (JACM)</em>, <em>56</em>(6), 1–40.</p><p>Shamir, A. (1979). <a href="https://dl.acm.org/doi/abs/10.1145/359168.359176">How to share a secret.</a> <em>Communications of the ACM</em>, <em>22</em>(11), 612–613.</p><p>Chen, Z., Wang, J., Chen, L., &amp; Song, X. (2014). <a href="https://onlinelibrary.wiley.com/doi/full/10.1155/2014/983862">A Regev‐Type Fully Homomorphic Encryption Scheme Using Modulus Switching</a>. <em>The Scientific World Journal</em>, <em>2014</em>(1), 983862.</p><p>Zhang, J., Cheng, X., Yang, L., Hu, J., Liu, X., &amp; Chen, K. (2024). <a href="https://arxiv.org/html/2212.01713v4">SoK: Fully homomorphic encryption accelerators</a>. <em>ACM Computing Surveys</em>, <em>56</em>(12), 1–32.</p><h3>Further Reading</h3><p>Multi-party computation</p><ul><li><a href="https://blog.zksecurity.xyz/posts/mpcith-intro/">A Gentle Introduction to the MPC-in-the-Head Transformation</a></li><li><a href="https://www.multipartycomputation.com/tpmpc-2025">Theory and Practice of Multi-Party Computation Workshop (TPMPC)</a></li></ul><p>Shamir’s secret sharing</p><ul><li><a href="https://evervault.com/blog/shamir-secret-sharing">Shamir’s Secret Sharing: Explanation and Visualization</a></li><li><a href="https://practicaldsc.org/wn25/guides/machine-learning/interpolation/#:~:text=of%20polynomials%20shortly.-,Point%20Representation,=%201%20+%202%20x%202%20.">Polynomial interpolation</a></li></ul><p>Garbled circuits:</p><ul><li><a href="https://cronokirby.com/posts/2022/05/explaining-yaos-garbled-circuits/">Explaining Yao’s Garbled Circuits</a></li><li><a href="https://medium.com/zkpass/yaos-classical-garbled-circuits-4cbdbadf288e">Understanding Yao’s Classical Garbled Circuits</a></li><li><a href="https://web.mit.edu/sonka89/www/papers/2017ygc.pdf">A Gentle Introduction to Yao’s Garbled Circuits</a></li></ul><p>Threshold decryption</p><ul><li><a href="https://www.researchgate.net/publication/372803956_Multi-Party_Threshold_Cryptography">Multi-Party Threshold Cryptography</a></li></ul><p>Oblivious transfer</p><ul><li><a href="https://www.youtube.com/watch?v=wE5cl8J27Is&amp;ab_channel=Computerphile">[Video] Oblivious Transfer — Computerphile</a></li><li><a href="https://courses.grainger.illinois.edu/cs498ac3/fa2020/Files/Lecture_20_Scribe.pdf">Oblivious Transfer</a></li></ul><p>Fully homomorphic encryption:</p><ul><li><a href="https://www.jeremykun.com/2024/05/04/fhe-overview/">A High-Level Technical Overview of Fully Homomorphic Encryption</a></li><li><a href="https://www.zama.ai/post/tfhe-deep-dive-part-1">TFHE Deep Dive</a></li><li><a href="https://crypto.stanford.edu/craig/craig-thesis.pdf">A Fully Homomorphic Encryption Scheme</a></li><li><a href="https://www.zama.ai/post/what-is-bootstrapping-homomorphic-encryption">Bootstrapping for Dummies</a></li><li><a href="https://eprint.iacr.org/2021/1402">Guide to Fully Homomorphic Encryption over the [Discretized] Torus</a></li><li><a href="https://www.icms.org.uk/sites/default/files/downloads/Anamaria.pdf?utm_source=chatgpt.com">Challenges and open problems in Fully Homomorphic Encryption</a></li><li><a href="https://www.jeremykun.com/fhe-in-production/">Fully Homomorphic Encryption in Production Systems</a></li></ul><p>Learning with errors</p><ul><li><a href="https://www.youtube.com/watch?v=K026C5YaB3A&amp;t=1s&amp;pp=ygUKY2hhbGsgdGFsaw%3D%3D">[Video] Learning with errors: Encrypting with unsolvable equations</a></li><li><a href="https://www.youtube.com/watch?v=QDdOoYdb748&amp;t=1s">[Video] Lattice-based cryptography: The tricky math of dots</a></li><li><a href="https://cims.nyu.edu/~regev/papers/lwesurvey.pdf">The Learning with Errors Problem</a>, Oded Regev</li><li><a href="https://notes.0xparc.org/notes/public-key-cryptography-from-lwe/">Public-key Cryptography from Learning with Errors</a>, 0xPARC</li></ul><p>General</p><ul><li><a href="https://github.com/0xPARC/0xparc-intro-book">0xPARC Four Easy Pieces</a> — Download <a href="https://github.com/0xPARC/0xparc-intro-book/releases/tag/v1.1.1">easy.pdf</a></li><li><a href="https://0xparc.org/blog/programmable-cryptography-1">Programmable Cryptography (Part 1)</a>, 0xPARC</li></ul><h3>Appendix</h3><h3>Garbled Circuits</h3><p>Garbled circuits can be broken down into the following flow:</p><ol><li>Representing the function as a boolean circuit</li></ol><p>a. The function that both parties want to jointly compute is converted into a boolean circuit made of logic gates. (Any computable function can be represented as a boolean circuit made of logic gates.)</p><p>b. Each gate has input and output wires, each carrying 1 bit.</p><p>2. Garbler assigning wire labels</p><p>a. For each wire in the circuit, the garbler creates two random cryptographic labels representing bit values, one representing 0 and the other representing 1. These labels look essential like random bit-strings.</p><p>3. Garbling the gates</p><p>a. For each gate in the circuit, the garbler encrypts the output wire for every input combination.</p><p>i. For example, an AND gate with inputs A and B would result in 4 encrypted output labels:</p><ol><li>Encₐ₀, ᵦ₀(C⁰)</li><li>Encₐ₀, ᵦ₁(C⁰)</li><li>Encₐ₁, ᵦ₀(C⁰)</li><li>Encₐ₁, ᵦ₁(C¹)</li></ol><p>ii. Based on the input values that the evaluator has, only one of these outputs will be decryptable by the evaluator.</p><p>4. Sending garbled circuit and garbled inputs to evaluator</p><p>a. The garbler sends the <strong>garbled circuit</strong>, made up of these encrypted gate tables, to the evaluator.</p><p>b. The garbler also sends their own input labels directly.</p><p>5. Using oblivious transfer for evaluator’s inputs</p><p>a. For each of the evaluator’s input bits, the garbler holds two input labels.</p><p>b. Using <strong>oblivious transfer</strong>, the evaluator receives the input label corresponding to their secret input bit without revealing their inputs.</p><p>6. Evaluating the encrypted circuit</p><p>a. The evaluator now has all the pieces for evaluating the garbled circuit: the garbler’s input labels, their own input labels (via oblivious transfer), and the garbled circuit, represented as an encrypted gate tables.</p><p>b. Starting with the input wires, the evaluator can evaluate each gate by finding the decryptable row in the gate table, eventually getting to the final gate, resulting in the decrypted output bits.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=441c1ea16179" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Benchmarking zkVMs: Current State and Prospects]]></title>
            <link>https://fenbushicapital.medium.com/benchmarking-zkvms-current-state-and-prospects-ba859b44f560?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba859b44f560</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Wed, 18 Jun 2025 17:43:59 GMT</pubDate>
            <atom:updated>2025-06-20T01:31:10.633Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gdmQK6ADzc3qtlnmKNBR9w.png" /></figure><p>Authored by @grandchildrice.</p><p>Special thanks to <a href="https://fenbushi.vc/">Fenbushi Capital</a> for supporting this research, <a href="https://www.novanet.xyz/">ICME Labs — the operator of Novanet — </a>for covering our ZKProof7 travel expenses, and <a href="https://x.com/ClankPan">ClankPan</a> and <a href="https://yuki.aoki.app/">Yuki Aoki</a> for their substantial code contributions.</p><ul><li><strong>Code:</strong> <a href="https://github.com/grandchildrice/zkvm-benchmarks">Benchmark repository</a></li></ul><h3>Abstract</h3><p>Zero‑Knowledge Virtual Machines (zkVMs) are emerging as fundamental building blocks for Ethereum scaling. However, most existing benchmarks — such as the SP1 launch post, the <a href="https://vac.dev/rlog/zkVM-testing/">Vac report</a>, and <a href="http://zkbenchmarks.com/">zkbenchmarks.com</a> — employ incomparable methodologies, thereby obscuring crucial performance trade‑offs.<br>We introduce a standardized test‑bed and evaluate eight zkVMs — <a href="https://github.com/succinctlabs/sp1">SP1</a>, <a href="https://github.com/risc0/risc0">RISC Zero</a>, <a href="https://github.com/openvm-org/openvm">OpenVM</a>, <a href="https://github.com/brevis-network/pico">Pico</a>, <a href="https://github.com/zkMIPS/zkm">ZKM</a>, <a href="https://github.com/a16z/jolt">Jolt</a>, <a href="https://github.com/nexus-xyz/nexus-zkvm">Nexus</a>, and <a href="https://github.com/ICME-Lab/zkEngine_dev">Novanet</a> — across four computational tasks and three performance metrics (prover time, proof size, and peak RAM utilization). Our findings elucidate which zkVM architectures are optimally suited for specific use‑cases within the Ethereum ecosystem.</p><h3>1. Introduction</h3><h3>1.1 The Concept of zkVM</h3><p>A Zero-Knowledge Virtual Machine (zkVM) represents a technological framework designed to cryptographically verify the correctness of program execution without disclosing inputs or intermediate computational states. The operational workflow typically encompasses three principal stages:</p><ol><li><strong>Execute:</strong> The program undergoes execution, generating a comprehensive record (trace) that captures instructions, CPU states, and memory values at each computational step.</li><li><strong>Prove:</strong> A cryptographic attestation (typically utilizing SNARKs — Succinct Non-interactive Arguments of Knowledge) is constructed based on the execution trace.</li><li><strong>Verify:</strong> The generated proof undergoes validation to confirm computational correctness without necessitating re-execution of the original program.</li></ol><p>A zkVM leverages SNARKs to validate the VM’s state transitions at each step by synthesizing three fundamental cryptographic assurances:</p><ul><li><strong>Read-Write Memory Consistency Proof:</strong> Ensures the integrity and proper manipulation of memory throughout the computation.</li><li><strong>Instruction Encoding Proof:</strong> Validates that executed instructions correspond precisely to their defined encodings.</li><li><strong>Instruction Proof:</strong> Confirms that each instruction produced the correct result according to the VM’s formal specification.</li></ul><p>It is imperative to recognize that zkVMs primarily serve to facilitate the efficient verification of computational integrity rather than privacy preservation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_hkdJovOIbXVpdPg.png" /></figure><h3>1.2 Why Ethereum needs zkVMs</h3><p>Currently, every Ethereum validator re‑executes all transactions, inherently coupling gas costs to raw computational complexity.<br>With zkVMs, an off‑chain prover executes computationally intensive operations, generates a cryptographic proof, and submits only the resultant state root alongside the proof on‑chain. Validators merely verify these proofs — an <em>O(1)</em> process regardless of code complexity. Consequently, transaction fees can remain relatively constant even for demanding operations such as machine learning inference or other computationally intensive smart contract executions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NXBZYjZfxfvH1jj4.png" /></figure><h3>1.3 Goal of this work</h3><p>Numerous zkVM projects assert superior performance metrics, frequently citing benchmarks conducted under heterogeneous conditions, thereby complicating objective comparative analysis. Systems marketed as “fastest” may harbor significant limitations, such as excessive memory requirements or restricted functionality applicable exclusively to specific program categories.</p><p>This research introduces a rigorously standardized benchmarking framework designed for comprehensive comparative analysis of multiple zkVM implementations. We evaluate performance across diverse metrics and representative computational tasks. This methodical approach enables the identification of current performance bottlenecks in zkVM technology and facilitates the formulation of best practices for their effective deployment and utilization.</p><h3>2. Classification and Techniques of zkVM Implementations</h3><h3>2.1 Classification of zkVMs</h3><p>Several proof system architectures predominate in the zkVM landscape:</p><ul><li><strong>FRI-STARK-based:</strong> These systems leverage Fast Reed-Solomon Interactive Oracle Proofs of Proximity (FRI) in conjunction with Scalable Transparent Arguments of Knowledge (STARKs). They typically employ a 32-bit extension field to optimize proof generation efficiency. Projects implementing this approach include SP1, RISC Zero, OpenVM, Pico, ZKM, and Valida.</li><li><strong>Nova-based:</strong> These implementations utilize a folding scheme, a technique that efficiently aggregates multiple proofs incrementally. Novanet and Nexus 1.0 exemplify this methodology.</li><li><strong>Lasso Lookup:</strong> This technique aims to mitigate computational overhead by retrieving execution data from precomputed lookup tables. Jolt employs this innovative approach.</li><li><strong>GKR:</strong> Based on the Goldwasser-Kalai-Rothblum protocol, GKR demonstrates circuit satisfiability layer by layer, typically proceeding from output to input, utilizing a sum-check protocol. Ceno represents a notable implementation of this methodology.</li></ul><p>zkVM architectures can be further categorized into two predominant paradigms:</p><ol><li><strong>vRAM Style:</strong> This architecture segregates the program execution trace according to instruction types (e.g., CPU instructions, memory operations). Circuit proofs are subsequently generated in a data-parallel manner for each group before final aggregation. SP1, RISC Zero, and ZKM exemplify this architectural approach.</li><li><strong>Modular Style:</strong> This methodology defines dedicated circuits or lookup tables for individual opcodes. These are sequentially executed as required, with the resultant proofs aggregated. OpenVM and Jolt adopt this modular architectural design.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MJ_tQAQ5Tz1GKjIc.png" /></figure><h3>2.2 Optimization Techniques</h3><p>Common optimization methodologies employed across diverse zkVM implementations include:</p><ol><li><strong>Precompiles:</strong> Utilizing specially optimized circuits to efficiently prove specific, frequently invoked operations such as large integer arithmetic, floating-point calculations, or cryptographic hash functions (e.g., Keccak).</li><li><strong>Continuation:</strong> Segmenting the program execution trace into multiple discrete sections. This facilitates a concurrent generation of proofs for each segment, potentially reducing overall proving time significantly, particularly for computationally extensive operations.</li><li><strong>À la carte Proving:</strong> Architecting the system such that proof generation costs are incurred exclusively for instructions actually executed during specific CPU cycles, rather than for all potential instructions.</li></ol><h3>3. Benchmark Results and Analysis</h3><p>Benchmarks were conducted on a Linux system equipped with Ubuntu 24.04, 8 virtual CPUs, 192GB of RAM, and an NVIDIA RTX 5090 GPU with 32GB of VRAM.</p><p>The four test programs utilized for evaluation comprised:</p><ol><li>Calculation of the 100,000th Fibonacci number.</li><li>SHA2–2048 hash computation.</li><li>ECDSA signature verification using the secp256k1 curve.</li><li>Simulation of 100 Ethereum Transfer transactions (ETHTransfer).</li></ol><h3>3.1 Execution Time Efficiency (Prover Time)</h3><p>The four bar charts presented below illustrate the proof generation times for each benchmark (ordered from top to bottom: Fibonacci, SHA2–2048, ECDSA k256, 100 ETHTransfer).</p><p>Overall, RISC Zero (GPU), SP1 (GPU), and OpenVM (CPU) demonstrated exceptional time efficiency. Conversely, Pico and Jolt exhibited considerable performance variability contingent upon the specific program being executed.</p><p>Comparative analysis with results obtained on an <a href="https://github.com/grandchildrice/zkvm-benchmarks/blob/master/benchmark_outputs/20250312.ipynb">EC2 g5.x16xlarge instance</a> underscores the dependency of SP1’s GPU performance on underlying hardware capabilities.</p><p>OpenVM’s remarkable performance in CPU execution merits particular attention, suggesting potential advantages inherent to its modular architectural design.</p><p><strong>Fibonacci (100k-th):</strong><br>Proof generation times were:</p><ul><li>SP1 (GPU): 3.4s</li><li>RISC Zero (GPU): 3.6s</li><li>OpenVM (CPU): 7.5s</li><li>Pico (CPU): 20s</li></ul><p>As this program involves straightforward iterative calculations with minimal memory access, the performance advantages conferred by GPU acceleration are particularly pronounced.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1ssSqbbqrq21Ck4G.png" /></figure><p><strong>SHA2–2048:</strong><br>Proof generation times were:</p><ul><li>RISC Zero (GPU): 0.54s</li><li>OpenVM (CPU): 0.99s</li><li>Jolt (CPU): 2.7s</li><li>SP1 (GPU): 12s</li><li>ZKM (CPU): 22s</li></ul><p>For cryptographic operations such as SHA2, precompile-based acceleration represents a prevalent optimization strategy. This explains the impressive speeds achieved by RISC Zero, OpenVM, SP1, and ZKM, which implement relevant precompiles. Jolt, currently lacking SHA2 precompiles, utilizes lookup tables to ameliorate the computational cost of large bit operations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EQUfsDhtd6CIjO8D.png" /></figure><p><strong>ECDSA Verification (secp256k1):</strong><br>Proof generation times were:</p><ul><li>RISC Zero (GPU): 1.0s</li><li>SP1 (GPU): 12s</li><li>OpenVM (CPU): 14s</li><li>Jolt (GPU): 83s</li></ul><p>Analogous to the SHA2 results, RISC Zero, SP1, and OpenVM exhibited superior performance attributable to their implementation of precompiles for requisite cryptographic operations. Jolt’s comparative inefficiency relative to its SHA2 performance stems from its current absence of lookup table support for the finite field arithmetic necessary for ECDSA.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*t6Ew0B_wivuWT0sG.png" /></figure><p><strong>100 ETHTransfer Transactions:</strong><br>Proof generation times were:</p><ul><li>RISC Zero (GPU): 7.3s</li><li>OpenVM (CPU): 7.6s</li><li>SP1 (GPU): 13s</li><li>Jolt (GPU): 82s</li></ul><p>Once again, RISC Zero, OpenVM, and SP1 demonstrated superior performance, benefiting from precompiles for essential EVM operations such as Keccak hashing. The capability to generate proofs for simulated EVM execution in approximately 7 seconds, even on desktop-class hardware, represents a significant technological milestone. For comparative analyses involving continuous EVM block execution, refer to the <a href="https://ethproofs.org/">ETHProofs website</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SFaaKcEeMQzv89Ey.png" /></figure><p><strong>Scalability with Input Size (Fibonacci):</strong><br>When increasing the Fibonacci input from the 10th term to the 100,000th term, most implementations exhibited proof generation time increases that were approximately linear (or slightly super-linear) with input size.</p><ul><li>SP1 (GPU) demonstrated a remarkably gradual increase, showing minimal performance degradation.</li><li>RISC Zero and OpenVM exhibited moderate performance scaling.</li><li>Pico, Jolt, and ZKM manifested the most significant increases in proving time.</li></ul><p>The sub-linear scaling observed in certain zkVMs might be attributed to overhead associated with proof aggregation, particularly in implementations lacking or possessing less optimized continuation mechanisms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HnThrOsG0TlMzDpV.png" /></figure><h3>3.2 Memory Efficiency (Peak Memory Usage)</h3><p>The graph below summarizes the comparative peak memory utilization (CPU execution only). For each project, four bars represent memory consumption for the Fibonacci, SHA2, ECDSA, and ETHTransfer benchmarks, respectively.</p><h3>Peak Memory Usage (Selected Examples):</h3><ul><li><strong>100k-Fibonacci:</strong></li><li>RISC Zero (GPU): 0.63GB (GPU Mem)</li><li>SP1 (GPU): 1.3GB (GPU Mem)</li><li>Nexus/Novanet (CPU): ~4GB</li><li>OpenVM (CPU): 8.2GB</li><li>RISC Zero (CPU): 9.2GB</li><li>Pico (CPU): 21GB</li><li>Jolt (CPU): 28GB</li><li>ZKM (CPU): 82GB</li><li><strong>ECDSA Verification (k256):</strong></li><li>RISC Zero (GPU): 0.49GB (GPU Mem)</li><li>SP1 (GPU): 1.3GB (GPU Mem)</li><li>Nexus/Novanet (CPU): ~4GB</li><li>RISC Zero (CPU): 4.5GB</li><li>OpenVM (CPU): 11GB</li><li>Pico (CPU): 20GB</li><li>SP1 (CPU): 21GB</li><li>Jolt (CPU): 58GB</li><li>ZKM (CPU): 84GB</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SzpxPY0Qajai7vOH.png" /></figure><p>SP1 (GPU), RISC Zero (GPU), Nexus, and Novanet demonstrated relatively constant memory consumption irrespective of the test program. However, other projects exhibited memory utilization varying significantly depending on program characteristics, likely influenced by data structures and the frequency of memory access operations.</p><p>GPU implementations generally exhibited reduced host (CPU) memory utilization but consumed substantial GPU memory; the benchmarked GPU-accelerated projects necessitated a minimum of 24GB of VRAM.</p><p>Nexus and Novanet maintained consistent memory utilization even as input size increased, though this consistency incurred significantly extended execution times.</p><p>Memory efficiency represents an active area of development, with improvements frequently resulting from the adoption of more efficient memory checking arguments (e.g., polynomial IOPs), utilization of smaller cryptographic fields, and implementation of techniques such as continuation.</p><h3>3.3 Proof Size</h3><p>The graph below presents the generated proof sizes in kilobytes (kB).</p><p><strong>Proof Sizes (Selected Examples — 100k Fibonacci):</strong></p><ul><li>RISC Zero: 222 kB</li><li>Jolt: 232 kB</li><li>ZKM: 415 kB</li><li>OpenVM: 838 kB</li><li>SP1: 1.8 MB</li><li>Novanet, Nexus: Tens of MB</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xiTxXpcQ9itOFpTe.png" /></figure><p>RISC Zero and Jolt consistently produced among the most compact proof sizes across the evaluated benchmarks.</p><p>Conversely, SP1 generated notably expansive proofs (exceeding 6 MB for tasks other than Fibonacci). Proof sizes surpassing 1 MB, as observed with SP1 and ZKM in certain cases, may indicate that their proof aggregation algorithms warrant further optimization.</p><h3>3.4 Performance Summary</h3><p>The following table summarizes the comprehensive performance results.</p><p>Figures representing optimal performance in each category are highlighted in <strong>green</strong>.</p><p>Peak memory utilization compares CPU executions exclusively, as GPU memory allocation is distinct and not directly comparable.</p><p>Overall, RISC Zero’s performance demonstrates exceptional consistency, while SP1, OpenVM, Pico, and Jolt each attain superlative performance in several individual categories.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*TZERtPc3rU0TbkaS.png" /></figure><h3>3.5 Bottleneck Analysis</h3><p>A detailed performance analysis reveals the following primary bottlenecks for each zkVM implementation:</p><ul><li><strong>Jolt:</strong> CPU limitations include Lasso’s Sumcheck protocol and Spartan proof generation. Memory constraints stem from the Multivariate Polynomial Extension (MLE) construction for lookup tables.</li><li><strong>RISC Zero, SP1, ZKM, OpenVM:</strong> CPU limitations primarily relate to proof recursion and polynomial commitment schemes. Memory constraints frequently arise from Merkle Tree construction.</li><li><strong>Ceno:</strong> The GKR Sumcheck protocol constitutes a bottleneck for both CPU and memory resources.</li><li><strong>Nexus, Novanet:</strong> The folding scheme employed for proof aggregation represents the principal limitation affecting both CPU and memory performance.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*36ncCKq9VNw2Wpgm.png" /></figure><h3>4. Best Practices for Selecting zkVMs</h3><p>Based on our comprehensive benchmark results, the following guidelines can facilitate the selection and configuration of appropriate zkVM technologies:</p><h3>4.1 Selection Criteria Based on Performance</h3><ul><li>For general-purpose applications requiring a balanced combination of speed, memory efficiency, and proof size, RISC Zero, SP1 (particularly with GPU acceleration), and OpenVM emerge as compelling candidates. RISC Zero, specifically, demonstrated consistently exceptional performance across all evaluated computational tasks.</li><li>For applications with highly specialized or computationally intensive requirements not addressed by standard precompiles, the implementation of custom precompiles within a selected zkVM framework is strongly recommended.</li></ul><h3>4.2 GPU Specifications</h3><ul><li>Utilizing the GPU-accelerated variants of SP1 and RISC Zero necessitates graphics processing units with a minimum of 24GB of VRAM for the benchmarked computational tasks.</li></ul><h3>4.3 Memory Specifications</h3><ul><li>A minimum of 32GB of system RAM is advisable for effective operation of most zkVMs in general-purpose applications.</li><li>For memory-intensive zkVMs such as Jolt or ZKM, particularly when processing complex programs, 128GB of RAM or greater might be necessary to prevent performance degradation or memory exhaustion errors.</li></ul><h3>5. Discussion</h3><h3>5.1 Trade-offs Between Security and Performance</h3><p>A legitimate concern regarding the equitability of these benchmark results pertains to varying levels of security assurance across the evaluated zkVM projects. Optimization exclusively for performance can potentially compromise security guarantees.</p><p>Currently, numerous zkVM projects remain in development phases and may lack comprehensive security validation. For example:</p><ul><li><a href="https://github.com/succinctlabs/sp1/tree/dev/audits">The latest version of SP1 has not undergone a complete audit.</a></li><li><a href="https://blog.lambdaclass.com/responsible-disclosure-of-an-exploit-in-succincts-sp1-zkvm-found-in-partnership-with-3mi-labs-and-aligned-which-arises-from-the-interaction-of-two-distinct-security-vulnerabilities/">Soundness bugs were previously reported in SP1 (since remediated).</a></li><li><a href="https://github.com/a16z/jolt/blob/945c0328a313c25d61975c576a17e368bed2421b/Cargo.toml#L3">Jolt remains in an early development stage (v0.1).</a></li></ul><p>Performance comparisons may prove misleading if security levels differ substantially. Future zkVM evaluations should ideally incorporate assessments of security maturity, including factors such as formal verification efforts, completed third-party audits, and the existence of rigorous safety proofs, to provide a more comprehensive comparative analysis.</p><h3>5.2 Technical Differences in GPU Implementations</h3><p>Both SP1 and RISC Zero offer GPU acceleration, yet their performance characteristics diverge significantly, suggesting fundamental architectural variations.</p><p>A comparative analysis based on their respective GitHub implementations reveals distinct design philosophies:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*N1-j0GuqdTy78GgK.png" /></figure><p>It should be noted that Jolt’s GPU implementation currently remains partial and subject to substantial architectural modifications.</p><p>Comparing the more mature GPU implementations, SP1’s approach, while potentially offering superior maintainability through abstraction of CUDA execution via the Moongate server, may incur additional overhead resulting from the separation of certain processes from the GPU kernel. Conversely, RISC Zero’s direct GPU access through its Hardware Abstraction Layer (HAL), despite potentially entailing higher initial implementation complexity when incorporating new CUDA-accelerated functions, aims to achieve performance closer to theoretical limits by minimizing abstraction layers.</p><h3>6. Conclusion</h3><p>This comprehensive study presents a standardized benchmark comparing eight zkVM implementations, evaluating their performance characteristics and potential as scaling solutions specifically relevant to the Ethereum ecosystem.</p><p>Our findings indicate that RISC Zero, OpenVM, and SP1 demonstrate particularly robust performance, especially in executing EVM-related computational tasks. This positions them as promising candidates for integration into Ethereum scaling solutions, such as Layer 2 rollups or specialized co-processors, thereby contributing directly to enhancing the network’s throughput and operational efficiency.</p><p>RISC Zero, notably, exhibited exceptional efficiency across key metrics relevant to blockchain applications. It completed proof generation for 100 simulated ETHTransfer executions in merely 7.3 seconds (GPU), maintained peak CPU memory utilization below 10GB (with GPU memory consumption under 1GB), and produced remarkably compact proofs (approximately 222kB for Fibonacci computation). These results highlight its significant potential for applications demanding high throughput, minimal resource consumption, and low on-chain verification costs — critical factors for enhancing Ethereum’s scalability and improving overall user experience.</p><p>The comparative benchmark results provide invaluable data for developers, researchers, and decision-makers within the Ethereum community. By offering a clearer understanding of current capabilities and performance trade-offs among diverse zkVM technologies, this research aims to inform the selection and development of efficient and robust scaling solutions, ultimately contributing to the continued evolution and expansion of the Ethereum network.</p><h3>References</h3><p>Additional sources are cited inline via hyperlinks.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba859b44f560" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[zkVM 基准评估：现状与前景]]></title>
            <link>https://fenbushicapital.medium.com/zkvm-%E5%9F%BA%E5%87%86%E8%AF%84%E4%BC%B0-%E7%8E%B0%E7%8A%B6%E4%B8%8E%E5%89%8D%E6%99%AF-cdb236785a96?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/cdb236785a96</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Tue, 17 Jun 2025 22:23:17 GMT</pubDate>
            <atom:updated>2025-06-17T22:23:17.999Z</atom:updated>
            <content:encoded><![CDATA[<p>特别感谢<a href="https://fenbushi.vc/">分布式资本（Fenbushi Capital</a>）对本项研究的支持，<a href="https://www.novanet.xyz/">ICME Labs</a>（Novanet 的运营方）为我们参加 ZKProof7 提供差旅资助，以及 <a href="https://x.com/ClankPan">ClankPan</a> 和 <a href="https://yuki.aoki.app/">Yuki Aoki</a>在代码方面的重大贡献。</p><ul><li>代码<strong>:</strong> <a href="https://github.com/grandchildrice/zkvm-benchmarks">Benchmark repository</a></li></ul><p>零知识虚拟机 (zkVM) 正在成为以太坊扩容的基础构件。然而，现有的基准测试 — — 例如 SP1 发布文章中的基准测试如the<a href="https://vac.dev/rlog/zkVM-testing/"> Vac report</a>, 和<a href="http://zkbenchmarks.com/"> zkbenchmarks.com</a> — 大多采用不一致的方法，从而掩盖了关键的性能权衡。</p><p>我们引入了一个标准化的测试平台，并评估了八种 zkVM — — <a href="https://github.com/succinctlabs/sp1">SP1</a>,<a href="https://github.com/risc0/risc0"> RISC Zero</a>,<a href="https://github.com/openvm-org/openvm"> OpenVM</a>,<a href="https://github.com/brevis-network/pico"> Pico</a>,<a href="https://github.com/zkMIPS/zkm"> ZKM</a>,<a href="https://github.com/a16z/jolt"> Jolt</a>,<a href="https://github.com/nexus-xyz/nexus-zkvm"> Nexus</a>,和<a href="https://github.com/ICME-Lab/zkEngine_dev"> Novanet</a> — — 在四种计算任务和三种性能指标（证明者时间、证明大小和峰值 RAM 利用率）上的表现。我们的研究结果阐明了哪些 zkVM 架构最适合以太坊生态系统中的特定用例。</p><h3>1. 引言</h3><h4>1.1 zkVM的概念</h4><p>零知识虚拟机 (zkVM) 代表了一种技术框架，旨在在不泄露输入或中间计算状态的情况下，以密码学方式验证程序执行的正确性。其操作流程通常包括三个主要阶段：</p><ol><li>执行（Execute）： 程序被执行，并生成一个完整的记录（称为执行轨迹），该记录捕捉了每一步计算中的指令、CPU 状态和内存数值。</li><li>证明（Prove）： 基于执行轨迹，构造一个加密证明（通常使用 SNARK — — 简洁非交互式知识论证）来证明计算过程的正确性。</li><li>验证（Verify）： 对生成的证明进行验证，以确认计算结果的正确性，而无需重新执行原始程序。</li></ol><p>zkVM 利用 SNARKs（简洁非交互式知识论证）来验证虚拟机每一步状态转换的正确性，具体通过构建以下三种基础的加密保障来实现：</p><ul><li>读写内存一致性证明（Read-Write Memory Consistency Proof）： 确保在整个计算过程中内存的一致性和操作的正确性。</li><li>指令编码证明（Instruction Encoding Proof）： 验证所执行的指令与其定义的编码严格一致。</li><li>指令执行证明（Instruction Proof）： 确认每条指令根据虚拟机的形式化规范产生了正确的结果。</li></ul><p>必须明确的是，zkVM 的核心用途在于实现对计算完整性的高效验证，而非隐私保护。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v_EMvD9sIeaDHhI3NP--Lg.png" /></figure><h4>1.2 为什么以太坊需要 zkVM</h4><p>目前，以太坊的每一个验证者都需要重新执行所有交易，这使得 Gas 成本与原始计算复杂度直接相关联。引入 zkVM 后，计算密集型操作由链下的证明者（prover）执行，并生成加密证明，然后只将结果状态根和该证明提交到链上。验证者仅需验证这些证明 — — 这是一个与代码复杂度无关、时间复杂度为 O(1) 的过程。因此，即便是如机器学习推理或其他高强度计算的智能合约执行，交易费用也能维持在一个相对稳定的水平。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T5hsRc60XVkuFzTTGLg58w.png" /></figure><h4>1.3 本研究的目标</h4><p>众多 zkVM 项目声称具备优越的性能指标，但这些指标往往是在不同测试条件下得出的，从而使得客观的对比分析变得复杂。某些被宣传为“最快”的系统，可能存在显著局限性，例如内存需求过高，或仅能支持特定类型程序的有限功能。</p><p>本研究提出了一个经过严格标准化的基准测试框架，旨在对多个 zkVM 实现进行全面对比分析。我们从多种性能指标出发，结合具有代表性的计算任务进行评估。这种系统化方法有助于识别当前 zkVM 技术的性能瓶颈，并为其高效部署与应用制定最佳实践。</p><h3>2. 零知识虚拟机（zkVM）实现的分类与技术</h3><h4>2.1 zkVM 的分类</h4><p>在 zkVM 领域，目前主要存在以下几种主流的证明系统架构：</p><ul><li><strong>FRI-STARK-based:</strong> 此类系统结合了快速里德-所罗门交互式接近性证明（FRI）与可扩展的透明知识论证（STARKs）。它们通常使用 32 位扩展域，以优化证明生成的效率。代表项目包括：SP1、RISC Zero、OpenVM、Pico、ZKM 和 Valida。</li><li><strong>Nova-based: </strong>这类实现采用折叠（folding）方案，一种能够高效地逐步聚合多个证明的技术。Novanet 和 Nexus 1.0 是此方法的代表。</li><li><strong>Lasso Lookup:</strong> 该技术通过从预计算的查找表中提取执行数据，以降低计算开销。Jolt 项目采用了这种创新方法。</li><li><strong>GKR:</strong> 基于 Goldwasser-Kalai-Rothblum 协议，GKR 技术通过逐层展示电路可满足性，通常从输出层向输入层推进，并借助 sum-check 协议来完成。Ceno 是此方法的重要实现。</li></ul><p>此外，zkVM 架构还可以进一步细分为两种主要范式：</p><ol><li><strong>vRAM Style:</strong> 这种架构根据指令类型（例如 CPU 指令、内存操作）将程序执行轨迹进行分组处理。每个组随后以数据并行方式生成电路证明，并最终进行聚合。SP1、RISC Zero 和 ZKM 采用了该架构风格。</li><li><strong>Modular Style:</strong> 该方法为每个操作码定义专用电路或查找表，并根据需要依次执行，再将所有结果证明进行聚合。OpenVM 和 Jolt 采用了此类模块化设计。</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cjVq7aUtO15PrQr9kEVSng.png" /></figure><h4>2.2 优化技术</h4><p>在各类 zkVM 实现中，常见的优化方法包括：</p><ol><li><strong>预编译（Precompiles</strong>）<strong>:</strong> 利用专门优化的电路，高效证明特定且经常调用的操作，例如大整数运算、浮点计算或加密哈希函数（如 Keccak）。</li><li><strong>连续性证明（Continuation）:</strong> 将程序的执行轨迹划分为多个独立的片段。这样可以并行地为每个片段生成证明，特别是在处理计算密集型操作时，有助于显著减少整体的证明时间。</li><li><strong>按需生成（À la carte Proving）:</strong> 通过架构设计，使得证明开销仅在实际执行的指令上发生，而非为所有可能的指令都生成证明，从而提升效率和灵活性。</li></ol><h3>3. 基准测试结果与分析</h3><p>本次基准测试在一台运行 Ubuntu 24.04 的 Linux 系统上进行，配备有 8 个虚拟 CPU、192GB 内存，以及一块配备 32GB 显存的 NVIDIA RTX 5090 GPU。</p><p>测试评估所使用的四个程序包括：</p><ol><li>计算第 100,000 个斐波那契数（Fibonacci）</li><li>进行 SHA2–2048 哈希计算</li><li>基于 secp256k1 椭圆曲线的 ECDSA 签名验证</li><li>模拟 100 笔以太坊转账交易（ETHTransfer）</li></ol><h4>3.1 执行时间效率（Prover Time）</h4><p>下方的四个柱状图展示了各项基准测试的证明生成时间（从上到下依次为：Fibonacci、SHA2–2048、ECDSA k256、100 ETHTransfer）。</p><p>整体而言，RISC Zero（GPU）、SP1（GPU）以及 OpenVM（CPU） 在时间效率方面表现出色。相比之下，Pico 和 Jolt 的性能则因具体程序而异，波动较大。</p><p>与在 <a href="https://github.com/grandchildrice/zkvm-benchmarks/blob/master/benchmark_outputs/20250312.ipynb">EC2 g5.x16xlarge instance</a>实例上的测试结果对比显示，SP1 的 GPU 性能在很大程度上依赖于底层硬件配置。</p><p>值得关注的是，OpenVM 在 CPU 执行方面的出色表现，这可能表明其模块化架构设计在效率上具有一定优势。</p><p><strong>Fibonacci（第 100,000 项）证明生成时间如下：</strong></p><ul><li>SP1 (GPU): 3.4s</li><li>RISC Zero (GPU): 3.6s</li><li>OpenVM (CPU): 7.5s</li><li>Pico (CPU): 20s</li></ul><p>由于该程序主要涉及简单的迭代计算且内存访问极少，因此 GPU 加速所带来的性能优势尤为明显</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fmcF65EGpvL7a-vfYiVapw.png" /></figure><p><strong>SHA2–2048 哈希计算证明生成时间如下：</strong></p><ul><li>RISC Zero (GPU): 0.54s</li><li>OpenVM (CPU): 0.99s</li><li>Jolt (CPU): 2.7s</li><li>SP1 (GPU): 12s</li><li>ZKM (CPU): 22s</li></ul><p>对于 SHA2 这类加密操作，基于预编译（Precompile）的加速机制是当前主流的优化策略。这也解释了 RISC Zero、OpenVM、SP1 和 ZKM 在该测试中实现较快速度的原因 — — 它们都实现了相关的预编译模块。Jolt 当前尚未支持 SHA2 的预编译优化，因此通过查找表（lookup table）来降低大规模位运算的计算开销，从而弥补性能不足。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Dz2LS6i9wU40v2k-UArZqA.png" /></figure><p><strong>ECDSA Verification (secp256k1)证明生成时间如下:</strong></p><ul><li>RISC Zero (GPU): 1.0s</li><li>SP1 (GPU): 12s</li><li>OpenVM (CPU): 14s</li><li>Jolt (GPU): 83s</li></ul><p>与 SHA2 的测试结果相似，RISC Zero、SP1 和 OpenVM 的出色性能同样得益于它们对相关加密操作的预编译支持（precompiles）。这些预编译模块大幅降低了执行复杂加密算法所需的计算成本。Jolt 在本项测试中表现较差，远不如其在 SHA2 测试中的相对效率，主要原因在于 其目前尚未支持用于 ECDSA 的有限域运算查找表（lookup tables），这导致其在处理该类复杂算术运算时计算开销显著上升。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6kCwtQhq5X3ACIDOSKFEuA.png" /></figure><p><strong>100 ETHTransfer 交易模拟证明生成时间如下：</strong></p><ul><li>RISC Zero (GPU): 7.3s</li><li>OpenVM (CPU): 7.6s</li><li>SP1 (GPU): 13s</li><li>Jolt (GPU): 82s</li></ul><p>再一次，RISC Zero、OpenVM 和 SP1 表现出色，得益于它们对 EVM 关键操作（如 Keccak 哈希）的预编译支持。这种优化显著提升了模拟以太坊交易执行的证明生成速度。值得强调的是：即便在桌面级硬件上，能够在大约 7 秒内完成模拟 EVM 执行的证明生成，已代表 zkVM 技术的一个重要里程碑。如需进一步比较连续 EVM 区块执行的性能，建议访问 <a href="https://ethproofs.org/">ETHProofs website</a> 网站 以获取更详细的数据与分析。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s3Yqa6_p9HA-_rXjMzf34w.png" /></figure><p><strong>Scalability with Input Size (Fibonacci) 随输入规模扩展的可伸缩性（Fibonacci 测试）:</strong><br>当 Fibonacci 序列的输入规模从第 10 项提升到第 100,000 项时，大多数 zkVM 实现的证明生成时间呈现出近似线性或略超线性增长趋势。</p><ul><li>SP1（GPU） 表现尤为出色，增长幅度极为缓慢，几乎没有明显的性能下降。</li><li>RISC Zero 和 OpenVM 的性能增长保持在中等水平，显示出一定的扩展能力。</li><li>Pico、Jolt 和 ZKM 的证明时间增长最为显著。</li></ul><p>在某些 zkVM 实现中观察到的亚线性（sub-linear）增长可能并非计算效率本身提升，而是与证明聚合过程中产生的额外开销有关，尤其体现在缺乏或未充分优化连续性证明机制（continuation mechanisms） 的架构中。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i4DRJKbnT5YQy9q44TrqNA.png" /></figure><h4>3.2 内存效率（峰值内存使用）</h4><p>下图（原文所附）展示了不同 zkVM 项目在四个基准任务中的 峰值内存使用情况（仅限 CPU 执行模式，部分含 GPU 数据）。对于每个项目，图中四个柱状分别代表其在 Fibonacci、SHA2、ECDSA 和 ETHTransfer 测试中的内存占用。</p><p><strong>峰值内存使用示例（部分项目）：</strong></p><ul><li><strong>100k-Fibonacci:</strong></li><li>RISC Zero (GPU): 0.63GB (GPU Mem)</li><li>SP1 (GPU): 1.3GB (GPU Mem)</li><li>Nexus/Novanet (CPU): ~4GB</li><li>OpenVM (CPU): 8.2GB</li><li>RISC Zero (CPU): 9.2GB</li><li>Pico (CPU): 21GB</li><li>Jolt (CPU): 28GB</li><li>ZKM (CPU): 82GB</li><li><strong>ECDSA Verification (k256):</strong></li><li>RISC Zero (GPU): 0.49GB (GPU Mem)</li><li>SP1 (GPU): 1.3GB (GPU Mem)</li><li>Nexus/Novanet (CPU): ~4GB</li><li>RISC Zero (CPU): 4.5GB</li><li>OpenVM (CPU): 11GB</li><li>Pico (CPU): 20GB</li><li>SP1 (CPU): 21GB</li><li>Jolt (CPU): 58GB</li><li>ZKM (CPU): 84GB</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*atjywwBFwR6mOGF-0EQ4Qg.png" /></figure><p>SP1（GPU）、RISC Zero（GPU）、Nexus 和 Novanet 表现出相对恒定的内存使用，无论测试程序如何变化。然而，其他项目的内存占用则因程序特性而大幅波动，这种差异很可能受到数据结构设计以及内存访问频率的影响。</p><p>GPU 实现方案通常会降低主机（CPU）内存的使用，但会消耗大量的 GPU 显存；在本次基准测试中，所有基于 GPU 加速的项目都至少需要 24GB 显存。</p><p>Nexus 和 Novanet 即使在输入规模扩大时，也能维持稳定的内存使用量，但这种一致性是以显著延长的执行时间为代价的。</p><p>内存效率仍是 zkVM 技术中的一个积极研究方向，常见的优化方法包括：采用更高效的内存一致性证明机制（如多项式交互式证明，polynomial IOPs）、使用更小的加密字段，以及引入如“continuation（连续证明）”等优化技术。</p><h4>3.3 <strong>证明大小（</strong>Proof Size）</h4><p>下图展示了各个 zkVM 项目生成的证明文件大小（单位为千字节 kB）。</p><p><strong>Proof Sizes (Selected Examples — 100k Fibonacci):</strong></p><ul><li>RISC Zero: 222 kB</li><li>Jolt: 232 kB</li><li>ZKM: 415 kB</li><li>OpenVM: 838 kB</li><li>SP1: 1.8 MB</li><li>Novanet, Nexus: Tens of MB</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OyFc7qN2JC7iycKKP_8Faw.png" /></figure><p>在所有基准测试中，RISC Zero 和 Jolt 一贯生成最为紧凑的证明文件，在文件体积方面表现优异。</p><p>相反，SP1 的证明显著更大（在 Fibonacci 以外的任务中超过 6 MB）。在部分场景下，SP1 和 ZKM 所生成的证明超过 1 MB，这可能表明其证明聚合算法仍有进一步优化的空间。</p><h4>3.4 <strong>性能总结</strong></h4><p>下表汇总了各 zkVM 项目的综合性能结果。</p><p>在每个指标类别中表现最优的数据以<strong>绿色</strong>标注（原文图表所示）。</p><p>峰值内存使用仅比较 CPU 执行模式，因为 GPU 的显存分配机制不同，无法直接进行对比。</p><p>综合来看，RISC Zero 的整体性能展现出极高的一致性与稳定性；与此同时，SP1、OpenVM、Pico 和 Jolt 也在多个单项指标中取得了优异的表现。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1VgVI9oKrRREwze3YUkd4A.png" /></figure><h4>3.5 <strong>瓶颈分析</strong></h4><p><strong>通过详细的性能分析，识别出各 zkVM 实现中的主要性能瓶颈如下：</strong></p><ul><li><strong>Jolt:</strong> 在 CPU 层面，其瓶颈主要来自 Lasso 的 Sumcheck 协议 和 Spartan 证明生成过程。内存方面的限制则主要源于用于查找表的多变量多项式扩展（MLE）构建。</li><li><strong>RISC Zero, SP1, ZKM, OpenVM:</strong> CPU 层面的主要瓶颈集中在 递归证明（proof recursion） 和 多项式承诺方案（polynomial commitment schemes） 上。内存开销方面，常见的瓶颈则源自 Merkle 树的构建。</li><li><strong>Ceno:</strong> GKR 的 Sumcheck 协议 同时对 CPU 和内存资源 构成了主要限制。</li><li><strong>Nexus, Novanet:</strong> 其采用的折叠式证明聚合方案（folding scheme） 是导致 CPU 与内存性能受限的主要因素。</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g6Qrg-AE3e8t1cfhER9EJw.png" /></figure><h3>4. <strong>选择 zkVM 的最佳实践</strong></h3><p>基于本次全面的基准测试结果，以下指南可用于帮助选择与配置适合的 zkVM 技术：</p><h4>4.1 <strong>基于性能的选择标准</strong></h4><ul><li>对于通用应用场景，若需在速度、内存效率与证明大小之间取得良好平衡，RISC Zero、SP1（尤其是在 GPU 加速模式下）以及 OpenVM 是极具吸引力的候选方案。尤其是 RISC Zero，在所有评估的计算任务中均展现出卓越且一致的性能表现。</li><li>对于那些具有高度专业化或计算密集型需求、且标准预编译无法满足的应用，强烈建议在选定的 zkVM 框架中实现自定义预编译。</li></ul><h4>4.2 <strong>GPU 配置要求</strong></h4><ul><li>在基准测试的计算任务中，使用 GPU 加速版本的 SP1 和 RISC Zero 需要配备至少 24GB 显存的图形处理器。</li></ul><h4>4.3 <strong>内存配置建议</strong></h4><ul><li>对于大多数 zkVM 的通用应用场景，推荐的系统内存为至少 32GB RAM。</li><li>对于内存使用密集的 zkVM，如 Jolt 或 ZKM，尤其是在运行复杂程序时，建议配置 128GB 或更高内存，以避免性能下降或内存耗尽错误。</li></ul><h3>5. <strong>讨论</strong></h3><h4>5.1 <strong>安全性与性能之间的权衡</strong></h4><p>关于这些基准测试结果的公平性，一个合理的担忧在于被评估的各个 zkVM 项目在安全性保障方面存在不同水平。若仅追求性能优化，可能会以牺牲安全性为代价。</p><p>目前，许多 zkVM 项目仍处于开发阶段，尚未完成全面的安全性验证。例如：</p><ul><li><a href="https://github.com/succinctlabs/sp1/tree/dev/audits">The latest version of SP1 has not undergone a complete audit.</a></li><li><a href="https://blog.lambdaclass.com/responsible-disclosure-of-an-exploit-in-succincts-sp1-zkvm-found-in-partnership-with-3mi-labs-and-aligned-which-arises-from-the-interaction-of-two-distinct-security-vulnerabilities/">Soundness bugs were previously reported in SP1 (since remediated).</a></li><li><a href="https://github.com/a16z/jolt/blob/945c0328a313c25d61975c576a17e368bed2421b/Cargo.toml#L3">Jolt remains in an early development stage (v0.1).</a></li></ul><p>如果安全级别存在显著差异，性能比较可能具有误导性。未来的 zkVM 评估理应纳入对安全成熟度的评估，包括形式化验证工作的开展情况、已完成的第三方审计，以及是否存在严格的安全性证明等因素，以提供更全面的对比分析。</p><h4>5.2 <strong>GPU 实现中的技术差异</strong></h4><p>SP1 和 RISC Zero 均提供了 GPU 加速功能，但它们的性能特性存在显著差异，暗示其底层架构存在根本性的不同。</p><p>通过对它们各自 GitHub 实现的比较分析，可以看出设计理念上的明显差异：</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/954/1*YFp8gLUKCrEzE1ckhcR25Q.png" /></figure><p>应注意，Jolt 的 GPU 实现目前仍不完整，且其架构可能会有重大调整。</p><p>在对比更为成熟的 GPU 实现时，SP1 的方法通过 Moongate 服务器对 CUDA 执行进行抽象，可能在可维护性方面具有优势，但这种方式也可能因将某些处理过程从 GPU 核心中剥离而引入额外开销。相反，RISC Zero 通过其硬件抽象层（HAL）直接访问 GPU，虽然在引入新的 CUDA 加速功能时初期实现复杂度可能更高，但该方式旨在通过减少抽象层，尽可能接近理论性能上限。</p><h3>6. <strong>结论</strong></h3><p>本研究通过标准化基准测试，对八种 zkVM 实现进行了全面比较，评估了它们的性能特征及其作为以太坊扩展解决方案的潜力。</p><p>我们的研究结果表明，RISC Zero、OpenVM 和 SP1 在执行与 EVM 相关的计算任务方面表现尤为强劲，使它们成为有望被集成至以太坊扩展方案（如 Layer 2 Rollup 或专用协处理器）的理想候选，从而直接提升网络的吞吐量与运行效率。</p><p>尤其值得注意的是，RISC Zero 在区块链相关关键指标上展现出卓越的效率：它在 GPU 模式下仅用 7.3 秒 即完成了 100 笔 ETHTransfer 模拟执行的证明生成，CPU 峰值内存消耗控制在 10GB 以下（GPU 显存消耗不到 1GB），并生成了异常紧凑的证明文件（如 Fibonacci 运算约 222kB）。这些表现凸显了其在高吞吐量、低资源占用与低链上验证成本方面的巨大潜力 — — 这些都是提升以太坊可扩展性与用户体验的关键因素。</p><p>本次对比测试结果为以太坊社区的开发者、研究人员与决策者提供了宝贵的数据支持。通过更清晰地了解不同 zkVM 技术的当前能力与性能权衡，本研究旨在为高效、稳健的扩展解决方案的选择与开发提供指导，最终推动以太坊网络的持续演进与扩展。</p><p><strong>参考文献</strong></p><p>额外的参考资料已通过超链接形式在正文中引用。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cdb236785a96" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[PayFi: Endeavoring to Realize Blockchain’s Original Vision]]></title>
            <link>https://fenbushicapital.medium.com/payfi-endeavoring-to-realize-blockchains-original-vision-31a7cf01022d?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/31a7cf01022d</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Thu, 31 Oct 2024 00:00:56 GMT</pubDate>
            <atom:updated>2024-11-23T12:02:56.698Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/902/1*ylYMSmOwOKvi4XuUdfWuPQ.png" /></figure><p><strong>Abstract</strong></p><p>Many modern economic services reflect the importance of the time value of money. Services from payday loans to complex cross-border supply chain financing seek to offer customers flexible solutions to optimize cash flow. Blockchain-based solutions have already begun to enable more timely, efficient, and democratic provision of these crucial services under the collective name “PayFi”. As PayFi grows it may present novel payment solutions and create new economies. However, potential hurdles remain regarding market and regulatory acceptance of PayFi solutions.</p><p><strong>Introduction to PayFi</strong></p><p>PayFi (short for Payment Finance) is a group of services aiming to realize <a href="https://youtu.be/vBc5oqZjZ-0?t=254">the original vision of blockchain</a> as a peer to peer electronic cash system. Unlike decentralized finance (“DeFi”) solutions, which focus on broader financial services through decentralized platforms, PayFi solutions are specifically designed to provide real-time transaction settlement.</p><p>PayFi seeks to leverage cutting-edge blockchain technology to enable instantaneous processing of financial transactions, a crucial feature in supply chain financing, cross-border payments, and invoice financing among others. By integrating blockchain’s transparency, security, and efficiency with rapid transaction processing, PayFi potentially offers a transformative approach to enhance the speed and reliability of real-world financial operations.</p><p>As a novel paradigm, PayFi looks to streamline financial transactions while also democratizing access to liquidity, reducing transaction costs, and expanding global financial accessibility. As PayFi continues to evolve, it hopes to redefine the financial landscape, bridging the gap between traditional financial systems and the new possibilities offered by blockchain.</p><p>PayFi’s emergence may also prove a crucial development for investors seeking to build healthy crypto investment portfolios. Over the last 10 years the amount of value stored on blockchains has soared.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g-L8C0_J6NOj9FvvNBlHhg.png" /><figcaption><a href="https://www.statista.com/statistics/730876/cryptocurrency-maket-value/">Figure 1</a>: The Total Market Cap of Cryptocurrencies from 21 June, 2010 through 29 May, 2024</figcaption></figure><p>Reports suggest that the <a href="https://www.forbes.com/digital-assets/crypto-prices/?sh=b96d18a24785">total value of crypto assets</a> in 2024 is approximately USD 2.41T. Estimates vary, but financial industry experts predict that the total value that will be stored on-chain by 2030 will be <a href="https://www.cointribune.com/en/the-crypto-market-is-heading-towards-10-trillion-in-market-cap/">USD 10T</a> to <a href="https://www.thestreet.com/crypto/markets/cathie-wood-predicts-crypto-price-explosion-to-25-trillion-in-2030">USD 25T</a>. With so much value on-chain, it is important that investors have diverse on-chain investment opportunities to maximize the utility of their assets. While current staking and liquidity pool solutions offer investors some options, on-chain investment opportunities still lag behind TradFi alternatives. PayFi looks to provide the next step in the maturation of on-chain investing by bringing established financial services on-chain and enabling investors to avail themselves of new return-earning opportunities. Importantly, because current on-chain investments are <a href="https://blockworks.co/news/the-investors-guide-to-crypto-correlation">generally correlated</a>, PayFi’s new opportunities may allow for greater diversification of on-chain portfolios.</p><p><strong>PayFi Verses TradFi Alternatives</strong></p><p>PayFi offers several potential advantages over TradFi solutions. Below is a table comparing some of PayFi’s features and advantages over its TradFi counterparts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yGCo_uYvO-ZaO9u5H1tvpA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c_mJtRamB_fv9gnln3bEJQ.png" /></figure><p><strong>Market Sectors for Potential PayFi Solutions</strong></p><p>Multiple market sectors and industries can benefit from well-designed PayFi solutions. Though PayFi is still relatively nascent, several PayFi solutions are available or are being built. The below market map, though not exhaustive, offers a glimpse into some of the projects looking to transform existing markets through PayFi solutions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RMrN-r8672CSIY-MEhsZWQ.jpeg" /></figure><p>In the cross-border payments sector, two PayFi companies, Arf Financial (“Arf”) and Huma Finance (“Huma”), have already gained significant traction.</p><p><em>Arf Financial</em></p><p>Arf is a global transaction services platform that first surpassed <a href="https://arf.one/arfs-onchain-liquidity-volume-surpasses-1-billion/">USD 1B in on-chain liquidity volume</a> in March 2024. Arf’s infrastructure, with native on-ramping and off-ramping, uses real-world asset (“RWA”) based stablecoins to provide quick and efficient cross-border payment solutions. Specifically, Arf, in partnership with Circle, offers short-term USDC-based working capital loans to licensed financial institutions. The institutions use the loans to settle cross-border payments immediately with no additional collateral or prefunding required. Because of the on-chain loaning and settlement, the transactions are transparent and traceable.</p><p>Arf has crafted a careful business model to deal with the regulatory uncertainty that surrounds crypto and PayFi. Regulated under Swiss law, Arf only offers its PayFi cross-border payment services to licensed financial institutions to better ensure the regulatory compliance of all parties in each transaction. Further, Arf checks its borrowers’ quarterly audited financials and performs daily credit risk checks.</p><p>Even with their careful approach, Arf currently processes <a href="https://youtu.be/Smqs0koYdbU?t=933">USD 1B per year in transactions</a>. However, Arf’s co-founder and CEO Al Erhat N. states that the current demand for cross-border payment solutions is USD 1B per month.</p><p>As of October 2024, <a href="https://dune.com/arf_financial/arf-financial">Arf</a> has an on-chain volume of USD 2B and a total loan volume of USD 1B with 0 defaults. Arf also boasts a capital turnover of 3.85 times per month.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0bcLUh5b4xihAg_HRX0ndg.png" /><figcaption><a href="https://dune.com/arf_financial/arf-financial">Figure 3</a>: Arf Financial Statistics</figcaption></figure><p><em>Huma Finance</em></p><p>Another company contributing to PayFi’s development is Huma. Huma is a credit platform that tokenizes receivables and provides risk management services. Huma currently offers cross-border payment financing and digital-asset backed credit cards. Further offerings of instant settlement for RWAs, trade finance, and decentralized physical infrastructure networks (“DePIN”) financing are coming soon.</p><p>As of October 2024, <a href="https://dune.com/huma-finance/huma-overview">Huma</a> has financed total payments of USD 332M and has suffered 0 losses.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z_NLpwvGgZkDh-qgns0PzA.png" /><figcaption><a href="https://dune.com/huma-finance/huma-overview">Figure 4</a>: Huma Financial Statistics</figcaption></figure><p>Moreover, Huma is doing more than just offering PayFi solutions, it is building a PayFi platform for other PayFi apps to operate on. <a href="https://blog.huma.finance/payfi-the-new-frontier-of-rwa">Huma’s PayFi Stack</a> is an open stack designed to aid in the building of PayFi solutions. The PayFi Stack’s consists of the following six layers: transaction layer; currency layer; custody layer; compliance layer; financing layer; and application layer. With the PayFi Stack, Huma aims to provide a blueprint for PayFi solutions to build upon while encouraging cooperation and innovation across each layer and the PayFi Stack itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4r5t065atMwD0hosKwmX4w.png" /><figcaption><a href="https://huma.finance/">Figure 5</a>: Huma Finance’s PayFi Stack Architecture</figcaption></figure><p>Another interesting PayFi project currently being worked on is BSOS’s supply chain finance solution.</p><p><em>Supply Chain Financing</em></p><p>An integral part of international commerce, supply chain financing (“SCF”) can potentially benefit from PayFi. While it can be considered a subsection of cross-border payments, SCF is a massive market and is deserving of individual attention. In 2023 the SCF market had <a href="https://www.coastlinesolutions.com/tfe/pdf/WSCFR_2024.pdf">a global value of USD 2.347T</a>. SCF is prominent because it presents a win-win-win solution for many businesses and financiers.</p><p>In SCF suppliers offer discounts to 3rd party financiers in return for immediate payment. The 3rd party financiers then take full payment from buyers according to their negotiated payment terms, be it 30, 60, or 90 days.</p><p>This process stabilizes the cash flow for both the supplier and the buyer. The supplier receives payment earlier than standard terms require, which may be vital to the continual operation of their business. Likewise, buyers improve their cash flow by delaying payment. Lastly, the financier receives the difference between the discounted price paid to the supplier and the full payment received from the buyer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9W5AFR15ehBBysl5bRTodA.png" /><figcaption><a href="https://www.lbbw.de/articlepage/experience-banking/optimizing-liquidity-supply-chain-finance_7ek5tnimt_e.html">Figure 6</a>: Supply Chain Finance Order Flow</figcaption></figure><p>Traditional SCF requires a complex web of contracts, letters of credit, and communication. PayFi can harness the advantages of blockchain to greatly simplify SCF. With smart contracts at the center of PayFi protocols, communication, documentation, and payments can be processed with much greater efficiency and speed as shown below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7hleAAy2Edq685C8pot3zg.png" /><figcaption><a href="https://www.mdpi.com/0718-1876/16/7/168">Figure 7</a>: Traditional SCF Model vs. Possible PayFi Model</figcaption></figure><p>Additionally, traditional SCF is only available to relatively sophisticated parties, often requiring multiple years of audited financials before financing is available. Under PayFi, however, SCF can be simplified, sped up, and automated according to the parties’ preferences. PayFi’s greater efficiency can bring SCF to more businesses that previously did not have access to the full array of services under the traditional financial system due to size, industry, or location. Such potential democratization of SCF may work toward leveling global playing fields and opening up new supply-chain lines. Small businesses, new businesses, and/or geographically isolated businesses can secure previously unavailable financial services and participate in previously unavailable markets.</p><p><a href="https://www.bsos.co/suplex">BSOS</a> is already working on transforming SCF through PayFi. BSOS seeks to collaborate with diverse capital providers to ensure continued, healthy working capital flows and promote a resilient and sustainable supply chain. So far, BSOS has transacted over USD 20M in volume for over 350 corporations including Adidas and Intel.</p><p>As PayFi continues to grow, opportunities in additional markets will likely become attractive targets for new PayFi solutions. Two markets that may eventually benefit from PayFi solutions are the repurchase agreement market (“the repo market”) and payday loans.</p><p><em>The Repo Market</em></p><p>The repos market’s requirements for speed and transparency are potentially a perfect fit for PayFi’s capabilities. The repo market is where financial institutions engage in short-term borrowing and lending, typically overnight, using securities such as bonds as collateral. It is used by financial institutions to manage liquidity and finance inventories of securities. Central banks also participate in the repo market to implement monetary policy by influencing short-term interest rates. The repo market is beyond massive, the U.S. repo market alone moves <a href="https://www.marketwatch.com/guides/banking/repo-market/">USD 3T every day</a>. In addition to speed and transparency, PayFi’s automation potential could add great value to the repo market and enable financial institutions to more efficiently borrow and lend money to each other to meet their liquidity and accounting demands.</p><p><em>Payday Loans</em></p><p>In the realm of personal finance, one market that may benefit from PayFi’s greater efficiency is payday loans. Payday loans are short-term, high-interest loans designed to provide immediate cash to individuals who need funds before their next paycheck. In 2021 the global payday loan market was worth approximately <a href="https://www.globenewswire.com/en/news-release/2022/05/03/2434258/0/en/2022-Statistics-Payday-Loans-Market-Will-Surpass-USD-42-6-Billion-at-4-1-CAGR-Growth-Vantage-Market-Research.html">USD 33.5B</a>. By 2028 the market is expected to be worth <a href="https://www.globenewswire.com/en/news-release/2022/05/03/2434258/0/en/2022-Statistics-Payday-Loans-Market-Will-Surpass-USD-42-6-Billion-at-4-1-CAGR-Growth-Vantage-Market-Research.html">USD 42.6B</a>. Specifically for the payday loan market, PayFi’s greater transaction speeds, greater accessibility, and greater transparency will be better able to serve borrowers. On the other side, payday loans’ risk-profile and short duration period (<a href="https://paydayloaninfo.org/how-payday-loans-work/">an average of two weeks</a>) will add needed variety and diversification options for crypto investors to consider.</p><p><em>PayFi’s Potential to Enable New Markets</em></p><p>PayFi’s advantages may also allow for the emergence of new markets.</p><p>“<a href="https://youtu.be/vBc5oqZjZ-0?t=680">Buy now pay never</a>” is an interesting concept combining both investment and credit services in one on-chain account. In “buy now pay never” the user deposits crypto assets into a liquidity pool. The user can then purchase goods with their deposited assets. However, instead of the payment being withdrawn from the user’s principal in the account to pay for their purchase, the amount owed is treated as a receivable and is paid off over time by the returns earned in the account.</p><p>PayFi also has the potential to create <a href="https://ethcc.io/archive/The-Emergence-of-PayFi-Realizing-the-Vision-of-Crypto">new economies for influencers and content creators</a>. Revenue for content creators flows in over time as views of their content accumulate. Unfortunately, even for steady creators the accumulation of views, and thus receipt of full payment, may take months or years. The drawn-out nature of the content creator revenue model can lead to inconsistent cash flow and liquidity issues. Much like SCF, PayFi can offer content creators a novel solution for short-term financing. Via liquidity pools, crypto investors can offer content creators advances on their revenue. Creators then will then pay back the advance as revenue flows in.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/678/1*yR7WGS5vSL3my0IrGddkcA.png" /><figcaption><a href="https://ethcc.io/archive/The-Emergence-of-PayFi-Realizing-the-Vision-of-Crypto">Figure 8</a>: Diagram of Potential PayFi Content Creator Financing Model</figcaption></figure><p>Such advances can provide content creators the liquidity they need to continue to create while providing a new investment option for crypto investors to realize returns on their on-chain assets.</p><p>However, such new markets are not without risks. Liquidity can evaporate quickly, borrowers can overestimate the size of future returns, and view counts with streaming revenue may never materialize. As new products are potentially launched both the PayFi solution providers and investors will need to move with caution.</p><p><strong>Hurdles and Obstacles to Overcome</strong></p><p>Though still nascent, PayFi has already shown some of its potential. As it matures, PayFi hopes to provide more and more benefit to businesses and investors. However, there are challenges hindering PayFi’s development.</p><p><em>Blockchain Liveness and Scalability</em></p><p>Payment services are vital to the international economy and any interruption to payments can create large-scale problems. Due to this blockchains need to showcase extreme reliability. Solana, reported as “the vanguard blockchain for institutional payment use-case adoption” has had <a href="https://public.bnbstatic.com/static/files/research/blockchain-payments-a-fresh-start.pdf">7 major outages</a> since its mainnet launch in 2020. While blockchain’s growing pains are understandable, liveness issues are a risk that will likely slow PayFi’s adoption. As blockchains mature and improve PayFi can win the trust of institutions and customers, but it may be a long process.</p><p>Similarly, scalability must improve for PayFi to reach its potential. Visa can process over <a href="https://usa.visa.com/solutions/crypto/deep-dive-on-solana.html">65,000</a> transactions per second (“TPS”). In 2023 Visa processed approximately <a href="https://www.statista.com/statistics/261327/number-of-per-card-credit-card-transactions-worldwide-by-brand-as-of-2011/#:~:text=Nearly%20four%20out%20of%2010,transactions%20per%20day%20in%202023">720M transactions per day</a> for a daily average TPS of 8,300. While Solana has a <a href="https://solana.com/solana-whitepaper.pdf">theoretical TPS of 710,000</a>, to date its highest recorded max daily average TPS is around <a href="https://dailycoin.com/solana-proves-to-be-the-fastest-blockchain-with-1504-tps/">1,500</a>. Currently, Ethereum’s max recorded transaction processing capabilities is <a href="https://chainspect.app/chain/ethereum">62 TPS</a> with the L2 Polygon hitting around <a href="https://chainspect.app/chain/polygon">429 TPS</a>. Arbitrum has hit a max recorded TPS of <a href="https://chainspect.app/chain/arbitrum">944</a>. Similarly, Tron’s max recorded TPS is reported at <a href="https://chainspect.app/chain/tron">272</a> while BNB Chain’s max reported TPS is <a href="https://chainspect.app/chain/bnb-chain">1,731</a>. At present blockchains do not have the scale to match PayFi’s traditional alternatives.</p><p><em>Regulation</em></p><p>PayFi, like all blockchains, is hampered by regulatory uncertainty. Each potential PayFi solution seemingly faces a unique combination of regulatory requirements. Such requirements include factors like the jurisdiction of use (or users) and the TradFi alternative that jurisdictions judge to be closest to the PayFi solution.</p><p>For any cross-border payment PayFi project in the U.S. multiple regulators may claim oversight including; the Federal Reserve, the Department of the Treasury, the Office of Foreign Assets Control, the Office of Comptroller of the Currency, the SEC, the Consumer Financial Protection Bureau, state regulators, and the Financial Crimes Enforcement Network.</p><p>Likewise, provision of PayFi personal credit services in the U.S. is subject to oversight from; the Consumer Financial Protection Bureau, the Office of the Comptroller of the Currency, the Federal Reserve, the Federal Deposit Insurance Corporation, the Federal Trade Commission, state banking regulators, the Office of Foreign Assets Control, and the Financial Crimes Enforcement Network.</p><p>The regulation of payroll services in the U.S. includes oversight by, the IRS, the Department of Labor, the Financial Crimes Enforcement Network, the Consumer Financial Protection Bureau, the Federal Trade Commission, and state regulators.</p><p>Due to this complexity, PayFi regulatory compliance must be approached methodically and on a case-by-case basis.</p><p>For the time being, companies like Arf are creating regulatory precedent and clarity for PayFi through careful and well-thought-out business practices as discussed above.</p><p><em>Institutional Resistance</em></p><p>Much of PayFi’s potential advantage over TradFi stem from greater automation and the cutting out of intermediary parties. These intermediaries, often banks and other well-organized entities, are incentivized to protect the status quo and their established revenue streams. As such, PayFi solutions may likely face institutional pressures, doubts, and dismissals from established TradFi powers.</p><p><em>Volatility</em></p><p>As mentioned above, greater on-chain liquidity can cut both ways. Capital in TradFi is generally stickier while on-chain assets can move quickly. PayFi solutions must be diligent in maintaining the liquidity they rely on to operate. Token and point reward schemes may be useful in maintaining liquidity, but PayFi projects must prioritize competitive returns or they may suddenly find that all of their liquidity has moved to a more efficient competitor.</p><p><strong>Concluding Remarks</strong></p><p>More so than DeFi, PayFi represents the promises of blockchain. PayFi also potentially represents a significant advancement over TradFi alternatives by offering real-time settlements, enhanced security, reduced costs, global accessibility, programmability, transparency, and improved scalability.</p><p>Through continued creativity, effort, and time, PayFi has the potential to become a transformative force in the financial industry that will provide more efficient, secure, and inclusive alternatives to conventional financial systems.</p><p>If anyone wishes to contact me with questions, comments, or observations about PayFi please feel free to DM on X <a href="https://x.com/Lawsker15">@Lawsker15</a>.</p><p><strong>Table Sources</strong></p><p>1 Blockchain Payments: A Fresh Start, p. 16. Binance Research, August, 2024. <a href="https://public.bnbstatic.com/static/files/research/blockchain-payments-a-fresh-start.pdf">Blockchain Payments: A Fresh Start</a></p><p>2 <em>Id</em>. at page 2.</p><p>3 Remittance Prices Worldwide Quarterly, p. 15. The World Bank, March 2024. <a href="https://remittanceprices.worldbank.org/sites/default/files/rpw_main_report_and_annex_q124_final.pdf">Remittance Prices Worldwide — Issue 49, March 2024</a></p><p>4 Expanding Financial Access for Mexico’s Poor and Supporting Economic Stability. World Bank Group, April 9, 2021. <a href="https://www.worldbank.org/en/results/2021/04/09/expanding-financial-access-for-mexico-s-poor-and-supporting-economic-sustainability">Expanding Financial Access for Mexico’s Poor and Supporting Economic Sustainability</a></p><p>5 SWIFT Transfers Explained: How Long Does a SWIFT Transfer Take? Fin.do, February 23, 2022. <a href="https://www.fin.do/blog/62_swift-transfers-explained">SWIFT Transfers Explained: How Long Does a SWIFT Transfer Take?</a></p><p>6 SWIFT vs. Global ACH. Modern Treasury. <a href="https://www.moderntreasury.com/learn/swift-vs-global-ach">SWIFT vs. Global ACH</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=31a7cf01022d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Future of Fiat-to-Crypto: zkP2P]]></title>
            <link>https://fenbushicapital.medium.com/the-future-of-fiat-to-crypto-zkp2p-e317f09b7c1e?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/e317f09b7c1e</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Thu, 27 Jun 2024 20:25:11 GMT</pubDate>
            <atom:updated>2024-06-27T20:25:11.270Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LV6Q6LhOKKAou6bT_Scktg.png" /></figure><h3>The Future of Fiat-to-Crypto: zkP2P</h3><p>In the evolving landscape of cryptocurrency, the need for seamless and secure on-ramping solutions has never been more critical. zkP2P introduces an innovative peer-to-peer fiat-to-crypto on-ramp that not only alleviates centralization in on-ramps, but is one of the few consumer-facing ZK applications that can be used today, no web3 indoctrination required.</p><p>zkP2P is a peer-to-peer fiat-to-crypto on-ramp that makes use of several foundational ZK packages to breathe new utility into existing internet infrastructure. Instead of convincing the Venmos, AliPays, and Revoluts of the world to opt into crypto, zkP2P takes the signatures already in use by their web2 infrastructure to make established fintech crypto-friendly. Users can use the online payment systems they already know and trust when on-ramping fiat. You can trustlessly say “Venmo me for crypto.”</p><h3>zkP2P Flow</h3><p>An example exchange on zkP2P would be exchanging USD on Venmo for USDC:</p><ul><li>A user requests an amount of USDC to be sent</li><li>The user is paired with a liquidity provider to venmo at their requested rate</li><li>The provider’s on-chain funds are held in escrow by the zkP2p smart contract, while the user sends the USD via venmo to the provider</li><li>The user submits the email receipt of the transaction from venmo to zkP2P (which can be done automatically if the user is signed in to zkP2p with their google account).</li><li>Two proofs are generated:</li><li>One using zkEmail to verify that the Venmo’s email server’s signature matches that of the receipt email. This ensures that the email can’t be falsified.</li><li>One using zkRegex to extract relevant information from the receipt email: that the correct amount was sent from the correct user to the correct recipient.</li><li>At this point, all checks have been completed and the funds in escrow are released to the user.</li></ul><h3>Why is this important?</h3><p>zkP2P shows what can be done when existing web2 infrastructure is brought into the blockchain arena. It further showcases how ZK can enable new consumer-facing tools across domains, no relationships needed.</p><h3>Trustless On-Ramping</h3><p>This is a topic that needs no introduction to anyone in web3. zkP2P solves both the itch of onboarding the next [moonshot amount here] users while being regulator friendly. For onboarding, this both de-risks users from scammy onramps and meets the user at their preferred online payment provider. Regulation is beautiful if not immediate: all major payment providers are already regulation-compliant and perform KYC (even offering protections for certain transactions in the case of Venmo). This allows zkP2P to actually use regulation to the user’s benefit; a rare case in the world of web3.</p><h3>Good ol’ IETF</h3><p>Another odd topic for a Web3 article. Another look at the steps above will show that beyond having a wallet (which is changing with support for account abstraction), the user never leaves their safe Venmo haven. This is because of the groundwork of the IETF (the Internet Engineering Task Force, the open governing body of the internet founded in 1986) in the steering of the internet. RFC’s (Requests For Comments; the inspiration for the ERC) govern the internet much the same way as ERCs do, with a similarly open governance process. ERC 6376 defined DKIM in September 2011, which made the e-mail authentication signatures that zkP2P relies on an internet standard. Thanks to this, every email server signs their emails; thanks to zkEmail, we can make proofs for the authenticity of emails, and now thanks to zkP2P we can use these signatures that were originally meant to prevent eavesdropping into useful trustless tools.</p><p>It is worth mentioning now that there is a newer RFC, RFC 9421 which proposes signing REST APIs in much the same way as emails are signed. This would allow proof of any API response; instead of relying on emails, the adoption of this RFC by industry would allow zkP2P and other hopeful ZK teams to verify any semantically meaningful data on the internet that was sent via an API endpoint.</p><h3>Consumer-Facing ZK in Production</h3><p>Other than zkRollups, there are very few consumer-facing ZK applications in existence. This can be attributed to a litany of reasons, all equally valid: a lack of ZK builders, a lack of ZK development tools and methodologies, or simply a lack of use cases that are meaningful to consumers. zkP2P demonstrates a case where even if the ZK isn’t meaningful to consumers, it’s <em>essential</em> to enable web3 to come to where it is convenient for them. Without the public key infrastructure and signatures that underpins the internet and the privacy-preserving proofs to authenticate them, an on-ramp that allows existing payment systems to be agnostic to it.</p><h3>The Future</h3><p>zkP2P is under active development. It’s currently in public alpha, supports USD via Venmo, INR via HDFC, TRY via Garanti, and is adding support for more currencies and payment providers (support for EUR/USDC via Revolut was added on May 31st). The team is funded by the Fenbushi Research Grant and the Ethereum Foundation and led by Sachin (<a href="https://x.com/0xSachinK">0xSachinK</a> on X). Sachin was previously with Set Protocol, having built smart contracts with over 500M TVL, and has received grants from the Ethereum Foundation to build ZK applications. His team started zkP2P at a ZKHack, an ongoing series of hackathons which he now supports as a judge.</p><p>Written by <a href="https://x.com/turanzv">Turan Vural</a>, a researcher at Fenbushi Capital and a PhD student in cryptography at UCLA specializing in MPC and distributed computing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e317f09b7c1e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Availability on FileCoin]]></title>
            <link>https://fenbushicapital.medium.com/data-availability-on-filecoin-3555fc121765?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/3555fc121765</guid>
            <category><![CDATA[filecoin]]></category>
            <category><![CDATA[data-availability]]></category>
            <category><![CDATA[ethereum]]></category>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Thu, 04 Apr 2024 18:24:08 GMT</pubDate>
            <atom:updated>2024-04-28T23:20:23.120Z</atom:updated>
            <content:encoded><![CDATA[<h3>Hot DA on Cold Storage: Building cost-effective DA on FileCoin</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dun1CXZTkr6R1JJ1pHlMeQ.jpeg" /></figure><p><em>Thanks to </em><a href="https://twitter.com/ballsyalchemist"><em>Yuki Yuminaga</em></a><em>, </em><a href="https://twitter.com/jnthnvctr"><em>Jonathan Victor</em></a><em>, </em><a href="https://twitter.com/mfreh418"><em>Matthew Frehlich,</em></a><em> </em><a href="https://twitter.com/colludingnode"><em>Connor Ohara</em></a><em>, and </em><a href="https://twitter.com/51N3P"><em>Kobby Chen</em></a><em> for great discussions and feedback. Thanks to </em><a href="https://twitter.com/0xtaetaehoho"><em>Terry</em></a><em> and </em><a href="https://twitter.com/_weidai"><em>Wei Dai</em></a><em> for their work on </em><a href="https://docs.google.com/spreadsheets/d/104SDzgfPaaSxwyZREaqAwCXSVnqzGJzaZ8NMjYa7s5Y/edit#gid=0"><em>Rollups and DA Execution</em></a><em>.</em></p><p><a href="https://ethereum.org/en/developers/docs/data-availability/">Data availability</a> (DA) is a core technology in the scaling of Ethereum, allowing a node to efficiently verify that data is available to the network without having to host the data in question. This is essential for the efficient building of rollups and other forms of vertical scaling, allowing execution nodes to ensure that transaction data is available during the settlement period. This is also crucial for sharding and other forms of horizontal scaling, a planned future update to the Ethereum network, as nodes will need to prove that transaction data (or <a href="https://www.eip4844.com/">blobs</a>) stored in network shards are indeed available to the network.</p><p>Several DA solutions have been discussed and released recently (e.g., <a href="https://celestia.org/">Celestia</a>, <a href="https://docs.eigenlayer.xyz/eigenda/overview">EigenDA</a>, <a href="https://www.availproject.org/">Avail</a>), all with the intent of providing performant and secure infrastructure for applications to post DA.</p><p>The advantage of an external DA solution over an L1 such as Ethereum is that it provides an inexpensive and performant vehicle for on-chain data. DA solutions often consist of their own public chains built to enable cheap and permissionless storage. Even with modifications, the fact remains that hosting data natively from a blockchain is extremely inefficient.</p><p>Thus, we find that it is intuitive to explore a storage-optimized solution such as <a href="https://filecoin.io/">FileCoin</a> for the basis of a DA layer. FileCoin uses its blockchain to coordinate storage deals between clients and storage providers but allows data to be stored off-chain.</p><p>In this post, we investigate the viability of a DA solution built on top of a Distributed Storage Network (DSN). We consider FileCoin specifically, as it is the most adopted DSN to date. We outline the opportunities that such a solution would offer, and the challenges that need to be overcome to build it.</p><p>A DA layer provides the following to services relying on it:</p><ol><li><strong>Client Safety: </strong>No node can be convinced that unavailable data is available.</li><li><strong>Global Safety</strong>: The un/availability of data is agreed upon by all except at most a small minority of nodes.</li><li>Efficient data retrievability.</li></ol><p>All of this needs to be done efficiently to enable scaling. A DA layer provides higher performance at a lower cost across the three points above. For example, any node can request a full copy of the data to prove custody, but this is inefficient. By having a system that provides all three of these, we achieve a DA layer that provides the security required for L2s to coordinate with an L1, along with stronger lower bounds in the presence of a malicious majority.</p><h4>Custody of Data</h4><p>Data posted to a <a href="https://notes.ethereum.org/@vbuterin/proto_danksharding_faq#If-data-is-deleted-after-30-days-how-would-users-access-older-blobs">DA solution has a useful lifetime</a>: long enough to settle disputes or verify a state transition. Transaction data needs to be available only long enough to verify a correct state transition or to give validators enough opportunity to construct fraud proofs. As of writing, Ethereum calldata is the most common solution used by projects (rollups) requiring data availability.</p><h4>Efficient Verification of Data</h4><p>Data Availability Sampling (<a href="https://arxiv.org/abs/1809.09044">DAS</a>) is the standard method of answering the question of DA. It comes with additional security benefits, strengthening network actors’ ability to verify state information from their peers. However, it relies on nodes to perform sampling: DAS requests must be answered to ensure mined transactions won’t be rejected, but there is no positive or negative incentive for a node to request samples. From the perspective of nodes that request samples, there is no negative penalty for not performing DAS. As an example, Celestia provides the first and only light client implementation to perform DAS, delivering stronger security assumptions to users and reducing the cost of data verification.</p><h4>Efficient Access</h4><p>A DA needs to provide efficient access to data to the projects using it. A slow DA may become the bottleneck for the services relying on it, causing inefficiencies at best and system failures at worst.</p><h3>Decentralized Storage Network</h3><p>A Decentralized Storage Network (DSN, as formalized in the <a href="https://filecoin.io/filecoin.pdf">FileCoin Whitepaper</a>¹) is a permissionless network of storage providers that offer storage services for users of the network. Informally, it allows independent storage providers to coordinate storage deals with clients that need storage services and provides cheap and resilient data storage to clients seeking storage services at a low price. This is coordinated through a blockchain that records storage deals and enables the execution of smart contracts.</p><p>A DSN scheme is a tuple of three protocols: Put, Get, and Manage. This tuple comes with properties such as fault tolerance guarantees and participation incentives.</p><p><strong>Put(data) → key<br></strong>Clients execute Put to store data under a unique key. This is achieved by specifying the duration for which data will be stored on the network, the number of replicas of the data that are to be stored for redundancy, and a negotiated price with storage providers.</p><p><strong>Get(key) → data<br></strong>Clients execute Get to retrieve data that is being stored under a key.</p><p><strong>Manage()<br></strong>The Manage protocol is called by network participants to coordinate the storage space and services made available by providers and repair faults. In the case of <a href="https://filecoin.io/">FileCoin</a>, this is managed via a blockchain. This blockchain records data deals being made between clients and data providers and proofs of correctly stored data to ensure that data deals are being upheld. Correctly stored data is proved via the posting of proofs generated by data providers in response to challenges from the network. A <a href="https://spec.filecoin.io/#section-glossary.fault">storage fault</a> occurs when a storage provider fails to generate a Proof-of-Replication or Proof-of-Spacetime promptly when requested by the Manage protocol, which results in the slashing of the storage provider’s stake. Deals can self-heal in the case of a storage fault if more than one provider is hosting a copy of the data on the network by finding a new storage provider to honor the storage deal.</p><h3>DSN Opportunities</h3><p>The work done thus far in DA projects has been to transform a blockchain into a platform for hot storage. Since a DSN is storage-optimized, rather than transforming a blockchain into a storage platform, we can simply transform a storage platform into one that provides data availability. The collateral of storage providers in the form of native <a href="https://coinmarketcap.com/currencies/filecoin/">FIL</a> token can provide crypto-economic security that guarantees data is stored. Finally, the programmability of storage deals can provide flexibility around the terms of data availability.</p><p>The most compelling motivation to transform the capabilities of a DSN to solve DA is the cost reduction in the data storage under the DA solution. As we discuss below, the cost of storing data on FileCoin is significantly cheaper than storing data on Ethereum. Given current Ether/USD prices, it costs over 3 million USD to write 1 GB of calldata to Ethereum, only to be pruned after 21 days. This calldata expense can contribute to over half of the transaction cost of an Ethereum-based rollup. However, 1 GB of storage on FileCoin costs less than .0002 USD per month. Securing DA at this or any similar price would bring transaction costs down for users and contribute to the performance and scalability of Web3.</p><h4>Economic Security</h4><p>In FileCoin, collateral is required to make storage space available. This collateral is slashed when a provider fails to honor its deals or uphold network guarantees. A storage provider that fails to provide services faces losing both its posted collateral and any profit that would have been earned from providing storage.</p><h4>Incentive Alignment</h4><p>Many of FileCoin’s protocol incentives align with the goals of DA. <a href="https://docs.filecoin.io/storage-providers/filecoin-economics/slashing">FileCoin provides disincentives for malicious or lazy behavior</a>: storage providers must actively provide proofs of storage during consensus in the form of <a href="https://filecoin.io/proof-of-replication.pdf">Proof-of-Replicas</a> and <a href="https://spec.filecoin.io/algorithms/pos/post/">Proof-of-Spacetime</a>, continuously proving that the storage exists without honest majority assumptions. Failure of a storage provider to provide proof results in stake slashing, and removal from consensus, among other penalties. Current DA solutions lack incentive for nodes to perform DAS, relying on ad-hoc altruistic behavior for proof of DA.</p><h4>Programmability</h4><p>The ability to customize data deals also makes a DSN an attractive platform for DA. Data deals can have varying durations, allowing users of a DSN-based DA to pay for only the DA that they need. Fault tolerance can also be tuned by setting the number of copies that are to be stored throughout the network. Further customization is supported via smart contracts on FileCoin (called <a href="https://docs.filecoin.io/basics/the-blockchain/actors#">Actors</a>), which are executed on the <a href="https://fvm.filecoin.io/">FEVM</a>. This leads to FileCoin’s growing ecosystem of DApps, from compute-over-storage solutions such as <a href="https://www.bacalhau.org/">Bacalhau</a> to DeFi and liquid staking solutions such as <a href="https://www.glif.io/en">Glif</a>. <a href="https://www.retriev.org/#/">Retriev</a> makes use of FileCoin Actors to provide incentive-aligned retrieval with permissioned referees. FileCoin’s programmability can be used to tailor DA requirements needed for different solutions, so that platforms that rely on DA are not paying for more DA than they need.</p><h3>Challenges to a DSN-Based DA Architecture</h3><p>In our investigation, we have identified significant challenges that need to be overcome before a DA service can be built on a DSN. As we now talk about the feasibility of implementation, we will use FileCoin as our main focus of the discussion.</p><h4>Proof Latency</h4><p>The cryptographic proofs that ensure the integrity of deals and stored data on FileCoin take time to prove. When data is committed to the network, it is partitioned into 32 gigabyte <a href="https://spec.filecoin.io/systems/filecoin_mining/sector/">sectors</a> and “<a href="https://spec.filecoin.io/systems/filecoin_mining/sector/sealing/">sealed</a>.” The sealing of data is the foundation of both the Proof-of-Replication (<a href="https://spec.filecoin.io/algorithms/pos/porep/">PoRep</a>), which proves that a storage provider is storing one or more unique<em> </em>copies of the data, and Proof-of-Spacetime (<a href="https://spec.filecoin.io/algorithms/pos/post/">PoST</a>), which proves that a storage provider stored a unique copy continuously throughout the duration of the storage deal. Sealing has to be computationally expensive to ensure that storage providers aren’t sealing data on demand to undermine the required PoReP. When the protocol presents the periodic challenge to a storage provider to provide proof of unique and continuous storage, sealing has to safely take longer than the response window so that a storage provider can’t falsify proofs or replicas on the fly. For this reason, it can take providers approximately three hours to seal a sector of data.</p><h4>Storage Threshold</h4><p>Because of the computational expense of the sealing operation, the sector size of the data being sealed has to be economically worthwhile. The price of storage has to justify the cost of sealing to the storage provider, and likewise, the resulting cost of data being stored has to be low enough at scale (in this case, for an approximately 32GB chunk) for a client to want to store data on FileCoin. Although smaller sectors could be sealed, this would drive up the price of storage to compensate storage providers. To get around this, <a href="https://docs.filecoin.io/smart-contracts/advanced/aggregated-deal-making">data aggregators</a> collect smaller pieces of data from users to be committed to FileCoin as a chunk close to 32 GB. Data aggregators commit to user’s data via a Proof-of-Data-Segment-Inclusion (<a href="https://docs.filecoin.io/smart-contracts/advanced/aggregated-deal-making#proof-of-deal-sub-piece-inclusion-podsi">PoDSI</a>), which guarantees the inclusion of a user’s data in a sector, and a sub-piece CID (pCID), which the user will be able to use to retrieve the data from the network.</p><h4>Consensus Constraints</h4><p>FileCoin’s consensus mechanism, <a href="https://spec.filecoin.io/algorithms/expected_consensus/">Expected Consensus</a>, has a block time of 30 seconds and finality within hours, which may improve in the near future (see <a href="https://fips.filecoin.io/FIPS/fip-0086.html">FIP-0086</a> for fast finality on FileCoin). This is generally too slow to support the transaction throughput needed for a Layer 2 relying on DA for transaction data. FileCoin’s block time is lower-bounded by storage provider hardware; the lower the block time, the more difficult it is for storage providers to generate and provide proofs of storage, and the more storage providers will be falsely penalized for missing the proving window for the proper storage of data. To overcome this, <a href="https://www.ipc.space/">InterPlanetary Consensus (IPC) subnets</a> can be leveraged to take advantage of faster consensus times. IPC uses <a href="https://tendermint.com/">Tendermint</a>-like consensus and <a href="https://drand.love/">DRAND</a> for randomness: in the case that DRAND is the bottleneck, we would be able to achieve a 3-second block-time with an IPC subnet. In the case of a Tendermint bottleneck, PoCs such as <a href="https://arxiv.org/abs/2105.11827">Narwhal</a> have achieved blocktimes in the hundreds of milliseconds.</p><h4>Retrieval Speed</h4><p>The final barrier-to-build is retrieval. From the constraints above, we can deduce that FileCoin is suitable for cold or lukewarm storage. However, the DA data is hot and needs to support performant applications. Incentive-aligned retrieval is difficult in FileCoin; data needs to be unsealed before it is served to clients, which adds latency. Currently, rapid retrieval is done via SLAs or the storage of un-sealed data alongside sealed sectors, neither of which can be relied on in the architecture of a secure and permissionless application on FileCoin. Especially with <a href="https://www.retriev.org/#/">Retriev</a> proving that retrieval can be guaranteed via the FVM, incentive-aligned rapid retrieval on FileCoin remains an area to be further explored.</p><h3>Cost Analysis</h3><p>In this section, we consider the cost that comes from these design considerations. We show the cost of storing 32GB as Ethereum calldata, Celestia blobdata, EigenDA blobdata, and as a sector on FileCoin using near-current market prices.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nxPSDoJu2czejNDgQ6bu-w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KvXkpeyKMWopkAjbGs8tCA.png" /></figure><p>The analysis highlights the price of Ethereum calldata: 100 million USD for 32 GB of data. This price showcases the cost of security behind Ethereum’s consensus, and is subject to the volatility of Ether and gas prices. The Dencun upgrade, which introduced Proto-Danksharding (<a href="https://eips.ethereum.org/EIPS/eip-4844">EIP-4844</a>), introduced blob transactions with a target of 3 blobs per block of approximately 125 KB each, and variable gas blob pricing to maintain the target amount of blobs per block. This upgrade cut the cost of Ethereum DA by ⅕: 20 million USD for 32 GB of blob data.</p><p>Celestia and EigenDA provide significant improvements: 8,000 and 26,000 USD for 32 GB of data, respectively. Both are subject to the volatility of market prices and reflect to some extent the cost of consensus securing their data: Celestia with its native <a href="https://coinmarketcap.com/currencies/celestia/">TIA</a> token, and EigenDA with Ether.</p><p>In all of the above cases, the data stored is not permanent. Ethereum calldata is stored for 3 weeks, with blobs stored for 18 days. EigenDA stores blobs for a default of <a href="https://docs.eigenlayer.xyz/eigenda/system-performance-and-customization#encoded-system-throughput">14 days</a>. As of the current Celestia implementation, blob data is stored indefinitely by archival nodes but only sampled by light nodes for a maximum of <a href="https://docs.celestia.org/learn/retrievability#data-retrievability-and-pruning-in-celestia-node">30 days</a>.</p><p>The final two tables are direct comparisons between FileCoin and current DA solutions. Cost equivalence first lists the cost of a single byte of data on the given platform. The amount of FileCoin bytes that can be stored for the same amount of time for the same cost is then shown.</p><p>This shows that FileCoin is orders of magnitude cheaper than current DA solutions, costing fractions of a cent to store the same amount of data for the same amount of time. Unlike Ethereum nodes and that of other DA solutions, FileCoin’s nodes are optimized to provide storage services, and its proof system allows nodes to prove storage, rather than replicate storage across every node in the network. Without accounting for the economics of storage providers (such as the energy cost to seal data), it shows that the basic overhead of the storage process on FileCoin is negligible. This shows a market opportunity in the millions of USD <em>per gigabyte</em> compared to Ethereum for a system that can provide secure and performant DA services on FileCoin.</p><h4><strong>Throughput</strong></h4><p>Below, we consider the capacity of DA solutions and the demand that is generated by major layer 2 rollups.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7aQ4DoDBnir_KYOGJUw0iw.png" /></figure><p>Because FileCoin’s blockchain is organized in <a href="https://spec.filecoin.io/systems/filecoin_blockchain/struct/tipset/#section-systems.filecoin_blockchain.struct.tipset">tipsets</a> with multiple blocks at every block-height, the number of deals that can be done is not restricted by consensus or block size. The strict data constraint of FileCoin is that of its network-wide storage capacity, not what is allowed via consensus.</p><p>For daily DA demand, we pull data from <a href="https://docs.google.com/spreadsheets/d/104SDzgfPaaSxwyZREaqAwCXSVnqzGJzaZ8NMjYa7s5Y/edit#gid=0">Rollups DA and Execution</a> from Terry Chung and Wei Dai, which includes a daily average across 30 days and a singular sampled day. This allows us to consider average demand while not overlooking aberrations from the average (for example, Optimism’s demand on 8/15/2023 of approximately 261,000,000 bytes was over 4x its 30 day average of 64,000,000 bytes).</p><p>From this selection, we see that despite the opportunity of lower DA cost, we would need a dramatic increase in DA demand to make efficient use of the 32 GB sector size of FileCoin. Although sealing 32 GB sectors with less than 32 GB of data would be a waste of resources, we could do so while still reaping a cost advantage.</p><h3>Architecture</h3><p>In this section, we consider the technical architecture that can be achieved if we were to build this today. We will consider this architecture in the context of arbitrary L2 applications and an L1 chain that the L2 is serving. Since this solution is an external DA solution, like that of Celestia and EigenDA, we do not consider FileCoin as example L1.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KC7wj9R-hJZS1zynnjYVJA.jpeg" /></figure><h4>Components</h4><p>Even at a high-level, a DA on FileCoin will make use of many different features of the FileCoin ecosystem.</p><p><strong>Transactions</strong>: Downstream users make transactions on a platform that requires DA. This could be an L2.</p><p><strong>Platforms Using DA</strong>: These are the platforms that use DA as a service. This could be an L2 which posts transaction data to the FileCoin DA and commitments to an L1, such as Ethereum.</p><p><strong>Layer 1</strong>: This is any L1 that contains commitments pointing to data on the DA solution. This could be Ethereum, supporting an L2 that leverages the FileCoin DA solution.</p><p><strong>Aggregator</strong>: The frontend of FileCoin-based DA solution is an <a href="https://docs.filecoin.io/smart-contracts/advanced/aggregated-deal-making">aggregator</a>, a centralized component which receives transaction data from L2’s and other DA clients and aggregates them into 32 GB sectors suitable for sealing. Although a simple proof-of-concept would include a centralized aggregator, platforms using the DA solution could also run their own aggregator,for example as a sidecar to an L2 sequencer. The centralization of the aggregator can be seen as similar to that of an L2 sequencer or <a href="https://docs.eigenlayer.xyz/eigenda/overview#how-eigenda-works">EigenDA’s disperser</a>. Once the aggregator has compiled a payload near 32GB, it makes a storage deal with storage providers to store the data. Clients are given a guarantee that their data will be included in the sector in the form of a PoDSI (Proof of Data Segment Inclusion), and a pCID to identify their data once it is on the network. This pCID is what would be included in the state commitments on the L1 to reference supporting transaction data.</p><p><strong>Verifiers</strong>: Verifiers request the data from the storage providers to ensure the integrity of state commitments and build fraud proofs, which are committed to the L1 in the case of provable fraud.</p><p><strong>Storage Deal</strong>: Once the aggregator has compiled a payload near 32GB, the aggregator makes a storage deal with storage providers to store the data.</p><p><strong>Posting blobs (Put)</strong>: To initiate a put, a DA client will submit their blob containing transaction data to the aggregator. This can be done in an off-chain manner, or an on-chain manner via an on-chain aggregator oracle. To confirm receipt of the blob, the aggregator returns a PoDSI to the client to prove that their blob is included in the aggregated sector that will be committed to the subnet. A pCID (sub-piece Content IDentifier) is also returned. This is what the client and any other interested party will use to reference the blob once it is being served on FileCoin.</p><p>Data deals would appear on-chain within minutes of the deal being made. The largest barrier to latency is the sealing time, which can take 3 hours. This means that although the deal has been made, and the client can be confident that the data will appear in the network, the data cannot be guaranteed to be queryable until the sealing process is complete. The <a href="https://lotus.filecoin.io/">Lotus</a> client has a <a href="https://docs.filecoin.io/networks/mainnet/network-performance#data-retrieval">fast-retrieval</a> feature in which an unsealed copy of the data is stored alongside the sealed copy that may be able to be served as soon as the unsealed data is transferred to the data storage provider, as long as a retrieval deal does not depend on the proof of sealed data to appear on the network. However, this functionality is at the discretion of the data provider, and is not cryptographically guaranteed as part of the protocol. If a fast-retrieval guarantee is to be provided, there would need to be changes to consensus and dis/incentive mechanisms in place to enforce it.</p><p><strong>Retrieving blobs (Get)</strong>: Retrieval is similar to a put operation. A retrieval deal needs to be made, which will appear on-chain within minutes. Retrieval latency will depend on the terms of the deal and whether an unsealed copy of data is stored for fast retrieval. In the fast retrieval case, the latency will depend on network conditions. Without fast retrieval, data will need to be unsealed before being served to the client, which takes the same amount of time as sealing, on the order of 3 hours. Thus without optimizations we have a maximum round-trip of 6 hours, major improvement in data serving would need to be made before this becomes a viable system for DA or fraud proofs.</p><p><strong>Proof of DA</strong>: proof of DA can be considered in two steps; via the PoDSI that is given when the data is committed to the aggregator while the deal is being made and then the continued commitment of PoRep and PoST that storage providers provide via FileCoin’s consensus mechanism. As discussed above, the PoRep and PoST give scheduled and provable guarantees of data custody and persistence.</p><p>This solution will make heavy use of bridging, as any client that relies on DA (regardless of the construction of proofs) will need to be able to interact with FileCoin. In the case of the pCID included in the state transition that is posted to the L1, a verifier can make an initial check to make sure that a bogus pCID wasn’t committed. There are several ways that this could be done, for example, via an oracle that posts FileCoin data on the L1 or via verifiers that verifies the existence of a data deal or sector that corresponds to the pCID. Likewise, the verification of validity or fraud proofs that get posted to the L1 may need to make use of a bridge to be convinced of a proof. Current available bridges are <a href="https://axelar.network/">Axelar</a> and <a href="https://cbridge.celer.network/1/56/USDC">Celer</a>.</p><h4>Security Analysis</h4><p>FileCoin’s integrity is enforced through the slashing of collateral. Collateral can be slashed in <a href="https://spec.filecoin.io/systems/filecoin_markets/onchain_storage_market/faults/#section-systems.filecoin_mining.sector.sector-faults">two cases</a>: <a href="https://spec.filecoin.io/#section-glossary.fault">storage faults</a> or <a href="https://spec.filecoin.io/#section-glossary.consensus-fault-slashing">consensus faults</a>. A storage fault corresponds to a storage provider not being able to provide proof of stored data (either PoRep or PoST), which would correlate to a lack of data availability in our model. A consensus fault corresponds to malicious action in consensus, the protocol that manages the transaction ledger from which the FEVM is abstracted.</p><ul><li>A <a href="https://spec.filecoin.io/#section-systems.filecoin_mining.sector.sector-faults">Sector Fault</a> refers to the penalty incurred from the failure to post proof of continuous storage. Storage providers are allowed a one-day grace period during which a penalty is not incurred for faulty storage. After 42 days from a sector becoming faulty, the sector is terminated. Incurred fees are burnt.</li></ul><p><em>BR(t) = ProjectedRewardFraction(t) * SectorQualityAdjustedPower</em></p><ul><li>A <a href="https://spec.filecoin.io/#section-systems.filecoin_mining.sector.sector-faults">Sector Termination</a> occurs after a sector has been faulty for 42 days or a storage provider purposefully terminates a deal. Termination fees are equivalent to the maximum amount that a sector has earned up to termination, with an upper bound of 90 days’ worth of earning. Unpaid deal fees are returned to the client. Incurred fees are burnt.</li></ul><p><em>max(SP(t), BR(StartEpoch, 20d) + BR(StartEpoch, 1d) * terminationRewardFactor * min(SectorAgeInDays, 140))</em></p><ul><li><a href="https://spec.filecoin.io/systems/filecoin_markets/onchain_storage_market/storage_market_actor/#section-systems.filecoin_markets.onchain_storage_market.storage_market_actor">Storage Market Actor Slashing</a> occurs in the event of a terminated deal. This is the slashing of the collateral that the storage provider puts up behind the deal.</li></ul><p>The security provided by FileCoin is very different from that of other blockchains. Whereas blockchain data is typically secured via consensus, FileCoin’s consensus only secures the transaction ledger, not the data referred to by the transaction. The data that is stored on FileCoin has only enough security to incentive-align storage providers to provide storage. This means that the data stored on FileCoin is secured via fault penalties and business incentives such as reputation with clients. In other words, a data fault on a blockchain is equivalent to a breach of consensus, and breaks the safety of the chain or its notion of the validity of transactions. FileCoin is designed to be fault tolerant when it comes to data storage, and therefore only uses its consensus to secure its dealbook and deal-related activities. The cost of a storage miner not fulfilling its data deal has a maximum of 90 days worth of storage reward in penalties, and the loss of the collateral put up by the miner to secure the deal.</p><p>Therefore, the cost of a data withholding attack being launched from FileCoin providers simply the opportunity cost a retrieval deal. Data retrieval on FileCoin relies on the storage miner being incentivized by a fee paid for by the client. However, there is no negative impact to a miner for not responding to a data retrieval request. To mitigate the risk of a single storage miner ignoring or refusing data retrieval deals, data on FileCoin can be stored by multiple miners.</p><p>Since the economic security behind the data being stored on FileCoin is considerably less than that of blockchain based solutions, the prevention of data manipulation must also be considered. Data manipulation is protected via FileCoin’s proof system. Data is referred to via CIDs, through which data corruption is immediately detectable. A provider therefore cannot serve corrupt data, as it is easy to verify whether the fetched data matches the requested CID. Data providers cannot store corrupted data in the place of uncorrupted data. Upon the receipt of client data, providers must provide proof of a correctly sealed data sector to initiate the data deal (check this). Therefore, a storage deal cannot be started with corrupt data. During the lifetime of the storage deal, PoSTs are provided to prove custody (recall that this proves both custody of the sealed data sector and custody since the last PoST). Since the PoST is reliant on the sealed sector at the time of proof generation, a corrupt sector would result in a bogus PoST, resulting in a sector failure. Therefore, a storage provider can neither store nor serve corrupted data, cannot claim reward for services provided for uncorrupted data, and cannot avoid being penalized for tampering with a client’s data.</p><p>Security can be strengthened through increasing the collateral committed by the storage provider to the Storage Market Actor, which is currently decided by the storage provider and the client. If we assume that this was sufficiently high enough (for example, the same stake as an Ethereum validator) to incentivize a provider not to default, we can think of what is left to secure (even though this would be extremely capital-inefficient, as this stake would be needed to secure each transaction blob or sector with aggregated blobs). Now, a data provider could choose to make data unavailable for maximums of 41-day chunks before the storage deal is terminated by the Storage Market Actor. Assuming a shorter data deal, we could assume that the data can be made unavailable until the last day of the deal. In the absence of coordinated malicious actors, this can be mitigated via replication on multiple storage providers so that the data can continue being served.</p><p>We can consider the cost of an attacker overriding consensus to either accept a bogus proof or rewrite ledger history to remove a deal from the orderbook without penalizing the responsible storage provider. It is worth noting however that in the case of such a safety violation, an attacker would be able to manipulate FileCoin’s ledger however they want. In order for an attacker to commit such an attack, they would need at least a majority stake in the FileCoin chain. Stake is related to storage provided to the network; with a current 25 EiB (10¹⁶ bytes) of data securing the FileCoin chain, at least 12.5 EiB would be needed for a malicious actor to offer its own chain that would win the fork-choice rule. This is further mitigated by slashing related to consensus faults, for which the penalty is the loss of all pledged collateral and block rewards and all suspension from participation in consensus.</p><p><strong>Aside: Withholding attacks on other DA solutions<br></strong>Although the above shows that FileCoin is lacking in protecting data from withholding attacks, it is not alone.</p><ul><li>Ethereum: In general, the only way to guarantee that a request to the Ethereum network is answered is to run a full node. Full nodes have no requirements to fulfill data retrieval requests outside of consensus — and therefore. Constructs such as <a href="https://ethresear.ch/t/peerdas-a-simpler-das-approach-using-battle-tested-p2p-components/16541">PeerDAS</a> introduce a peer scoring system for a node’s responses to data retrieval in which a node with a low enough score (essentially a DA reputation) could be isolated from the network.</li><li>Celestia: Even though Celestia has much stronger security per-byte against withholding attacks in comparison to our FileCoin construction, the only way to take advantage of this security is to host your own full node. Requests to Celestia infrastructure that are not owned and operated in-house can be censored without penalty.</li><li>EigenDA: Similar to Celestia, any service can run an EigenDA Operator node to ensure retrieval of their own data. As such, any out protocol data retrieval request can be censored. Also note that EigenDA has a centralized and trusted dispenser in charge of data encoding, KZG commitment, and data dispersal, similar to our aggregator.</li></ul><h4>Retrieval Security</h4><p>Retrievability is necessary for DA. Ideally, market forces motivate economically rational miners to accept retrieval deals, and compete with other miners to keep prices for clients low. It is assumed that this is enough for data providers to provide retrieval services, however given the importance of DA, it is reasonable to require more security.</p><p>Retrieval is currently not guaranteed via the economic security stipulated above. This is because it is cryptographically difficult to prove that data <em>wasn’t</em> received by a client (in the case where a client needs to refute a storage miner’s claim of sending data) in a trust-minimized manner. A protocol-native retrieval guarantee would be required in order for retrieval to be secured through the FileCoin’s economic security. With minimal changes to the protocol, this means that retrieval would need to be associated with a sector fault or deal termination. <a href="https://www.retriev.org/">Retriev</a> is a proof-of-concept which was able to provide data retrieval guarantees by using trusted “referees” to mediate data retrieval disputes.</p><p><strong>Aside: Retrieval on other DA solutions<br></strong>As can be seen above, FileCoin lacks the protocol-native retrieval guarantees necessary to keep storage (or retrieval providers) from acting selfishly. In the case of Ethereum and Celestia, the only way to guarantee that data from the protocol can be read is to self-host a full node, or trust a SLA from an infrastructure provider. It is not trivial to guarantee retrieval as a FileCoin storage provider; the analogous setting in FileCoin would be to become a storage provider (requiring <a href="https://docs.filecoin.io/storage-providers/infrastructure/reference-architectures">significant infrastructure cost</a>) and successfully accept the same storage deal as a storage provider that was posted as a user, at which point one would be paying themselves to provide storage to themselves.</p><h4>Latency Analysis</h4><p>Latency on FileCoin is determined by several factors, such as network, topology, storage mining client configuration, and hardware capabilities. We provide a theoretical analysis which discusses these factors, and the performance that can be expected by our construct.</p><p>Due to the design of FileCoin’s proof system and lack of retrieval incentives, FileCoin is not optimized to provide high-performance round trip latency from the initial posting of data to the initial retrieval of data. High performance retrieval on FileCoin is an active area of research that is constantly changing as storage providers increase their capabilities and as FileCoin introduces new features. We define a “round trip” as the time from the submission of a data deal to the the earliest moment the data submitted to FileCoin can be downloaded.</p><p><strong>Block Time<br></strong>In FileCoin’s Expected Consensus, data deals can be included within the block-time of 30 seconds. 1 hour is the typical time for confirmation of sensitive on-chain data (such as coin transfers).</p><p><strong>Data Processing<br></strong>Data processing time varies widely between storage providers and configurations. The sealing process is designed to take 3 hours with standard storage mining hardware. Miners often outperform this 3 hour threshold via special client configurations, parallelization, and investing in more capable hardware. This variation also affects the duration of sector un-sealing, which can be circumvented altogether by quick retrieval options in FileCoin client implementations such as Lotus. The quick retrieval setting stores an unsealed copy of data alongside sealed data, significantly speeding up retrieval time. Based on this, we can assume a worst-case delay of three hours from the acceptance of a data deal to when the data is available on-chain.</p><h3>Conclusion and Future Directions</h3><p>This article explores building a DA by leveraging an existing DSN, FileCoin. We consider the requirements of a DA with respect to its role as a critical element of scaling infrastructure in Ethereum. We consider building on top of FileCoin for the viability of DA on a DSN, and use it to consider the opportunities that a solution on FileCoin would provide to the Ethereum ecosystem, or any that would benefit from a cost-effective DA layer.</p><p>FileCoin proves that a DSN can dramatically improve the efficiency of data storage in a distributed, blockchain-based system, with a proven saving of 100 million USD per 32 GB written at current market prices. Even though the demand for DA is not yet high enough to fill 32 GB sectors, the cost advantage of a DA still holds if empty sectors are sealed. Although current latency of storage and retrieval on FileCoin is not appropriate for the hot storage needs, storage miner-specific implementations can provide reasonable performance with data being available in under 3 hours.</p><p>The increased trust in FileCoin storage providers can be tuned via variable collateral, such as in EigenDA. FileCoin extends this tunabel security to allow for a number of replicas to be stored across the network, adding tunable byzantine tolerance. Guaranteed and performant data retrieval would need to be solved in order to robustly deter data withholding attacks, however like any other solution, the only way to truly guarantee retrievability is to self-host a node or trust infrastructure providers.</p><p>We see opportunities for DA in the further development of PoDSI, which could be used (alongside FileCoin’s current proofs) in place of DAS to guarantee data inclusion in a larger sealed sector. Depending on how this looks, this may make slow turnaround of data tolerable, as fraud proofs could be posted in a window of 1 day to 1 week, while DA could be guaranteed on demand. PoDSIs are still new and under heavy development, and so we make no implication yet on what an efficient PoDSI could look like, or the machinery needed to build a system around it. As there are solutions for compute on top of FileCoin data, the idea of a solution that computes a PoDSI on sealed or unsealed data may not be out of the realm of near-future possibilities.</p><p>As both the field of DA and FileCoin grows, new combinations of solutions and enabling technologies may enable new proof of concepts. As <a href="https://filecoin.io/blog/posts/filecoin-news-83/">Solana’s integration with the FileCoin network</a> shows, DSNs hold potential as a scaling technology. The cost of data storage on FileCoin provides an open opportunity with a large window of optimization. Although the challenges discussed in this article are presented in the context of enabling DA, their eventual solution will open a plethora of new tools and systems to be built beyond DA.</p><p><em>¹ Although this isn’t the construction of FileCoin, it is useful for those who are unfamiliar with programmable decentralized storage.</em></p><p><em>Graph data from </em><a href="https://spec.filecoin.io/systems/filecoin_blockchain/struct/tipset/"><em>FileCoin spec</em></a><em>, </em><a href="https://eips.ethereum.org/EIPS/eip-4844#gas-accounting"><em>EIP-4844</em></a><em>, </em><a href="https://blobs-goerli.eigenda.xyz/"><em>EigenDA</em></a><em>, </em><a href="https://github.com/celestiaorg/celestia-app/blob/main/pkg/appconsts/initial_consts.go#L18"><em>Celestia implementation</em></a><em>, </em><a href="https://celenium.io/gas"><em>Celenium</em></a><em>, </em><a href="https://dashboard.starboard.ventures/dashboard"><em>Starboard</em></a><em>, </em><a href="https://file.app/"><em>file.app</em></a><em>, </em><a href="https://docs.google.com/spreadsheets/d/104SDzgfPaaSxwyZREaqAwCXSVnqzGJzaZ8NMjYa7s5Y/edit#gid=0"><em>Rollups DA and Execution</em></a><em>, and current approximate market prices.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3555fc121765" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ending LP’s Losing Game: Exploring the Loss-Versus-Rebalancing (LVR) Problem and its Solutions]]></title>
            <link>https://fenbushicapital.medium.com/ending-lps-losing-game-exploring-the-loss-versus-rebalancing-lvr-problem-and-its-solutions-38099d700f9b?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/38099d700f9b</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Wed, 21 Feb 2024 23:55:22 GMT</pubDate>
            <atom:updated>2024-02-21T23:55:51.012Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AYGb7XaEBPq5lqW2p-tswQ.png" /></figure><ul><li>Thanks to <a href="https://twitter.com/ballsyalchemist">Yuki Yuminaga</a>, <a href="https://twitter.com/0x94305">Alex Nezlobin</a>, <a href="https://twitter.com/0xShitTrader">Eugene Chen</a>, <a href="https://twitter.com/AnthonyLeeZhang">Anthony Lee Zhang</a>, <a href="https://twitter.com/atiselsts_eth">Atis</a>, <a href="https://twitter.com/0xvanbeethoven">Ludwig</a>@<a href="https://twitter.com/SorellaLabs">Sorella Lab</a>, and <a href="https://twitter.com/MaxResnick1">Max Resnick</a> for great discussions and feedback.*</li><li>Original article link: <a href="https://mirror.xyz/0xbdA5bCe76bF62d97D9C9dF0425CC10079Df1aD75/bWOCccjVC7eoYKOzgmjXFdhWDc8rrUL6Yei-eugF52s">https://mirror.xyz/0xbdA5bCe76bF62d97D9C9dF0425CC10079Df1aD75/bWOCccjVC7eoYKOzgmjXFdhWDc8rrUL6Yei-eugF52s</a></li></ul><h3>Key Takeaways:</h3><ul><li>LVR is important because 1) it disincentivizes AMM LPs from providing liquidity to the pool, and 2) it leads to the centralization of block building because searcher-builders who are performing CEX-DEX arbitrages can generate far more valuable blocks than neutral builders due to their exclusive orderflow.</li><li>There are broadly two design philosophies to tackle the LVR problem: LVR minimization/reduction and LVR redistribution. LVR minimization/reduction can either be achieved via a low-latency oracle to inform CFMM LP of the current market price or via an LVR rebate parameter that allows arbitrageurs to execute a portion of the CEX-DEX arbitrage trade. LVR redistribution includes the dynamic fee model and the auction model: the dynamic fee model imposes price discrimination on informed and uninformed order flows based on pattern recognition. The auction model allows CEX-DEX arbitrageurs to pay LPs for the right to be the first trade per block at the application level.</li><li>The primary concern with the LVR reduction approach is the reduced orderflow directed to the LVR minimization/reduction pool. This results in LPs incurring greater losses compared to traditional CFMM pools. Instead of merely minimizing LVR, the future seems to be in redistributing LVR back to the LPs, effectively converting LVR into “Gain-versus-rebalancing(GVR)”.</li><li>In the realm of LVR redistribution, both the dynamic fee and auction approaches present distinct trade-offs: the dynamic fee model may cause uninformed flows to suffer from higher swap fees during high market volatility, while the auction model presents greater execution challenges, including building a competitive bidding market in a timely manner to allow uninformed traders to trade with the pool.</li></ul><h3>Introduction</h3><p>There have been continuous discussions around the profitability of liquidity provider (LP) at Constant Function Market Maker (CFMM) across crypto communities and academic researchers. Late last year, <a href="https://twitter.com/thiccythot_/status/1589022227437039616">this thread</a> estimated that Uniswap V3 LPs on ETH/USDC pools lost approximately $100 million dollars after fees on top of their original LP holdings by using markout on 5m/1h/24h/7d horizons (it is worthwhile to notice that the data presented by <a href="https://twitter.com/thiccythot_">@thiccythot_</a> could arguably be improved by using pool prices as markout prices as discussed <a href="https://crocswap.medium.com/usage-of-markout-to-calculate-lp-profitability-in-uniswap-v3-e32773b1a88e">in this article</a>, which tends to increase the estimates of LP profitability. Nonetheless, it is not a myth that most passive LPs in the CFMM V3 pool are losing money currently). The thread went viral in the DeFi community with many voices <a href="https://twitter.com/0xShitTrader/status/1579510561267929089">discouraging retail users from passively providing AMM liquidity</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ARcxX0l2m_i6SBcbjucgJA.png" /><figcaption>Uniswap V3 LPs markout on ETH/USDC pools: <a href="https://dune.com/thiccythot/uniswap-markouts">https://dune.com/thiccythot/uniswap-markouts</a></figcaption></figure><p>Meanwhile, many researchers took the initiatives to unveil the fundamental reasons behind passive LPs’ money-losing game by introducing the concepts of Loss-versus-rebalancing (LVR) and <a href="https://uniswap.org/flair.pdf">fee liquidity-adjusted instantaneous returns (FLAIR)</a>. LVR looks at the problem from the liquidity pool’s perspective, quantifying LPs’ adverse selection cost due to the sophisticated counterparty (informed trader). FLAIR, on the other hand, looks at the problem from an individual LP’s perspective to explain how the competition among LPs affects individual LP performance. These two concepts provide us with a more holistic understanding of the factors hindering passive LPs from making sustainable profits.</p><p>In this article, I will pinpoint my focus on LVR. <strong>Specifically, I want to explain how LVR leads to an unsustainable DeFi ecosystem due to misaligned incentives and how we can potentially realign those incentives to build a healthier DeFi architecture.</strong></p><h3>Understanding LVR</h3><p><strong>LVR refers to CFMM LPs’ adverse selection cost due to the information asymmetry between the maker (LP) and taker (informed swapper).</strong> This concept was first brought forth in <a href="https://arxiv.org/abs/2208.06046">this paper by Milionis, Moallemi, Roughgarden, and Zhang</a>. Adverse selection is a phenomenon that arises when one party has information that the counterparty does not have. In the context of CFMM, adverse selection occurs because high-frequency traders (HFTs) have access to the most up-to-date market prices on centralized exchanges (the most up-to-date market prices, for short-tail tokens at least, are at centralized exchange currently because centralized exchange has the largest amount of volumes) that the passive LPs do not have. Consequently, HFTs are able to exploit passive LPs’ information failure by making top-of-the-block arbitrages between CFMM stale prices and CEX prices. From the passive LPs’ perspective, they are consistently providing liquidity (quoting) to HFTs at worse-than-market prices, bearing losses when the price slippage is higher than the swap fees they charge to the HFTs. As HFTs perform CEX-DEX arbitrages, asset prices at CFMM will eventually converge to the market prices in a perfectly competitive market disregarding fees. <strong>Therefore, LVR can be measured as the gap between CFMM LPs executing risk assets at CFMM stale prices versus at market prices (equal to the top-of-the-block arbitrage profits extracted by HFTs), which essentially is the cost of price discovery for CFMM.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Oim-mqM-DEzEsc0v7Lspyg.png" /><figcaption>LVR Formula</figcaption></figure><p>The following example from the <a href="https://arxiv.org/abs/2208.06046">first LVR paper</a> mentioned previously can help us better contextualize LVR: suppose the x and y axes of the graph denote the quantity of X (risky asset) and quantity of Y (USD). Currently, CFMM LPs (the black invariant curve) are holding reserves at point A, corresponding to an instantaneous price Pt represented by the red line. Similarly, there is a rebalancing strategy that holds the same amount of reserves as CFMM LPs but trades at the centralized exchange along the brown line crossing point A. Assume that the CEX price for X is now at Pt + dPt, the arbitrageurs are thus incentivized to trade with CFMM, buying dxt amount of X from CFMM and selling dxt amount of X in CEX, moving CFMM from A to B along the invariant curve. From CLMM LPs’ perspective, they sold dxt amount of X at Pt+ dPt/2 (the slope of the scant purple line connecting A and B). If the rebalancing strategy makes the same trade as CFMM LPs but at CEX prices, they will sell dxt amount of X at Pt + dPt without slippage (assuming there is deep enough liquidity on CEX), eventually ending up at point B*. <strong>Therefore, the LVR can be figuratively captured by the height between B and B*; quantitatively, it would equal to dxt * ((Pt + dPt) — Pt+ dPt/2), which is also the (maximum) profits captured by arbitrageurs (GVR).</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p8kdia7ao-iL0xlRqrhSnQ.png" /><figcaption>LVR Explanation Graph: <a href="https://arxiv.org/abs/2208.06046">https://arxiv.org/abs/2208.06046</a></figcaption></figure><h3>Why is this a problem?</h3><p>In traditional finance, adverse selection is a known risk to market makers. Market makers facilitate trading by always being ready to buy (bid) or sell (ask) securities at publicly quoted prices, thereby providing liquidity to the market. In return, market makers profit from the bid-ask spread but bear the risk of not updating their quoted prices faster than the counterparty either due to information failure or execution speed. <strong>Because of this risk, the current market-making landscape has become centralized with a few high-frequency market making firms</strong> that are fast enough and informed enough to not get picked off by other traders.</p><p>Therefore, If the LVR issue continues, retail LPs will realize that their strategies become unprofitable and exit. This will cause CFMM to mirror the centralized market-making seen in TradFi, dominated by a few sophisticated LPs using just-in-time (JIT) liquidity. The absence of abundant, affordable liquidity from retail LPs will reduce overall liquidity, impairing trade execution and causing DeFi to lose significant volume to CEX. In essence, for DeFi to remain competitive, it requires efficient trade execution supported by profitable retail LPs.</p><p><strong>Additionally, LVR also poses a threat to the infrastructure level as it can lead to the centralization of block builders.</strong> For HFTs to successfully capture CEX-DEX arbitrages, they need to ensure that their arbitrage bundle is at the top of the block. Therefore, HFTs become builders themselves (also known as integrated searcher-builder), organizing their CEX-DEX arbitrage to be included at the top of the block. Currently, CEX-DEX arbitrages account for <a href="https://frontier.tech/a-tale-of-two-arbitrages">60% of the total MEV revenue</a>, which means that these integrated searcher-builders are able to construct more valuable blocks than neutral builders. Integrated builders can thus dominate the PBS auction to get their block included by the block proposer (this claim can be empirically supported by <a href="https://arxiv.org/abs/2305.19150">this paper</a>). In another word, integrated builders are able to dominate the PBS auctions because of the exclusive orderflow from their CEX-DEX arbitrages. As such, CEX-aware integrated builders exploit the information failure of CFMM LPs to generate disproportionately valuable blocks, which leads to the <strong>centralization of block builders</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lDBm2CZtup9v3wpvCp1vWw.png" /><figcaption>LVR Value Chain</figcaption></figure><h3>Potential Solutions</h3><p>The nature of the LVR problem lies in the <strong>value leaked from the application level to the infrastructure level</strong>. Since integrated builders need to pay bribe fees to the proposer to win the PBS auction, CFMM LP’s adverse selection cost (value leaked from the application level) goes to HFTs and proposers (it is estimated that <a href="https://frontier.tech/a-tale-of-two-arbitrages">37–77% of CEX-DEX arbitrage profits go to the validators currently</a>). This creates a misalignment of interest within the application level, namely HFTs and proposers are gaining profits at the expense of LPs, leading to CFMM LPs being unincentivized to participate.</p><h3>Solution Overview</h3><p>Thus, the LVR solution design generally takes in two forms:</p><ul><li>LVR reduction/minimization: this form has two lines of approaches:</li></ul><ol><li>Eliminate the value leaked (LP’s adverse selection cost) by cutting the need for informed/toxic flow to help with CFMM’s price discovery. This line of approach generally employs a low-latency oracle to inform CFMM LP of the current market price, solving the information failure at its root.</li><li>Minimize the LP’s adverse selection cost by only allowing arbitrageurs to execute a portion of the CEX-DEX arbitrage trade by introducing an LVR rebate parameter. The designated AMM pool price still aligns with the external price by removing corresponding tokens from the pool to a vault. The attained tokens in the vault will be gradually re-added back to the pool in subsequent blocks.</li></ol><ul><li>LVR redistribution: let LPs capture part of the LVR to realign the interests of LPs. This line of approach generally takes in two forms:</li></ul><ol><li>A dynamic fee model that charges more fees when there is more informed flow (similar to how in traditional finance market makers often widen the bid-ask spread during periods of high volatility).</li><li>An auction that requires HFTs to pay a fee to be the first trade per block, redirecting the value leaked to the proposer back to the LPs.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Sy1fMosdlUNqxV1XaeCTQA.png" /><figcaption>LVR Solution Design Landscape</figcaption></figure><h3>LVR Minimization: Diamond Protocol</h3><p>The Diamond Protocol was initially presented in <a href="https://arxiv.org/abs/2210.10601">this paper</a>, followed by a proof-of-concept demonstration in an <a href="https://ethresear.ch/t/lvr-minimization-in-uniswap-v4/15900">Ethresearch post</a> that differs in implementation. Recently, Arrakis Finance <a href="https://mirror.xyz/0x916563f8476b988855af0b8b8A3D56072E1917FA/ZhTQJ6qiurBTHBNuF08GudWBQ8x9p3P-12vo-C9czKE">announced plans</a> to adopt this concept using hooks. Given the variations in implementation across these sources, my discussion will center on the Arrakis Diamond Hook, accessible on this <a href="https://github.com/ArrakisFinance/minimize-lvr-hook-poc/blob/main/documentation/code-walkthrough.md">GitHub page</a>.</p><p>In the Diamond Protocol, builders must first commit to the end-of-the-block price at the start of the block, which should equate to the LVR maximizing price (i.e., the external market price), before any swaps can occur. Then, the Diamond protocol will update the pool price to the committed price via the openPool() function in the Hook contract. Specifically, this function will calculate the amount of trade size X required to move the starting pool price to the committed price. However, the actual trade size the builder can execute is discounted based on the LVR-rebate parameter, β, which ranges between 0 and 1. For instance, if β is 0.75, the builder can only execute a trade size of (1-0.75)X = 0.25X. To move the price in the pool to the committed price as if the entire trade size X is executed, the diamond protocol will remove some tokens from the pool to a vault contract. The tokens in the vault will be gradually re-added back to the pool per block at a percentage between 1-5%. It decides to slowly rebate the vault token back to the pool because re-adding tokens from the pool to the vault creates an expected arbitrage opportunity in the next block. By re-adding fewer tokens, the losses will be reduced, which increases the profitability of the pool.</p><p>After the openPool() function is completed, builders are required to post collateral to the hedger contract before any swap can occur. The deposited collateral must be sufficient to adjust the pool&#39;s price back to the committed price at any given point. In other words, the swap size should not exceed the tokens available in the hedger contract. This is important because If, by the end of the block, the pool&#39;s price doesn&#39;t return to the committed price, there are enough tokens in the hedger contract to bring the pool back to balance. The tokens in the hedger are used to adjust the pool&#39;s price in the next block when the pool is accessed.</p><p>The crux of how diamond protocol redistributes LVR back to the LPs lies in the collection and gradual readdition of LVR rebates back to the pool. When the vault fills up with LVR rebates, <strong>these rebates represent retained tokens from LP positions that would otherwise have been adversely selected against by the top-of-the-block arbitrage.</strong> Every time the vault accumulates a certain amount of tokens, denoted as Xi for blocki, the percentages of Xi owed to the respective LPs are directly proportional to the percentage of liquidity that each LP provided over the pool move in blocki. Over time, these rebated tokens are gradually reintroduced into the pool.</p><p>However, there are also potential concerns and issues stemming from the design of the Diamond Protocol (some of these observations are based on comments from <a href="https://twitter.com/ballsyalchemist/status/1712206997179810091">this Twitter post</a>):</p><ul><li><strong>Bounded order size in Diamond Protocol leading to LP’s unprofitability</strong>: Orders executed in the diamond pool must be smaller than the available tokens in the hedger contract to ensure rebalancing to the committed price. This limits the order size, potentially reducing order flow and fee earnings for LPs compared to a vanilla CFMM pool.</li><li><strong>Volatility throughout the block can go adversely to builder’s committed price at the start of the block</strong>: because the builder need to commit to the end-of-the-block price at the start of the block, any external market price fluctuations during the block can cause the pool price to diverge from the committed price. Given the volatility of crypto market, the likelihood of this happen will be quite high. In such cases, builders will be forced to take the other side of the trade, as they have an obligation to align the pool price with their committed price at the end of the block. This could result in losses for the builders, potentially disincentivizing them from participating in the diamond pool in the first place.</li><li><strong>Potential execution delay due to the exclusive reliance on a single block builder</strong>: If the designated builder fails to secure a win in the PBS auction, it could result in transaction delays within the Diamond pool. This becomes even more concerning considering the committed builder’s reduced competitiveness in the PBS auction, given that they can only capture a fraction of the implied LVR.</li><li><strong>Technical specification including which builder will be granted the exclusive right to perform the executing and ordering of the block.</strong> The document did not specify which builder would be granted the exclusive right to perform the executing and ordering of the block. Will this depend on how much collateral they are willing to put? (this is also another question that remains to be answered which is how much collateral the builder needs to put).</li></ul><h3>LVR Reduction: Oracle-Informed AMM</h3><p>Oracle-informed AMM can employ a low-latency oracle to inform CFMM LPs of the current market price before the CEX-DEX arbitrage. Specifically, it can work by including an active liquidity manager module that adjusts its liquidity concentration dynamically to the updated oracle price (like an automated JIT) before the CEX-DEX arbitrages. This is also called a <a href="https://medium.com/@alexnezlobin/solving-order-flow-toxicity-d388126cf69a">dealer-like</a> model of liquidity provisioning.</p><p>For oracle-informed AMM to prevent toxic orderflow, the oracle needs to be fast enough and accurate enough to prevent CEX-DEX arbitrageurs. Specifically, the speed of the oracle price and the subsequent adjustment of liquidity provision price should be before the CEX-DEX arbitrage. Additionally, the accuracy of the oracle price needs to be smaller than the pool’s swap fee. This is because an arbitrage is unprofitable whenever fees made from the trade are smaller than the price different from the market price. In this model, instead of acting as passive market makers who are waiting to get their stale price picked up, CFMM LPs become active market makers who have the same amount of information as the HFTs to actively provide quote prices compatible with the market prices.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4ywCB3x4vDdG1pBxMaPqzw.png" /><figcaption>Oracle-Informed AMM</figcaption></figure><p>However, this model, if implemented without modification, can lead to <strong>unbounded losses for LPs in the event of inaccurate oracle-fed prices due to oracle manipulation or oracle liveness issues.</strong> Since oracle-informed AMM loses its internal price discovery mechanism, it becomes a secondary price discovery platform that is vulnerable to oracle manipulation/shut-down. Oracle manipulation involves intentional interference with the data source, skewing the accuracy of the prices relayed. Additionally, if the oracle experiences downtime or liveness issues, it can’t provide real-time price updates. Therefore, when the oracle is down or provides an inaccurate price due to oracle manipulation, LPs will face an unbounded loss because they will aggregate their liquidity based on these flawed prices.</p><p><strong>Given this concern, oracle-informed AMM should implement relevant modifications to cap the potential unbounded loss.</strong> One approach to achieve this is by limiting the trade size executable within the pool, similar to <a href="https://twitter.com/xdev_10/status/1597494441614643202">how GMX caps the long/short open interest</a>. However, this method introduces a ceiling to the growth of the protocol. An alternative modification could involve using oracle-fed information to guide the pool’s fee instead of the price. This brings us to the next topic of discussion: the dynamic fee model.</p><h3>LVR Redistribution: Dynamic Fee</h3><p>The underlying philosophy of the dynamic fee model is to exert <strong>price discrimination</strong> upon informed and uninformed orderflow based on pattern recognition. It needs to <strong>1) utilize all sorts of signals to derive a pattern that can differentiate informed and uninformed orderflow and then 2) charge a higher fee to toxic orderflow and a lower fee to retail orderflow.</strong> In theory, the dynamic fee model redistributes part of the LVR back to the LPs because LPs can charge a higher fee for CEX-DEX arbitrages. In other words, CFMM LPs under the dynamic fee model can widen their bid-ask spread to informed traders and tighten their bid-ask spread to uninformed traders.</p><p>One example of the dynamic fee model is proposed by Alex Nezlobin <a href="https://twitter.com/0x94305/status/1674857993740111872">in this thread</a>. <strong>It recognizes an autocorrelated trading pattern of toxic flow that tends to keep picking on the ask side of the AMM if the market price pushed the AMM’s ask price in the previous block.</strong> To illustrate this, assume an AMM pool of ETH/USDC and ETH market price increases that jump outside of the AMM’s bid-ask spread. This will attract CEX-DEX arbitrages to buy ETH at AMM and sell ETH at CEX, making AMM’s best ask price equals to the CEX price. From the arbitragers’ perspective, they are more likely to continue buying ETH (pick on the ask side) than selling ETH (pick on the bid side) in the AMM. This is because it is more likely to profit from buying ETH than selling ETH considering the CEX price only needs to increase by a marginal point for buying ETH to be profitable but the CEX price needs to decrease by 2f for selling ETH to profit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QtmJnxXcAIdqnzG9JKSZ8A.png" /></figure><p>Compared with the autocorrelated trading pattern of informed flow, the trading direction of uninformed flow should be fairly random. Therefore, if the AMM price increases by Δ in block t, the model will raise the fee for buys and lower the fee for sells by δ=cΔ for c&gt;0 (to make neither fee negative) at block t+1, making informed flow pay more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9L6galW0V6vy95jIChllaw.png" /><figcaption>Graphic Representation of Alex Nezlobin’s Dynamic Fee Model: <a href="https://twitter.com/0x94305/status/1674857993740111872">https://twitter.com/0x94305/status/1674857993740111872</a></figcaption></figure><p>Empirically, Alex ran a Python simulation that changed fees by 0.75 if the AMM price went up in the previous block in the ETH/USD pool ($50,000 liquidity per basis point, 5 bps fee, 5% daily volume). <strong>The result shows that the losses of LPs in the dynamic fee model are around 10% lower</strong> than with the fixed fee model depending on the gas fee per swap.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F_5LEb6d_aoy-WT3LqKO4w.png" /><figcaption>Alex Nezlobin’s Dynamic Fee Model Simulation Result: <a href="https://twitter.com/0x94305/status/1674858006205571072">https://twitter.com/0x94305/status/1674858006205571072</a></figcaption></figure><p>Alex’s model demonstrates a neat way to exert price discrimination without using an oracle. However, there is still some downside including the arbitrageur who caused the first strong price shift will not be price discriminated against and the uninformed flow also faces this higher fee if they want to pick on the ask side. To further improve upon Alex’s model, there are other signals that AMMs can use to differentiate informed and uninformed flow to a greater extent (some of the following signals were discussed in the <a href="https://twitter.com/specialmech/status/1691139040119312384">LVR Reduction Twitter Space</a>):</p><ul><li><strong>Oracle-fed External world information</strong>: this refers to the information that CFMM passive LPs are unaware of, including market price, market volatility, etc. This information can be fed to the AMM using a low-latency oracle. For example, since LVR is greater when volatility is high (CEX-DEX price slippage is higher under high volatility), we can use an oracle that feeds us with updated market volatility signals to increase the swap fees when market volatility is high or vice versa. <a href="https://docs.ambient.finance/users/dynamic-fees">Ambient Finance</a> currently uses this model. This line of approach minimizes the effect of oracle compared to the oracle-informed AMM model because it still maintains internal price discovery that does not have a liveness issue. The problem with this approach is that high-frequency price movements on CEX can occur intra-block, rendering the oracle’s snapshot of the market outdated by the time it’s available for the AMM’s decision-making. Consequently, this misalignment creates a window of opportunity for arbitrageurs, as they can capitalize prior to an oracle update. Additionally, this could also lead to a fall in overall volumes because it neutralizes the increased fees to both informed and uninformed flow, discouraging them from participating in this pool at high volatility and choosing fixed-price pools instead, leading to a welfare loss to all.</li><li><strong>Source of orderflow:</strong> this refers to the channel in which the orderflow is routed from (e.g., CoWSwap, UniswapX, Metamask, etc.). In traditional finance, Robinhood monetizes by selling its orderflow to high-frequency market makers because most orderflow coming from Robinhood are retail flow. Similarly, in crypto, the source of the ordeflow can also be an important signal to differentiate informed and uninformed flow. One example is CoWSwap: CoWswap runs a batch auction that introduces a delay that makes them unattractive for toxic flow. Therefore, AMM pools could lower the prices for orderflow routed from CoWSwap solvers as there is a high possibility that the orderflow is uninformed. Currently, <a href="https://snapshot.org/#/balancer.eth/proposal/0xd991e9f3c6edd148bd37c600d7ada3d28db1758e3cfd703c02d290f502906f05">Balancer reduces the 50–75% swap fees for all orderflow coming from CoWSwap</a>. In the future, there could be a reputation system for all the relayers based on the toxicity of their orderflow and exert price discrimination based on that</li><li><strong>Existing orderflow</strong>: This refers to the trading pattern observed from the existing orderflow that can be used to differentiate informed and uninformed flow. One example is Alex Nezlobin’s dynamic fee model as discussed above: it discovers a trading pattern that informed flow has a higher possibility of picking on the ask side than on the bid side if the market price pushed the AMM’s ask price in the previous block, which allows the model to exert price discrimination accordingly. Additionally, one can employ a co-processor to perform computation based on the history of on-chain transactions to differentiate informed and uninformed flow. For example, the protocol can leverage <a href="https://docs.axiom.xyz/">Axiom</a> to analyze the historical record of on-chain transactions. Axiom provides trustless access to all on-chain data and offers verified compute capabilities over it. This means that developers can query historic data from Ethereum via Axiom, and Axiom will trustlessly fulfill ZK-verified results on-chain. By doing so, the dynamic fee model can further differentiate between informed and uninformed transaction flows without introducing additional trust assumptions.</li></ul><p><strong>Ideally, the most efficient dynamic fee model should combine all these signals to differentiate informed and uninformed flow to the highest probability.</strong> However, the dynamic fee model, even if it intakes all sorts of signals, is probabilistic in nature in differentiating informed and uninformed flow. Therefore, it still neutralizes its increased fees to both the informed and uninformed traders to a certain extent. The consequence of this is that uninformed flow might be discouraged from participating in these pools at times of high volatility (higher fees). This could be a concern under the adoption of intent-based architecture because solvers will be competing to generate the lowest possible price for the users, begging the question of <strong>how much orderflow will be routed to dynamic-fee AMM pools at times of higher fees.</strong> Additionally, a dynamic fee model naturally <strong>introduces additional friction</strong> to informed traders because the cost of arbitrage is higher at times of higher volatility (swap fee is higher), which can make the price discovery of the AMM pool less efficient.</p><h3>LVR Redistribution: Auction</h3><p>Different from the dynamic fee model that extracts value from the top-of-the-block arbitragers, the <strong>auction aims to redistribute LVR back to LPs from the hands of proposers</strong>. The reason why the value was leaked to the proposer was that HFT integrated builders need to compete in the PBS auction for block inclusion for their top-of-the-block arbitrage, hence paying priority fees to the proposers. Therefore, the gist of the auction model is to shift the power of being at the top-of-the-block to the AMM application level, thereby HFTs, instead of paying priority fees to the proposers, pay auction to the AMM for the right of the first trade.</p><p>One example, as proposed by <a href="https://ethresear.ch/t/mev-capturing-amm-mcamm/13336">Josojo in Ethresearch</a>, is to auction off the right to be the first trade per block at the application’s smart contract level. The winning bid for the auction is then distributed back to the LPs in the AMM pools. This design idea is similar to CEX provides faster execution speed for their VIP-tier customers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AeaciggN2HBx72qhNDnaLA.png" /><figcaption><a href="https://ethresear.ch/t/mev-capturing-amm-mcamm/13336">https://ethresear.ch/t/mev-capturing-amm-mcamm/13336</a></figcaption></figure><p>Specifically, it introduces a global contract that oversees the first execution right for all McAMMs. This contract, optimized for gas efficiency, ideally functions as the standard router for all McAMM pools. The designated party with the first execution right, termed the “leadsearcher,” is identified by this global contract. The contract maintains a record, per block, of whether the leadsearcher has traded on any of the deployed AMM contracts. Other participants are permitted to trade with the McAMM only after the leadsearcher’s trade; otherwise, trades are reverted.</p><p>To acquire the leadsearcher status, one must win the auction for the first execution right, structured similarly to the <a href="https://docs.edennetwork.io/slot-tenants/getting-started#becoming-a-slot-tenant">Eden network’s auction for the block’s initial slot</a>. Leadsearchers are automatically deselected by the router contract if they miss more than three blocks, preventing prolonged trade blockages. If leadsearchers opt not to trade in a block, they must still signal their interaction with the McAMMs, enabling others to trade against them. This signaling incurs a minimal gas cost for the leadsearcher. To ensure the leadsearcher captures all arbitrage opportunities, they are exempted from standard AMM fees.</p><p>Theoretically, block-builders’ interests are also aligned with this specification because they are <strong>motivated to propose blocks with maximal priority fee gas consumption, enhancing their chances of winning the MEV boost auction.</strong> In other words, block builders want the transactions to consume as much gas as possible so that their bidding price in the PBS auction is higher. Therefore, if the builders put other transactions in front of the leadsearcher’s transaction, those transactions would revert, leading to less gas and hence fewer fees. Therefore, block-builders are incentivized to ensure the correct trade sequence, prioritizing the leadsearcher’s transaction. Moreover, McAMM users are encouraged to exclusively broadcast their trades to block-builders who respect the McAMM’s enforced ordering, avoiding the costs of failed transactions. These combined incentives are expected to drive all block-builders to adhere to the necessary ordering for McAMMs.</p><p>On the other hand, there are some potential problems and further design specifications needed in the auction mechanism:</p><ul><li><strong>How to allocate the auction fee to the LPs fairly</strong>: One question with the McAMM design and auction design in general is how to allocate the auction fee among LPs across McAMM pools. Josojo discusses some possibility around accumulating the auction fee to a DAO that subsequently distribute the auction fees to the relevant parties.</li><li><strong>How to construct a competitive market for bidders</strong>: in order for LPs to capture the maximal amount of LVR value possible, we need to construct a competitive bidding market. This depends on when should the auction take place and finish. Currently, there are broadly two times where an auction can take place: <a href="https://youtu.be/lBntkdgoHr8">ex-ante where the leadsearcher for BlockN is determined in BlockN-1 prior to the CEX price movement ,and ex-post where the leadsearcher for BlockN is determined after the CEX price movement but before BlockN is committed.</a> Compared with the ex-ante auction, the <strong>ex-post auction has a relatively lower barrier of entry</strong> because searchers know how much CEX-DEX arbitrage value they can capture whereas searchers in ex-ante auction need to have option pricing ability because the volatility is not realized. Therefore, an ex-post auction should have a more competitive bidding market than the ex-ante auction. On the other hand, the ex-post auction has a shorter time to determine the leadsearcher compared with the ex-ante auction because the auction begins after the CEX price movement and needs to finish no later than the PBS auction. Thus, there seems to be a tradeoff in establishing a more time-contingent ex-post auction that could lead to a more competitive bidding market versus an ex-ante auction that could have a more buffer time but the higher barrier of entry for bidders. Additionally, how to construct a censorship-resistant auction mechanism (e.g., <a href="https://a16zcrypto.com/posts/article/how-auction-theory-informs-implementations/">first-price sealed auction with a commit reveal scheme</a>) within such a short period of time while attracting a significant amount of bidders could be challenging too.</li><li><strong>On-chain auction vs off-chain auction</strong>: whether we would want to run the auction on-chain or off-chain is another important metric that worth further consideration. In general, on-chain auction smart contracts offer the advantage of reduced overhead costs compared to establishing an off-chain trust-minimized entity for auctions. <strong>However, they introduce two challenges:1) gas costs and 2) the uncertainty tied to latency, which is influenced by network congestion</strong>. The imposition of gas costs on every bid <a href="https://a16zcrypto.com/posts/article/how-auction-theory-informs-implementations/">effectively adjusts the de facto bid value</a>. Since gas prices can fluctuate throughout the auction’s duration, <strong>a potential highest bidder might be sidelined due to prohibitive gas fees.</strong> The uncertainty tied to latency boils down to the time lag between a bid’s broadcast and its blockchain inclusion. If this bid is relayed to a public mempool rather than a private transaction pool, it can trigger a <a href="https://a16zcrypto.com/posts/article/nft-sales-market-clearing-gas-wars-auction-mechanism-design-for-builders/">gas bidding war</a> on top of the auction, leading to further complexity. These reasons are why current on-chain auctions <a href="https://eprint.iacr.org/2023/1454.pdf">struggle to scale with an increasing number of bidders, leading to network bottlenecks, higher transaction costs, and extended confirmation times</a>. The off-chain auction, on the other hand, suffers from the execution complexity around building a censorship-resistant and fair off-chain entity <strong>but benefits from its independence without worrying about the gas price cost and the associated uncertainty in latency due to potential network congestion</strong>. <a href="https://twitter.com/SorellaLabs">Sorella Lab</a> is designing an ex-post off-chain auction that strikes a good balance between these tradeoffs.</li></ul><h3>Conclusion</h3><p>Overall, the primary concern with the LVR reduction approach is the reduced orderflow directed to the LVR minimization/reduction pool. This results in LPs incurring greater losses compared to traditional CFMM pools. Instead of merely minimizing LVR, the future seems to be in redistributing LVR back to the LPs, effectively converting LVR into “Gain-versus-rebalancing(GVR)”.</p><p>In the LVR redistribution realm, both the auction and dynamic fee models present distinct trade-offs. The auction mechanism can more deterministically differentiate between informed and uninformed flows, ensuring that the uninformed flows aren’t subjected to higher fees as they would under the dynamic fee model. However, the auction approach comes with non-trivial execution challenges, as outlined above. Additionally, auction design must strike the right balance between building a competitive bidding market and ensuring the auction concludes quickly, allowing uninformed traders enough time to swap in the AMM pool before the next auction begins. I envision the future of auctions as off-chain, ex-post auctions settled on a sidechain or suave.</p><p>Another consideration is the amount of value that can be redistributed back to the LPs. In the auction model, the value that LPs can extract is contingent upon the competitiveness of the bidding market. Meanwhile, in the dynamic fee model, the key factor is its ability to differentiate between informed and uninformed flows, ensuring that it doesn’t levy higher fees on the latter, which otherwise could deter uninformed traders from interacting with the pool. With the adoption of zk-coprocessor, I believe it can aid the current oracle-based dynamic fee approach to differentiate informed and uninformed flows to a higher extent by querying on-chain data in a trust-minimized way.</p><p>All in all, I am very excited to see this new evolution of AMM DEX design to solve the adverse selection cost of LPs. This could further motivate passive LPs to join AMM, potentially offering execution prices that can rival the off-chain CEX platforms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZX6azrYbErmGLwP6ec7zaw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=38099d700f9b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analyzing Missed Slot Through The Lens of Relays — A potential path for relay incentivization]]></title>
            <link>https://fenbushicapital.medium.com/analyzing-missed-slot-through-the-lens-of-relays-a-potential-path-for-relay-incentivization-33b0b724a812?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/33b0b724a812</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Wed, 21 Feb 2024 22:46:39 GMT</pubDate>
            <atom:updated>2024-02-21T22:46:39.862Z</atom:updated>
            <content:encoded><![CDATA[<h3>Analyzing Missed Slot Through The Lens of Relays — A potential path for relay incentivization</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IzFvjqSW8ov4f4MeSdhWuQ.png" /></figure><p><em>Thanks to </em><a href="https://twitter.com/nero_eth"><em>Toni 2</em></a><em>, </em><a href="https://twitter.com/_julianma"><em>Julian 2</em></a><em>, </em><a href="https://twitter.com/mikeneuder"><em>Mike 2</em></a><em>, </em><a href="https://twitter.com/ralexstokes"><em>Alex 4</em></a><em>, </em><a href="https://twitter.com/_danielmarzec"><em>Dan 5</em></a><em>, </em><a href="https://twitter.com/MaxResnick1"><em>Max 4</em></a><em>, </em><a href="https://twitter.com/barnabemonnot"><em>Barnabé 2</em></a><em>, and </em><a href="https://twitter.com/blairlmarshall"><em>Blair 2</em></a><em> for the review and extensive discussions.</em></p><p><em>Special thanks to </em><a href="https://twitter.com/sui414"><em>Danning 3</em></a><em> for sharing the queried raw data for my analysis.</em></p><p><em>Additional thanks to </em><a href="https://twitter.com/alextes"><em>Alex 5</em></a><em> (Ultra sound) and </em><a href="https://twitter.com/0xMozart"><em>Karthik 9</em></a><em> for the discussion and comments.</em></p><p>Original article link: <a href="https://ethresear.ch/t/analyzing-missed-slot-through-the-lens-of-relays-a-potential-path-for-relay-incentivization/18257">https://ethresear.ch/t/analyzing-missed-slot-through-the-lens-of-relays-a-potential-path-for-relay-incentivization/18257</a></p><h3>Introduction</h3><p>Relays in MEV-Boost are doubly-trusted stakeholders within the existing PBS landscape and yet they are not properly incentivized. People have discussed and are concerned over many ways in which validators and builders game the current system, such as slot proposal delay for timing games and colocation for latency optimization, all to maximize their MEV revenue. However, relays have been looked over as a public good who reap no benefit in being honest and trusted. Relays could easily <a href="https://collective.flashbots.net/t/disclosure-mitigation-of-block-equivocation-strategy-with-early-getpayload-calls-for-proposers/1705">unbundle transactions and steal MEV 4</a> if their rewards surpass any reputational cost and loss of future revenue. But so far, most relays don’t dare to collude — they may prefer to just quit relaying instead. Hence why I have <a href="https://mirror.xyz/0xE21b1e6f471EDeF18264e9BBe51b7fA7643EE6B5/0Sh7BDW7qgH_nadfqF8bpmnjxnfoYzPFvRdmoIoi9mg">shed some 5</a>light on relays’ importance and explored the incentive dynamic across PBS with the hope of devising a sustainable plan for relay incentivization. With this piece, I would like to point out a new direction for thinking about this relay incentive problem from the lens of missed slots and propose approaches for profitable competitive and cooperative relaying.</p><h3>Problems with existing proposals</h3><p>Largely in two parts:</p><ol><li>Incentive misalignment across validators and/or builders:</li><li>Relay centralization — builders are incentivized to send bids to one relay only.</li></ol><p>Most of the proposals to date (check the <a href="https://mirror.xyz/0xE21b1e6f471EDeF18264e9BBe51b7fA7643EE6B5/0Sh7BDW7qgH_nadfqF8bpmnjxnfoYzPFvRdmoIoi9mg">here 5</a> for more details) are trying to carve up a piece of builders’ bids for relays, which means that builders and/or validators are losing potential future profit by accommodating relays. Implementing such proposals will mean that you have to ask at least one of the two parties to make some compromises and requires a large amount of off-chain coordination work, which is hard to execute if not incentive-aligned.</p><p>Alternatively, building a competitive relay market could enable a path to monetization by hyper-optimizing relay latency. However, the current proposal of competitive relays captures values by out-competing other relays in latency and aggregating more bids. For context, here is an explainer diagram for competitive relay.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cILCqICq3XZZ5oW-L7eNvw.png" /></figure><p><strong>Figure 1</strong>. Diagram illustrating the bid flow among competitive relays with rebates</p><p>In this diagram, winning builder 4 sends the bid to relay 3 who then rebates a <strong>maximum of 0.029 ETH</strong> back to the builder 4. The bid transfer of the block is modified such that the relay’s bid amount and the rebate can be adjusted accordingly. Builder 4 will receive back the rebates, the delta minus the relay fee. And if there are competing relays, a market-drive relay fee reduction is expected. As such, competitive relay may lead to relay centralization where builders are incentivized to send their bid exclusively to a single relay that gives the most rebates with the best latency. If we were to extrapolate the competitive relay market, we will likely see a future where each block will be broadcasted by a single winning relay, instead of the more collaborative one that we see today. This raises a few questions:</p><p><strong>1. How bad would the slot miss rate be if we assume all the blocks to be relayed by a single winning relay under the competitive relay market?<br>2. What is the cost of missed slots and what would that be for competitive relays?<br>3. What alternatives there are to improve the miss rate while better incentivizing relays?</strong></p><h3>Analysis of Slot Miss Rate via Single Relay Broadcasting</h3><p>Existing competitive relay proposals attempt to double down on the latency game where they charge the delta between the highest bid and the second-highest bid from <strong>other relays</strong>. The builder will receive up to the delta amount of rebates, and the relay subsequently charges a % fee. Most builders today send their bids and the block to multiple relays with the hope of maximizing the block inclusion. However, with bid rebates offered via competitive winning relay, builders now have an incentive not to submit their bids to other relays. Therefore, bids will converge over one or two relays that most builders trust. One of the immediate problems that may arise from such centralization is the <strong>higher chance of missed slots</strong> for the network. Since there are fewer relays to broadcast the block payload when called getPayload from proposers, there will be a higher slot miss rate. However, whether it’s actually higher, how much higher, and how it correlates to the number of relays that propagate the payload is yet to be investigated thoroughly. To provide more context in an attempt to answer those questions, here are the results based on analyzing 30 days of relay data (across Nov ~ Dec 2023).</p><p>To determine whether single-relay broadcasting, where only one relay broadcasts the payload, results in more missed slots, I have drawn Figure 2. which shows the % of slot miss rate within its corresponding relay type, single vs cooperative. The results here show that single relay broadcasting results in <strong>0.215%</strong> of missed slots which is <strong>3.18 times</strong> higher than that of cooperative relay broadcasting, which is <strong>0.0675%</strong>. The chart also indicates that single relay broadcasting constitutes about <strong>74.6%</strong> of the total missed slot across the network, while the other <strong>25.4%</strong> comes from cooperative payload broadcasting. Essentially 3 out of 4 missed slots are due to single relay broadcasting. We can observe that the slot miss rate is significantly higher for single-relay broadcasting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ns8DyZsMMQq4wFQaaierTA.png" /></figure><p><strong>Figure 2.</strong> Corresponding % of slot miss rate with payload broadcasted via single VS multiple relays (cooperative)</p><p>So far, I have grouped all cooperatives relaying under one bracket. But in reality, there could be 2, 3, or more relays participating in broadcasting, all of which may result in different slot miss rates. Hence, I have also categorized the slot miss rate under different numbers of broadcasting relays in Figure 3. This diagram shows that the slot miss rate decreases exponentially with an increasing number of broadcasting relays, and there are <strong>0 missed slots with 5 or more relays broadcasting the payload</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NWQvN2uKJYsytK5F0mLkww.png" /></figure><p><strong>Figure 3</strong>. Corresponding % of slot miss rate with different numbers of broadcasting relays</p><p>Some relays may have different network connectivity with diverse sets of registered validators, hence varied performance in payload broadcasting. To understand how different relays perform, Figure 4 categorizes the slot miss rate per relay during its single-relay broadcasting. The results suggest significant discrepancies in slot miss rate across relays where the highest miss rate is <strong>1.52%</strong> (manifold) while the lowest is <strong>0%</strong> (aestus). Other major relays like Flashbots, Ultra sound, and bloXroute have a miss rate of around <strong>0.2%</strong>. It’s worth noting that there is also a significant discrepancy across the sampling size of the payload delivered for each relay. Smaller relays like Eden and Manifold tend to have higher miss rates due to the lower number of total payloads delivered as noted in Table 1. Additionally, Aestus and Agnostic have only <strong>4~6%</strong> of their payloads broadcasted without any cooperative relays, which is considerably lower than others. This explains why those two have a relatively lower slot miss rate in Figure 4.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n7lBz3eJDBujrbSzG6p2SA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kqxQ4GoBN2ADbDrmSu_LYQ.png" /></figure><p><strong>Figure 4</strong>. % of missed slots per relay without any cooperative relay</p><p>Few observations have become clear after going through the data.</p><p><strong>1. Single relay broadcasting has an exponentially higher slot miss rate than cooperative relay broadcasting.<br>2. ¾ of the missed slots today come from single relay broadcasting.<br>3. Relays have varied slot miss rates. Well-connected relays (Flashbots, bloXroute, Ultra sound) have an average of 0.2% and less-connected relays 4–7x higher miss rate. Also, note that some less-connected relays can still achieve a low (in this case 0) miss rate by always participating in cooperative relaying.</strong></p><p>It requires more data to attribute the exact cause for the higher miss rate among single relay broadcasting. What we are seeing here is <strong>observations that show some correlations between single relay broadcasting and missed slots. But does not definitively provide the cause nor the attributions as to why such a correlation exists.</strong> But here are some possible reasons that I see so far.</p><ol><li><strong>Network topology:</strong> Some proposers might be positioned quite far from a relay. The roundtrip of sending getPayload and receiving and broadcasting that payloads take additional time that results in a higher chance of validators missing the attestations.</li><li><strong>Validator registrations:</strong> Some solo validators may only register with a few relays and have a narrower view of the incoming bids/builders. As such, it results in a higher miss rate when proposing due to other relays not receiving the getPayload calls.</li><li><strong>Relay Operation:</strong> Relay may have downtime, bugs, and bad network connectivity. This could be remediated via low-latency propagation networks like <a href="https://docs.bloxroute.com/introduction/bdn-latencies">bloXroute’s BDN 1</a> or <a href="https://fiber.chainbound.io/docs/use-cases/">Chainbound’s Fiber Network 5</a>. But relays are not well-incentivized at the end of the day ;((</li><li><strong>Builder’s bid bias:</strong> Certain builders could be sending bids only to a single relay, which contributes to a higher miss rate.</li><li><strong>Timing game:</strong> Proposer might be delaying the slot with the hope of receiving higher value blocks.</li></ol><p>Now that we have a better picture of the slot miss rate, I can also extrapolate what the miss rate under competitive relay will look like. Assuming all the bid fragments and split into corresponding relays of their choice to maximize the relay rebates, it will have a slot miss rate that ranges from ~0.13% to ~1.52% (Figure 4 &amp; Table 1). Given over 90% of the slots are broadcasted by major relays (Flashbots, Ultra sound, bloXroute), it’s safe to say that the current projected slot miss rate under a competitive market will be around 0.2%. With this number in mind, we can quantify the cost of missed slots and explore ways to incentivize cooperative relaying to minimize the slot miss rate.</p><p>Note: The number here does not account for missed slots induced by the timing game. Likely, the slot miss rate will be higher as the proposer timing game becomes more prevalent. Also, the number I have projected here is based on the Nov-Dec data and may not reflect the real missed slot since the relay and validator situations are constantly changing.</p><h3>Cost of competitive relays = Cost of proposers missing slots</h3><p>Currently, there are no penalties for missed slots for proposers and validators. The main “cost” here would be the opportunity cost of not receiving the rewards issuance. Here is the current rewards issuance scheme:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8P_AH3lblAJJXE6fsf-DPQ.png" /></figure><p><strong>Table 2</strong>. Rewards issuance weighting by <a href="https://eth2book.info/capella/part2/incentives/rewards/">eth2book 1</a></p><p>The rewards are calculated per epoch where each epoch consists of 32 slots. The total available rewards issuance across all validators N are:</p><p>NRA+32(512RY+RAp+RYp)=Tb</p><p>where N is the number of validators, RA is the rewards for a single attestation, RY is the reward for a single sync committee contribution, RAp is the reward for a block proposal due to attestations, RYp is the reward for a block proposal due to sync committee contributions, and Tb is the maximum issuance per epoch in Gwei. A proposer missing a slot results in the loss of RAp+RYp rewards per slot.</p><p>Let’s take the number of validators as 900,000 and calculate the total proposer rewards per slot. With <em>Eq 5</em> from the <em>Appendix</em>, it will be around <strong>0.042 ETH</strong> per slot in the issuance rewards with an additional few thousand Gwei of priority fees and builder’s bid rewards from MEV-Boost that could range from a few ETHs to tens of ETHs (an average bid of 0.14 ETH). <strong>The total opportunity cost on average will sum up to be around 0.1442ETH per missed slot</strong>.</p><p>The current average slot miss rate overall is:</p><p>Total missed slot / Total slot =359350323×100=0.10%</p><p>out of which 27.3% of the payloads are delivered via a single relay. Based on the current miss rate, the cost of missed slots per day is:</p><p>Cost of missed slot per day=7200×0.0010×0.144</p><p>=1.0ETH</p><p>Under competitive relay with single relay broadcasting, I assume the overall slot miss rate tends towards around 0.2%. As such the cost of missed slot per day will be:</p><p>Cost of missed slot per day=7200×0.0020×0.144</p><p>=2.0ETH</p><p>Cost of missed slot per year=2×365</p><p>=730ETH</p><p>While this number is a rough estimate, an extra few hundred ETH per year could certainly cover the cost of multiple relay operations while incentivizing cooperative relaying. In the future where relay centralization happens among competitive relays, <strong>the cost of missing slots will double for proposers. This is where cooperative relaying comes to the rescue.</strong> If proposers have <strong>4 or more relays cooperatively broadcasting the blocks, there would be almost no missed slot</strong>, reducing the missed slot cost for proposers. Note that the costs we calculated are for proposers, hence should not conflate it as a network cost.</p><h3>Missed slot = Relay Incentives??</h3><p>Missing slots result in an opportunity cost, which sets the upper bound for what relays could be feasibly paid to prevent such losses. Paying relays in this manner functions similarly to insurance. For solo validators especially, missing a slot incurs higher opportunity cost given the lower chance of proposing compared to large operators. To help reduce missed slots,I envision a viable path for relays to become profitable by offering Broadcasting as a Service (BaaS). To achieve this, however, we need a <strong>relay mesh network</strong> where a winning competitive relay could rapidly distribute the payloads post-commitment so that other relays can help propagate the payloads and minimize the missed slot.</p><h3>Direct Channel Relay Communication Network (Relay Mesh Network):</h3><p>To ensure that a competitive winning relay can quickly distribute block payloads and corresponding fees for other relays participating in BaaS, it is necessary to establish dedicated, low-latency direct communication channels between relays. These channels need to be significantly faster than the standard devp2p. One potential approach could be to set up a modified TCP protocol or utilize UDP for improved latency between known relays (reference <a href="https://observablehq.com/@libp2p-workspace/performance-dashboard">here 2</a> for more). By creating a low-latency relay communication network, we can enhance relay incentivization in two ways: firstly, by incentivizing the competitive relay through bid rebates, and secondly, by distributing a portion of the fees across other relays for cooperative broadcasting. There are existing low-latency networks offered, such as <a href="https://docs.bloxroute.com/introduction/bdn-latencies">BDN by bloXroute 1</a> that service both subscribing relays, traders/searchers, and validators. While the purpose is different from the relay mesh network, the goal of delivering trades/bids/payload fast is the same.</p><h3>Enshrining Missed Slot Penalties (a nice-to-have minimal viable enshrinement for relay incentivization):</h3><p>I originally started writing this piece to explore possible enshrinement for enabling relay incentivization. However, I realized after looking through data on missed slots that there is a relatively simple minimal viable enshrinement that will increase the relay profitability: missed slot penalty on proposers. Max has also prompted this idea nicely in his tweet here.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/740/1*rI7lj4LVhwypfOXyYknbZg.png" /><figcaption>Max’s <a href="https://twitter.com/MaxResnick1/status/1737574165480132824">post 1</a> on missed slot penalties on the proposer timing game</figcaption></figure><p>Although his comment on missed slot penalty was aimed at fixing the <a href="https://arxiv.org/abs/2305.09032">proposer timing game 1</a>, I believe it is beneficial for relays as well. There is a clear opportunity cost for missing a slot as a proposer, which sets the max bound for the fee that relays could earn. However, this bound could be further increased if there is a missed slot penalty on top of it. Assuming we take 15M base fee (100 Gwei) for the penalty as Max suggested, missing slots would cost additional 1,500,000,000 Gwei = 1.5 ETH which increases the cost of missed slots per day from about 1 ETH to 11.8 ETH per day. If under competitive relay, this number could increase up to 23.7 ETH per day. This is a very significant increase in cost but certainly should be taken with a grain of salt given the assumptions around penalty calculations and constant slot miss rate. Nonetheless, <strong>increasing the cost of missing slots will increase the value of BaaS and hence enable better cooperative relay incentivization on top of competitive relay rebates.</strong></p><h3>Challenges ahead With Missed Slot Penalties</h3><p>One major concern around missed slot penalties is raising the barrier of entry and disproportionate penalties to solo stakers. Unlike large operators, solo stakers have potentially sub-optimal network connectivity, and less capital/manpower to maintain the validators. According to Figure 5 &amp; 6, which look at missed slots from a validator angle, solo stakers roughly have a 4~6% miss rate while professional operators like Figments have 0.2~0.4% miss rate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/740/1*rI7lj4LVhwypfOXyYknbZg.png" /></figure><p>Max’s <a href="https://twitter.com/MaxResnick1/status/1737574165480132824">post 1</a> on missed slot penalties on the proposer timing game</p><p>Although his comment on missed slot penalty was aimed at fixing the <a href="https://arxiv.org/abs/2305.09032">proposer timing game 1</a>, I believe it is beneficial for relays as well. There is a clear opportunity cost for missing a slot as a proposer, which sets the max bound for the fee that relays could earn. However, this bound could be further increased if there is a missed slot penalty on top of it. Assuming we take 15M base fee (100 Gwei) for the penalty as Max suggested, missing slots would cost additional 1,500,000,000 Gwei = 1.5 ETH which increases the cost of missed slots per day from about 1 ETH to 11.8 ETH per day. If under competitive relay, this number could increase up to 23.7 ETH per day. This is a very significant increase in cost but certainly should be taken with a grain of salt given the assumptions around penalty calculations and constant slot miss rate. Nonetheless, <strong>increasing the cost of missing slots will increase the value of BaaS and hence enable better cooperative relay incentivization on top of competitive relay rebates.</strong></p><h3>Challenges ahead With Missed Slot Penalties</h3><p>One major concern around missed slot penalties is raising the barrier of entry and disproportionate penalties to solo stakers. Unlike large operators, solo stakers have potentially sub-optimal network connectivity, and less capital/manpower to maintain the validators. According to Figure 5 &amp; 6, which look at missed slots from a validator angle, solo stakers roughly have a 4~6% miss rate while professional operators like Figments have 0.2~0.4% miss rate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1006/1*a54t0IOz4XOI86dDA8EBqw.png" /></figure><p>Based on some napkin math, blindly imposing every validator with missed slot penalties will punish the solo stakers magnitudes more than professional stakers like figments. This will likely incentivize solo stakers to migrate to staking pools in an attempt to minimize penalties. Hence, if the missed slot penalties are not designed carefully, it could exacerbate the validator centralization problems already faced by Ethereum today. One possible approach could be to have discriminatory missed slot penalties based on the stake size, assuming the validator cap is raised. This echoes the two-tiered staking and more nuanced staking mechanisms proposed by Vitalik in this <a href="https://ethresear.ch/t/sticking-to-8192-signatures-per-slot-post-ssf-how-and-why/17989/22">post 1</a>. While such an approach is less egalitarian, professional operators who hold large amounts of ETH and have a highly optimized slot miss rate may be still sufficiently incentivized. This is still dependent on how the missed slot is calculated. There is a possibility that operators may break up their stake across many smaller validators despite the cap raise in an attempt to avoid missed slot penalties.</p><h3>Concluding thoughts</h3><p>Through observing the missed slots from various angles, we have evidence to suggest that</p><ol><li>Slot miss rate will double under competitive relay with single broadcasting, not accounting for timing games.</li><li>Slot miss rate can be improved exponentially with an increasing number of cooperatively broadcasting relays.</li><li>Missed slots result in an opportunity cost for the proposers and hence the winning competitive relay.</li></ol><p>From these observations, I have proposed two ideas to incentivize relays that enable both competitive and cooperative relaying.</p><ol><li>Creation of a low-latency relay mesh network for rapid payload distribution for further propagation, aka BaaS.</li><li>Enshrining missed slot penalty on proposers as a minimal viable enshrinement for relay incentivization, giving more reasons for a proposer to not miss a slot, aka use BaaS.</li></ol><p>While relay incentivization is challenging, it’s certainly not impossible with coordinated efforts by relay operators and modification to the current incentive for missed slots. I wish this piece uncovers the path to a future with profitable relays and gives some hope to those who are operating or thinking about operating a relay to support Ethereum PBS.</p><p>There is much more research work to be done to uncover the path forward for relays &amp; PBS. Here is an incomplete list of some open questions that I have so far:</p><ol><li>What network topology results in a higher missed slot for single relay broadcasting?</li><li>Could there be bid biases among builders submitting to multiple relays VS builders that don’t? And how does that contribute to the missed slots?</li><li>How and how much should we penalize proposers for the missed slot without exacerbating validator centralization? Could there be heterogenous penalties applied based on the stake amount, similar to the tiered staking model proposed by Vitalik?</li><li>Is this relay/acc???</li></ol><p>I am looking forward to further discussions among the communities and many work that comes after this. Please also feel free to DM <a href="https://twitter.com/ballsyalchemist">me 1</a> for any discussions.</p><h3>Appendix:</h3><p><strong>Eq 1</strong>. RAp+RYp=WpW∑−WpRAN32+512WpW∑−WpRY</p><p>where RAN32 is the total attestation rewards per slot (proposer reward scale with number of attestors)</p><p><strong>Eq 2</strong>. RA=14+26+1464×32b</p><p><strong>Eq 3</strong>. RY=Wy32×512×W∑Tb</p><p><strong>Eq 4</strong>. Base reward (b)=1,000,000,000×64√32,000,000,000×900,000=377Gwei</p><p><strong>Eq 5</strong>. RAp+RYp=900,00032×864−8×14+26+1464×32b+512864−8×232×512×64×900,000×32b</p><p>=40,897,768+1,514,732</p><p>=42,412,500Gwei</p><p>=0.042ETH (Using Eq 1, 2, 3, 4 in Appendix)</p><h3>Relevant resources:</h3><ul><li>[<a href="https://ethresear.ch/t/sticking-to-8192-signatures-per-slot-post-ssf-how-and-why/17989/22%5C">https://ethresear.ch/t/sticking-to-8192-signatures-per-slot-post-ssf-how-and-why/17989/22\ 1</a>](<a href="https://ethresear.ch/t/sticking-to-8192-signatures-per-slot-post-ssf-how-and-why/17989/22%5C">https://ethresear.ch/t/sticking-to-8192-signatures-per-slot-post-ssf-how-and-why/17989/22\ 1</a>)</li><li><a href="https://mirror.xyz/0xE21b1e6f471EDeF18264e9BBe51b7fA7643EE6B5/0Sh7BDW7qgH_nadfqF8bpmnjxnfoYzPFvRdmoIoi9mg">The Pursuit of Relay Incentivization — Ballsyalchemist 5</a></li><li><a href="https://ethresear.ch/t/timing-games-implications-and-possible-mitigations/17612">Timing Games: Implications and Possible Mitigations</a></li><li><a href="https://www.youtube.com/watch?v=kumD7njaCcU">https://www.youtube.com/watch?v=kumD7njaCcU 1</a></li><li><a href="https://bloxroute.com/pulse/introducing-the-validator-gateway-boost-your-ethereum-validator-rewards/">Introducing the Validator Gateway: Boost Your Ethereum Validator Rewards — bloXroute 3</a></li><li><a href="https://ethresear.ch/t/increase-the-max-effective-balance-a-modest-proposal/15801">Increase the MAX_EFFECTIVE_BALANCE — a modest proposal</a></li><li><a href="https://arxiv.org/abs/2305.09032">[2305.09032] Time is Money: Strategic Timing Games in Proof-of-Stake Protocols 1</a></li><li><a href="https://ethresear.ch/t/sticking-to-8192-signatures-per-slot-post-ssf-how-and-why/17989/22">Sticking to 8192 signatures per slot post-SSF: how and why — #22 by aivarasko 1</a></li><li><a href="https://observablehq.com/@libp2p-workspace/performance-dashboard">https://observablehq.com/@libp2p-workspace/performance-dashboard 2</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=33b0b724a812" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Pursuit of Relay Incentivization]]></title>
            <link>https://fenbushicapital.medium.com/the-pursuit-of-relay-incentivization-0d0e521d7108?source=rss-7c9ec2c710ab------2</link>
            <guid isPermaLink="false">https://medium.com/p/0d0e521d7108</guid>
            <dc:creator><![CDATA[Fenbushi Capital | 分布式资本]]></dc:creator>
            <pubDate>Wed, 21 Feb 2024 22:33:31 GMT</pubDate>
            <atom:updated>2024-02-21T22:33:31.281Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MEE4zXs7MQ456aGKGzr9Aw.png" /></figure><p><em>Thanks to </em><a href="https://twitter.com/tzhen?lang=en"><em>Tina Zhen</em></a><em>, </em><a href="https://twitter.com/drakefjustin"><em>Justin Drake</em></a><em>, </em><a href="https://twitter.com/uriklarman"><em>Uri Klarman</em></a><em>, </em><a href="https://twitter.com/mcutler"><em>Matt Cutler</em></a><em>, </em><a href="https://twitter.com/alextes"><em>Alex Tesfamichael</em></a><em>, </em><a href="https://twitter.com/mikeneuder"><em>Mike Neuder</em></a><em>, </em><a href="https://twitter.com/hasufl"><em>Hasu</em></a><em>, </em><a href="https://twitter.com/titanbuilderxyz"><em>Fabrizio</em></a><em>, </em><a href="https://twitter.com/0xkydo"><em>Kydo</em></a><em>, </em><a href="https://twitter.com/ralexstokes"><em>Alex Stokes</em></a><em>, and </em><a href="https://twitter.com/barnabemonnot"><em>Barnabé Monnot</em></a><em> for invaluable feedback and discussions.</em></p><p>Original article link: <a href="https://mirror.xyz/0xE21b1e6f471EDeF18264e9BBe51b7fA7643EE6B5/0Sh7BDW7qgH_nadfqF8bpmnjxnfoYzPFvRdmoIoi9mg">https://mirror.xyz/0xE21b1e6f471EDeF18264e9BBe51b7fA7643EE6B5/0Sh7BDW7qgH_nadfqF8bpmnjxnfoYzPFvRdmoIoi9mg</a></p><h3>Opening remark</h3><p>On September 26th, 2023, Blocknative announced its decision to halt its block building and relay operations. When I heard the news ahead of the announcement, I was surprised. Even though I was well aware that companies that were operating the relays for PBS were bleeding cash, I did not expect anyone to actually shut down. When Matt jokingly said, “We may only have a few relays left after next year without a proper incentive scheme” back at ETHCC in Paris, I assumed the community devs and researchers would eventually identify a sustainable solution for relay incentivization before we get to that state.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/994/1*oie5vV1Z-C7_jk4Ltkn4Mg.png" /><figcaption>“I care about you”</figcaption></figure><p>But now I realize that perhaps such a thought was a bit naive. I had been taking relays for granted and had thrown away all the rational and economic assumptions behind relays in the name of public good. That’s not good enough. So below is my attempt at reevaluating the relay incentive from the first principle with the goal of figuring out a viable pathway forward for relays without sacrificing decentralisation.</p><h3>Relay responsibilities</h3><p>Let’s take a brief look at the following graph illustrating the relay’s role within PBS.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cdMxG0nqyLvKzIcjbPjJiA.png" /><figcaption>Role of the relay by Kydo</figcaption></figure><p>In a nutshell, the builder submits the bid to relays, relays send that bid over to the proposer, and the proposer commits to the most valuable block. The relay then reveals the block body, enforces the builder’s payment to the proposer, and propagates the block to the rest of the network. For more details, please refer to <a href="https://ethresear.ch/t/mev-boost-merge-ready-flashbots-architecture/11177">this post</a> by Stephane. Clearly, relays play a critical role in our transaction supply network today. So why aren’t they getting properly incentivized? Here are a few reasons that I see so far.</p><p><strong>1. Relays were a byproduct of solving the proposer monopoly/scalability problems via PBS</strong></p><p>PBS was originally proposed to resolve the following risks associated with proposers:</p><ul><li><strong>Pre-PBS Validator monopoly over the block proposal and transaction sequencing resulting in protocol-level censorship risk.</strong> To illustrate, a sophisticated validator could earn more fees than solo stakers by extracting MEV. More rewards allow the spin-up of more validators, creating a snowball effect that leads to validator centralization. With PBS, the validator (now proposer) would no longer have visibility in the block sequencing pre-commitment and could focus on increasing the censorship resistance (CR) property of Ethereum via inclusion list (IL), which is ongoing research. It is likely that validator centralization occurs with or without PBS, but I see PBS as a mitigation to an extreme level of centralization that we already observe in the builder landscape.</li><li><strong>An increase in the node requirement for running bundle-merging algorithms and low-latency infra for bundle and orderflow aggregations.</strong> The pursuit of a revenue-maximising block may result in the centralization of the validator network via proposers’ increased workload and latency requirements associated with MEV capturing. As such, PBS offloads the sequencing work and latency game to builders to keep validator nodes lightweight.</li><li><strong>Added compute overhead from KZG proof.</strong> In the case of Darksharding, proposers will need to compute KZG proofs for up to 64MB of rollup data in less than 1 second. PBS allows specialized provers, possibly builders, to efficiently provide the proofs, thereby facilitating a lightweight implementation of validators.</li></ul><p>I believe PBS is a necessary step to pave the way for a more robust, scalable, and CR consensus network, but at the same time, it has successfully “offloaded” an awful lot of problems to builders and relays. For example, the following picture from <a href="https://censorship.pics/">censorship.pic</a> suggests that while validators are mostly non-censoring, there exist significant censorship problems associated with relays and builders (less decentralised and no IL).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WlKwDC4ZCXPOaFB3-xOiaw.png" /><figcaption>OFAC censorship by Toni, see <a href="https://censorship.pics/">https://censorship.pics</a></figcaption></figure><p>Furthermore, relay and builder problems were not the main focal point of the PBS. And among the two, builders had explicit economic incentives (MEV, CEX-DEX arbs) which has not been the case for relays.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K28ZesqUCFD-7k9Nl9GyTw.png" /><figcaption>Relays Questioning Their Existence by Matt Cutler</figcaption></figure><p><strong>2. “Relay is temporary, enshrining it will resolve the relay problems”</strong></p><p>ePBS is still being researched and has many <a href="https://notes.ethereum.org/@mikeneuder/infinite-buffet">open questions</a> that need to be answered. Different approaches bring different tradeoffs and protocol changes, some are more invasive than others, such as requiring longer slot time. There is a lot more work to be done to reach a conclusion/consensus on whether or not Ethereum should enshrine PBS at all. For the foreseeable future, relays are necessary to sustain the current PBS setup, and I think it’s paramount that we prioritise aligning the incentive of the existing stakeholders in the transaction supply network rather than waiting for a solution that may not ever be realised.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/992/1*B5PKgJn2tul6VWL7AFJj0Q.png" /><figcaption>Illustrating the life of a relay</figcaption></figure><p><strong>3. ‘Public good’ =&gt; Avoidance of incentive alignment issues among builders and validators =&gt; Toxic price competition</strong></p><p>Given those uncertainties mentioned above, running relay as a public good understandably became a temporary solution to avoid dealing with incentive alignment issues around fee charges within the txs supply network. However, that also led to a toxic price competition among those who are trying to monetize relay.</p><p>bloXroute attempted to monetise in the past by offering additional services, such as a low-latency network for block propagation. Their aim was to provide a relay network that is latency optimized, allowing them to charge fees based on the profit that builders gained when using the bloXroute relay. While the added revenue barely covered relay maintenance and operations costs (ranging from $100k-500k), it was a modest success in demonstrating the ability of a relay to charge a fee. However, as more public-good relays (like Agnostic and Ultrasound) capable of providing relatively low latency (<a href="https://github.com/michaelneuder/optimistic-relay-documentation/blob/main/proposal.md">optimistic relay</a>) enter the scene, the competition for users willing to pay for these services intensifies, making bloXroute’s fee model less effective over time. Without an immediate solution, the relay market will continue to be a contest to determine who can sustain the public good the longest with the deepest pockets.</p><h3>Future of the relay market</h3><p>So far, there are two approaches to reaching a more sustainable relay market under different assumptions.</p><p><strong>1. Double down on public good:</strong> Relays can collect donations from large DAOs and protocols to fund the operations. This is probably easier to execute but it’s uncertain how much can be gathered and for how long the public good funding will last. Simple but not sure how sustainable it is, unless the fund is in a large size (maybe ETF funds?!?). One great effort on that front is <a href="https://collective.flashbots.net/t/pbs-guild-proposal-v3-wip/2223">PBS guild</a> led by Tina, raising more awareness around the relay incentivization issue while funding relay research and operations via its grants. It’s certainly possible to have sustainable public good grants via PBS guild if there is an entity, like an ETF fund, that is willing to donate a portion of its profit to the guild like what <a href="https://blockchain.news/news/vaneck-pledges-10-percent-of-ethereum-etf-profits-to-protocol-guild">VanEck did for Protocol guild</a>. However, getting such a donation is extremely difficult and perhaps even unrealistic at this stage. Therefore, public good efforts like PBS guild could be considered as an added incentive for conducting more research and development around the relay incentivization mechanism.</p><p><strong>2. Resolve the incentive alignment issues:</strong> Form a relay union &amp; reach a <strong>social consensus</strong> to collectively devise a fee model that attempts to enable relay monetization while aligning the incentive of builders and validators. This requires coordination and collaboration among the stakeholders in our txs supply network, and it is the focus of this article to explore the means through which this can be achieved.</p><p><strong>Relay Union: an attempt to move on from public good via social consensus</strong></p><p>Some relay operators have <a href="https://hackmd.io/@KuDeTa/relay_guild_mvp">attempted</a> to align everyone’s incentives, including validators and builders, to find a sustainable path forward. To start off the discussion, here are important questions we need answered:</p><ol><li><strong>Who should pay for the relay cost?</strong></li><li><strong>How should the relay be paid? (e.g. who sets the fee and how?)</strong></li></ol><p>So let’s explore those two questions here.</p><h3>Who should pay for the relay cost?</h3><p>Intuitively, it seems reasonable that builders and validators, who heavily rely on and benefit from the services provided by relays, should bear the associated costs. In practical terms, both parties could collaborate by allowing relays to deduct a portion of bids, but there’s a need for one of them to facilitate this payment. It’s important to recognize that each party’s participation also introduces varying levels of influence and leverage for the relays. As a result, it becomes essential to determine which party is more inclined, if at all, to accommodate and compensate the relays. Essentially, we must assess <strong>the extent to which well-connected relays can persuade builders or validators to take the lead in facilitating these payments</strong>.</p><p>Note: I refer to “well-connected” to mean relays that have over one-third of the validator registered across the network.</p><p><strong>Relay’s leverage over builders:</strong></p><ul><li>Well-connected relays can deliver the bid to more validators, increasing the reach of builders’ bids.</li><li>Large builders, like Titan, Rsync, and Beaver, could be incentivized to create a vertically integrated builder-relay to avoid paying any additional relay fee as long as the <strong>cost of operating &amp; developing relay — additional profit from latency advantage in co-locating with relay &lt;&lt; cost of paying external relay fee</strong>. Large builders who have a significant amount of CEX-DEX activities and/or associated trading partners may profit from improved latency with integrated relay such that the cost of relay operations can be amortized. However, builder-relay in general comes with the following risk factors, hence confining the possible candidates for launching a successful builder-relay to only reputable and trustworthy builders.</li><li><strong>Large conservative validators may not connect to new relays</strong> <strong>out of security/risk concerns (esp large ones like Coinbase and Lido)</strong>. BD partnership and nepotism might help, but getting large numbers of validators to connect requires a good reputation for performance and reliability.</li><li><strong>A builder-relay has the potential to manipulate their bids</strong>. A builder-relay might claim a bid of 100 ETH when the actual payout, after the proposer commitment, is only 1 ETH. This kind of manipulation becomes more challenging when the relay operates as a separate, trusted entity, unless colluded with builders. It’s worth noting that such manipulation is less likely to occur with the top three builders due to concerns about their reputation. Therefore, it’s primarily the most reputable builders who could establish builder-relays that validators would be more likely to trust.</li><li><strong>A builder-relay could face the risk of being orphaned by validators if the same builder also submits bids through existing relays</strong>. Even when the builder runs its own builder-relay, it may still need to submit bids through the incumbent relays for broader reach and efficient broadcasting. This could potentially disincentivize validators from connecting to a new builder-relay, considering they can continue to receive bids from the same builders. Consequently, builders must differentiate their own relays from others to increase their chances of attracting more validators. This differentiation strategy can come at a significant cost; builders may need to consistently send higher bids across their own relay (which may be less-connected) and lower bids across others (which are better-connected), if they choose to do so. Additionally, to initiate validator connections, builders must engage in extensive business development partnerships and marketing efforts resulting in more cost overhead.</li></ul><p><strong>Relay’s leverage over validators:</strong></p><ul><li>Reputable relays broaden bid coverage from builders.</li><li>Listening to multiple relays increases the redundancy for payload deliveries and increases the validator’s coverage on non-overlapping bids submitted across relays. This table shows various statistics around relays over the past 7 days (10/20/2023) to illustrate the non-overlapping bids coming from different builders across the relays with validator registration counts. Based on Builder Count (Total) and Builder Count (Top4), we observe that relays that regularly receive bids from top 3 &amp; 4 builders have a high % of successful payload delivery. Note that this is a correlation and does not suggest any causal relationship between the high % of payload delivery and receiving bids from top builders. However, this data does suggest that a relay who has a good reputation and performance tends to receive a broader range of builder bids and achieve a better rate of payload delivery. As such, validators should listen to multiple relays to maximize the bid value.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h8KymEIC5WJSJgV_APh8LQ.png" /></figure><p>At this point, I hope it’s clear that both validators and builders need relays, and it is hard to say which party depends more on relays than the other. Builders selectively send their bids to relays of their choice and do not blindly send to every relay as people have suggested previously. Similarly, validators selectively listen to relays of their choice and do not blindly register with every relay, also due to regulatory concerns.</p><p>One could argue that relays may hold more leverage over builders for two reasons: first, it’s relatively easier to persuade builders to submit bids when there are enough validator registrations due to the performance advantages; and second, building a builder-relay is a non-trivial task. However, it remains inconclusive as to who should accommodate to facilitate the payment and bear the relay cost. Therefore, our next step is to explore how the fees can be distributed.</p><h3>How should the relay be paid? (e.g. who sets the fee and how?)</h3><p>There are two possible modes of payment: real-time vs deferred payment, where the former settles the payment while the payload delivery is performed and the latter settles the payment after. With that in mind, let’s discuss the technical feasibility of possible payment approaches by builders and validators, from which we will also explore who and how the fee is set.</p><p><strong>If builders pay:</strong> <strong><em>Real-time payment</em></strong></p><ul><li>Ultrasound has recently proposed a solution that enables real-time builder payment that also attempts to better align builders’ incentives. Here is an overview of how it works:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZzT7Af0lmAWZxkllK41_mg.png" /><figcaption>Diagram illustrating the real-time payment by builders</figcaption></figure><ul><li>When a builder wins the bidding, it typically submits a payload whose last transaction pays the bid to the proposers. During this process, relays perform the role of making sure that the last transaction is actually a bid transfer and that it is the bid amount the proposer agreed to. The Relay has full visibility and control of the payload during this process and has to be trustworthy. Ultrasound attempts to leverage that trust and modify the bid transfer where the relay, instead of the builder, pays the proposer up to the <strong>second highest bid</strong> seen across <strong>other relays</strong> such that the delta between the first and second bid can be distributed among builders and the relay. This allows bid rebates to builders, but enable this, builders need to opt-in by providing the following:</li><li>Post some collateral such that the relay would have the liquidity to pay the second-highest bid amount to proposers. Alternatively, the relay could also settle directly with builders after the payload deliveries, which may require the relay to be well-capitalized. The post-settlement of the liquidity is essentially a short-term credit provided by the relay to the builders. Therefore, the relay could explore charging interest on the short-term credit as another avenue of monetization.</li><li>Merkle path to the last transactions such that relay can modify the last transactions and hence the header.</li><li>This approach takes full advantage of the latency competition that is already happening among the builders/relays and tries to benefit from it. There are, however, a few concerns:</li><li>Since the winning builder receives rebates on the delta between their bid and the second-highest bid submitted via <strong>other relays</strong>. This will incentivize top builders to submit their bids ONLY to the best-performing relays (the lowest latency + broad validator coverage) to maximize their rebates. If most builders opt into Ultrasound’s approach, there will be little to no incentive for builders to submit their bids to multiple relays since that would reduce their rebates to zero, leading to a relay centralization where a single entity handles over a majority of the payload deliveries. This could potentially make relay space even more centralized than the already centralized builder space. Even if all top relays reach a social consensus on applying this fee model, it would result in <strong>bid and relay centralizations</strong>.</li><li>Relay competitions may lead to an inflation of % rebate, resulting in a race to the bottom on providing the highest rebate (lowest relay revenue). To mitigate this, it may require some social consensus among top competing relays to agree on rebate % and adjustment mechanisms.</li></ul><p>This proposed approach is fascinating and worth exploring more ideas on top of it. For example, instead of calculating the delta between the highest bids from other relays, we could explore using the delta of the top and second highest bids received within the same relay. As such, builders could submit bids to multiple relays as usual, and the fastest relay that delivered the top bids and the payload can charge the fee and give rebates. This will still demand latency competitions among relays but can prevent the over-centralization of relays by not disincentivizing builders’ bid submissions to competing relays.</p><p><strong>If builders pay: Deferred payment</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n-OByedY5ZrUOOLcBd4wLA.png" /><figcaption>Diagram illustrating the deferred payment by builders</figcaption></figure><ul><li>Relays will prepare a smart contract where builders could post a few ETH of fee collateral as a form of prepayment for relay usage. Every month (or some specific time interval), relays can check the total number of bids and winning bids submitted by a specific builder. Relays can then decide to charge builders according to their fee specifications, such as % charge on winning bids or fixed fee cost per winning bids etc.. The trust assumption here is that relays do not withhold the builders’ collaterals and do not charge additional fees beyond the fee specification that was agreed upon originally. The assumption here is fair for builders given that they are already trusting relays for many important tasks, such as relaying the bids/payloads and not unbundling and stealing MEV. This approach enables a flexible expression of relay fee preference that could better align builders’ and relays’ incentives while preventing over-centralization among relays. However, there are some concerns:</li><li>Deferred builder payment does not provide a rebate, hence not competitive against relays that adopt Ultrasound’s real-time payment approach (Latter provides the rebate while the former doesn’t)</li><li>Builders could try to bypass this deferred payment by running their own relay, but for the factors I mentioned previously, it may not be worthwhile for builders to spend the cost and effort to run their own relays if the fee charge is reasonable (based on winning bid etc.) and sufficiently low (5% per winning bids).</li><li>Deferred payment is simple and flexible, requiring only relatively small builder collateral as opt-in prepayment. However, attracting builders could be more challenging if the alternative relays offer rebates. Due to the smaller collateral requirement, it may better cater for builders with fewer resources who aren’t willing to post large collaterals.</li></ul><p><strong>If validators pay: Real-time payment</strong></p><ul><li>MEV-Boost currently maps the received bids with the corresponding relays via</li></ul><pre>relays[BlockHashHex(bidInfo.blockHash.String())] = append(relays[BlockHashHex(bidInfo.blockHash.String())], relay)</pre><p>such that proposers can keep track of all the received bid-relay pairs. This enables proposers to have visibility into the incoming bids with associated relays and the number of overlapping bids coming from different relays. However, the fee-recipient field in MEV-Boost is for distributing validator rewards to a designated address, usually the validator themselves. Hence the current MEV-Boost construct keeps track of the relay’s bids but does not allow payment to additional entities like relays. To enable that, there are two possible solutions:</p><ul><li>Modify MEV-Boost to enable adding an additional <strong>fee-recipient</strong> for relays. Also need to add necessary logic to determining the fee distribution mechanism, such as First-Come-First-Serve or equal splits among relays who handled the winning bid. However, going down this route would require Herculean effort in updating the MEV-Boost across the existing validators in PBS and is not realistic from the incentive alignment point; it makes little sense for validators to update their client so that they can get charged a relay fee.</li><li>Validators opt into PEPC/Eigenlayer where they commit to relay fee payment, e.g. % of the highest bid. This approach does not require much coordination but might be ineffective in delivering meaningful relay revenue. Validators may opt-in due to public good but have to bear the slashing risk, which clearly doesn’t align with the validator incentive. Regardless of whether having MEV-Boost modification or not, real-time payment by validators requires opt-in from thousands of node operators, and many may not choose to do so due to the reduced fee (bids) that they will receive in both scenarios.</li></ul><p><strong>If validators pay:</strong> <strong>Deferred payment</strong></p><ul><li>Similar to builder’s deferred payment, relays can set up contracts for validators to post ETH collateral and charge the fee according to fee preferences. This will likely be an opt-in model due to the difficulty in convincing and aligning thousands of node operators (much easier done with builders with collective effort). However, this could be an extension of the builder’s deferred fee mechanism where benevolent validators could contribute to the collateral pool in addition to the builders’.</li></ul><h3>Concluding thoughts</h3><p>After exploring the incentive dynamics around builders/validators and approaches for sustainable relay incentivization, a few things are clear:</p><p><strong>1.</strong> <strong>Public good (not charging a fee) will always lead to a race to the bottom</strong>: Even if Ultrasound and bloXroute try to capture value on the latency advantages, the value capture will be forced to zero if there are <strong>equally competitive and neutral</strong> public good relays providing similar latency-optimization.</p><p><strong>2.</strong> <strong>A social consensus among major neutral relays to pull out of public good is necessary</strong>: At least all the major relays of today (most of which are neutral) should collectively put an end to free relays and start experimenting with opting-in on proposed incentivization schemes.</p><p><strong>3. Easier to charge builders</strong>: Enforcing validator payments requires a tremendous amount of coordination and incentive alignment across thousands of node operators. This is made even harder given that the relay revenue always comes out of a proposer’s bid revenue. Additionally, validators hold leverage over relays in that more validator registrations enables more competitive relay, thus attracting more bids from builders. Hence, I believe it’s more practical to convince 20–30 builders and try to align their incentives.</p><p><strong>4.</strong> <strong>The tug-of-war between relay incentivization and centralization</strong>: In devising our relay incentivization scheme, we have to consider the centralization risk of the relays that emerge from latency competition. Ideally, we would want a coopetitive relay market that ensures collaborative payload deliveries for maximum liveness while still allowing monetization, which could be derived from latency advantages. It is also worth noting that the emergence of distributed block builders like SUAVE helps alleviate the need for relays as it offloads the trust overhead on txs validation to hardened TEE.</p><p>While there is much more work to be done in the quest to find the right approach for sustainably incentivizing relays, I hope that this writing can help clarify some of the under-explored incentive dynamics and technical challenges associated with the various approaches to resolving this relay incentive problem. I am also looking forward to seeing more future discussions around relay incentivization to push this initiative forward.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0d0e521d7108" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>