<?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 Candy798 on Medium]]></title>
        <description><![CDATA[Stories by Candy798 on Medium]]></description>
        <link>https://medium.com/@lixian_58397?source=rss-c8413ba2a07e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Rlxlu-aIH4AW-46aWCTTug.jpeg</url>
            <title>Stories by Candy798 on Medium</title>
            <link>https://medium.com/@lixian_58397?source=rss-c8413ba2a07e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 16:26:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@lixian_58397/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[AI Switches for 100G/200G/400G and 800G Ports By Asterfusion]]></title>
            <link>https://medium.com/@lixian_58397/ai-switches-for-100g-200g-400g-and-800g-ports-by-asterfusion-010ac7ddc1cb?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/010ac7ddc1cb</guid>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Fri, 30 Aug 2024 08:45:42 GMT</pubDate>
            <atom:updated>2024-08-30T08:45:42.935Z</atom:updated>
            <content:encoded><![CDATA[<h3>Table of Contents</h3><ul><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#the-ai-revolution-transforming-our-world">The AI Revolution: Transforming Our World</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#asterfusion-leading-the-charge-in-ai-networking-solutions">Asterfusion: Leading the Charge in AI Networking Solutions</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#asterfusion-the-ultra-ethernet-consortium">Asterfusion &amp; The Ultra Ethernet Consortium</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#asterfusion-ai-switches-portfolio">Asterfusion AI Switches Portfolio</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#asterfusion-your-ultimate-solution-for-ai-networking">Asterfusion: Your Ultimate Solution for AI Networking</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#unmatched-platform-flexibility">Unmatched Platform Flexibility</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#advanced-networking-capabilities-with-aster-nos">Advanced Networking Capabilities with AsterNOS</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#industry-leading-low-latency-with-cx-n-ultra-low-latency-switches">Industry-Leading Low Latency with CX-N Ultra Low latency Switches</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#promoting-an-open-ecosystem">Promoting an Open Ecosystem</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#efficient-and-flexible-ethernet-based-ai-networks">Efficient and Flexible Ethernet-Based AI Networks</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#innovative-technologies-for-optimal-performance">Innovative Technologies for Optimal Performance</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#asterfusion-ai-switches-offers-forward-compatible-products-with-uec-standard">Asterfusion AI Switches Offers Forward-compatible Products with UEC Standard.</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#embrace-the-future-of-ai-ml-data-center-network-infrastructure-with-asterfusion-ai-switches">Embrace the Future of AI/ML Data Center Network Infrastructure with Asterfusion AI Switches</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#compute-fabric-scalability-and-efficiency">Compute Fabric: Scalability and Efficiency</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#storage-fabric-redundancy-and-efficiency">Storage Fabric: Redundancy and Efficiency</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#in-band-management-network-seamless-connectivity">In-Band Management Network: Seamless Connectivity</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#out-of-band-management-network-comprehensive-control">Out-of-Band Management Network: Comprehensive Control</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#for-more-about-ai-network-traffic-and-ai-data-center-network">For more About AI Network traffic and AI Data center Network</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#unveiling-ai-data-center-network-traffic">Unveiling AI Data Center Network Traffic</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#embrace-the-future-of-ai-data-center-network-infrastructure-with-asterfusion">Embrace the Future of AI Data Center Network Infrastructure with Asterfusion</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#a-comparative-analysis-of-infini-band-and-ro-ce-for-ai-data-centers">A Comparative Analysis of InfiniBand and RoCE for AI Data Centers</a></li><li><a href="https://cloudswit.ch/blogs/ai-switches-for-100g-200g-400g-and-800g-ports/#cost-effective-aigc-network-solution-by-asterfusion-roc-ev-2-ready-switch">Cost Effective AIGC Network Solution by Asterfusion ROCEv2 Ready Switch</a></li></ul><h3>The AI Revolution: Transforming Our World</h3><p>Artificial Intelligence (AI) is not just a buzzword; it’s a groundbreaking force reshaping industries and our everyday lives. From healthcare to finance, and entertainment to education, AI is the catalyst for change, driving innovation and efficiency like never before. Imagine a world where real-time gaming, virtual reality, and the metaverse aren’t just futuristic concepts but integral parts of our daily interactions. This is the reality AI is crafting, fundamentally altering how networking, computing, and data storage systems operate.</p><h3>Asterfusion: Leading the Charge in AI Networking Solutions</h3><p>In this rapidly evolving landscape, Asterfusion stands out as a pioneer, offering cutting-edge networking solutions tailored for AI and machine learning workloads. As AI applications grow exponentially, the demand for robust, energy-efficient interconnects becomes critical. Asterfusion’s IP/Ethernet solutions are designed to meet these demands, ensuring seamless integration and optimal performance across vast networks of processors and storage systems.</p><p>Modern AI applications require networks that are not only fast but also reliable and scalable. We’re talking about interconnecting hundreds or thousands of accelerators at lightning speeds of 100Gbps, 200Gbps, and even up to 800Gbps. Asterfusion’s switches are engineered to deliver this high-bandwidth, low-latency performance, making them the go-to choice for building next-generation network fabrics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/666/1*giD_1cfRq85oL_h1Sjig7g.png" /></figure><h3>Asterfusion &amp; The Ultra Ethernet Consortium</h3><p>In March 2024, Asterfusion took a significant step forward by joining the Ultra Ethernet Consortium (UEC), a collaborative effort led by the Linux Foundation and some of the world’s top tech companies. The UEC aims to revolutionize Ethernet technology, breaking through traditional performance bottlenecks to make it suitable for AI and high-performance computing networks.</p><p>As a full member, Asterfusion is at the forefront of developing the next-generation communication stack architecture, ensuring Ethernet’s continued relevance and efficiency in the AI era.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XtD-vhB2mKfMmGs4FLS-BA.png" /></figure><h3>Asterfusion: Your Ultimate Solution for AI Networking</h3><p>When it comes to AI networking, Asterfusion stands out with a suite of advantages designed to deliver unmatched performance and flexibility. Let’s dive into why Asterfusion is at the forefront of this transformative technology.</p><h3>Unmatched Platform Flexibility</h3><p>Asterfusion offers a versatile range of box platforms, from compact 1U units to robust 2U setups. These can be deployed individually for smaller clusters or combined to support expansive topologies of over 100,000 accelerators. The CX-N series is particularly noteworthy, featuring a vast array of ports including 800G, 400G, 200G, and 100G, with capacities ranging from 2T to an astounding 51.2Tbps. Their latest 800G AI switch is a game-changer, boasting ultra-large capacity with 64 x 800G Ethernet ports with a total switching capacity of 51.2T,having the world’s fastest switch with port-to-port latency under 560ns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/474/1*TORMBx4KPLKKmf0Tlw-KAg.png" /></figure><p><a href="https://cloudswit.ch/product/800gbe-switch-512tbps-enterprise-sonic/">Asterfusion 800GbE Switch with 64x OSFP ports, 51.2Tbps, Enterprise SONiC Ready</a></p><h3>Advanced Networking Capabilities with AsterNOS</h3><p>At the heart of Asterfusion’s offerings is the self-developed <a href="https://cloudswit.ch/product/sonic-enterprise-distribution/">Asterfusion Enterprise SONiC Distributio</a>n — AsterNOS. This is the cornerstone of cloud network solutions for next-gen data centers and high-performance AI networks. AsterNOS is equipped with features that ensure high reliability, quality, and performance, delivering lossless, high-bandwidth, low-latency networks. It supports ROCEv2 and EVPN-multihoming, crucial for implementing low-latency, lossless networks and running high-value workloads efficiently.</p><h3>Industry-Leading Low Latency with CX-N Ultra Low latency Switches</h3><p>The <a href="https://cloudswit.ch/product-category/asterfusion-ultra-low-latency-sonic-based-data-center-switch-by-marvell-teralynx/"><strong>Asterfusion CX-N</strong> <strong>series</strong></a> is engineered for those who demand the best in low-latency switching capabilities. Leveraging <a href="https://www.marvell.com/content/dam/marvell/en/public-collateral/switching/marvell-teralynx-7-product-brief.pdf">Marvell Teralynx 7</a> and <a href="https://www.marvell.com/content/dam/marvell/en/public-collateral/switching/marvell-teralynx-10-data-center-ethernet-switch-product-brief.pdf">Marvell Teralynx 10 </a>technologies, it offers unparalleled performance across various applications, achieving latencies as low as ~400ns. Its extensive on-chip packet cache, over 200MB in capacity, drastically reduces storage and forwarding latency of RoCE traffic during collective communications, making it ideal for cutting-edge AI applications.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/1*6DWjICWvS-zg5E6HdkthFQ.png" /></figure><h3>Promoting an Open Ecosystem</h3><p>Asterfusion is committed to fostering an open ecosystem for AI networks, collaborating with industry leaders to ensure flexibility and interoperability. With a broad Ethernet-based ecosystem that includes multiple system vendors, silicon providers, interconnects, and optics, Asterfusion empowers customers with complete freedom of choice. This open ecosystem supports REST API integration, enabling seamless collaboration with computing devices and facilitating automatic GPU cluster deployment.</p><h3>Efficient and Flexible Ethernet-Based AI Networks</h3><p>Asterfusion’s Ethernet-based networks are designed for interoperability, streamlining efficient and flexible designs that eliminate compatibility issues. This versatility allows Asterfusion’s solutions to be consistently deployed across general-purpose compute, data center, storage, and AI networks, avoiding the pitfalls of inter-domain gateways and pipeline bottlenecks.</p><h3>Innovative Technologies for Optimal Performance</h3><ul><li><strong>INNOFLEX Programmable Forwarding Engine</strong>: This innovative engine dynamically adjusts the forwarding process in real-time based on business needs and network status, minimizing packet loss from congestion and failures.</li><li><strong>FLASHLIGHT Traffic Analysis Engine</strong>: It measures packet delay and round-trip time in real-time, utilizing intelligent CPU analysis for adaptive routing and congestion control.</li><li><strong>High-Precision Time Synchronization</strong>: With 10 nanosecond PTP/SyncE synchronization, Asterfusion ensures synchronous computing across all GPUs, enhancing collaborative processing capabilities.</li><li><strong>Load-aware Balancing</strong>:The system’s load-aware per flowlet/packet balancing prevents congestion, ensuring efficient bandwidth utilization.</li><li><strong>Active Queue Management:</strong> active queue management, utilizing Explicit Congestion Notification (ECN), proactively avoids traffic congestion, maintaining smooth network operations</li></ul><h3>Asterfusion AI Switches Offers Forward-compatible Products with UEC Standard.</h3><p>As the Ultra Ethernet Consortium (UEC) completes its expansion to improve Ethernet for AI workloads, Asterfusion is building products that will be ready for the future. The Asterfusion CX-N AI data centre switch portfolio is the definitive choice for AI networks, leveraging standards-based Ethernet systems to provide a comprehensive range of intelligent features. These features include dynamic load balancing, congestion control, and reliable packet delivery to all ROCE-enabled network adapters. As soon as the UEC specification is finalised, the Asterfusion AI platform will be upgradeable to comply with it.</p><h3>Embrace the Future of AI/ML Data Center Network Infrastructure with Asterfusion AI Switches</h3><p><a href="https://cloudswit.ch/solution/rocev2-ai-solution-with-dgx-superpod/">RoCEv2 AI Solution with NVIDIA DGX SuperPOD</a></p><p>In the world of AI and machine learning, the NVIDIA DGX SuperPOD™ with NVIDIA DGX™ H100 systems represents the pinnacle of data center architecture. Asterfusion is pioneering a new approach, harnessing the power of RoCEv2-enabled 100G-800G Ethernet switches to build DGX H100-based superPODs.<br>This innovative solution replaces traditional InfiniBand switches, encompassing compute fabric, storage fabric, and both in-band and out-of-band management networks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mDVFwth2-qQt6IW5j4fpbA.png" /></figure><h3>Compute Fabric: Scalability and Efficiency</h3><p>When embarking on a research project or proof of concept, a small-scale cluster is often the ideal starting point. With fewer than four DGX H100 nodes, a single <a href="https://cloudswit.ch/product/32-port-400g-qsfp-dd-data-center-switch-enterprise-sonic-teralynx/">32x400G switch</a> can connect the entire cluster, utilizing QSFP-DD transceivers on switch ports and OSFP transceivers on ConnectX-7 NICs, linked by MPO-APC cables.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XpRAA-vlKs_1ucSgBn9ERg.png" /></figure><p>As the number of GPU servers grows to eight, a 2+4 CLOS fabric configuration optimizes rail connectivity, maximizing the efficiency of GPU resources. <a href="https://cloudswit.ch/blogs/what-is-leaf-spine-architecture-and-how-to-build-it/">Spine and leaf </a>switches interconnect via QSFP-DD transceivers with fiber or QSFP-DD <a href="https://cloudswit.ch/blogs/aoc-dac-acc-aec-modules-complete-overview/">AOC/DAC/AEC cables.</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J6oCws5uca3RjZ37rsFKCQ.png" /></figure><p>For even greater capacity, the forthcoming <a href="https://cloudswit.ch/product/800gbe-switch-512tbps-enterprise-sonic/">CX864E-N switch</a> (51.2Tbps Marvell Teralynx 10 based 64x800G switch, available in 2024Q3) can support up to 16 GPU servers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CpADriEvHGSh8TeVVhCjdw.png" /></figure><p>Asterfusion’s scalable compute fabric is built upon modular blocks, enabling rapid deployment at multiple scales. Each block comprises a group of GPU servers and 400G switches (such as 16 DGX H100 systems with 8 CX732Q-N switches). Rail-aligned, non-oversubscribed block design ensures optimal compute fabric performance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vfT59YaER9GcXEqtHXVpPw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SkBATCfrPwlv3cFaHnZLeA.png" /></figure><p>To further scale the fabric, higher-throughput switches like the CX864E-N can replace spines, expanding the network to a maximum of 16 blocks and 256 nodes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hVT-C0rSlJuz2R751Pc1kA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r2-fVADcCq8jOiEbBEANvw.png" /></figure><p><em>Table 1. Compute fabric component count of different scales</em></p><h3>Storage Fabric: Redundancy and Efficiency</h3><p>Storage specifications vary by vendor and product. Assuming each node offers 45GB/s read performance via 4x200G interfaces, each DGX H100 node requires 8GB/s read performance (as per NVIDIA documentation), totaling 256GB/s across the cluster (necessitating six storage nodes in this scenario).</p><p>While MC-LAG is typically used for redundancy, Asterfusion recommends<a href="https://cloudswit.ch/blogs/asterfusion-enterprise-sonic-nos-evpn-multi-homing/"> EVPN Multi-Homing</a> to minimize switch port requirements and enhance storage fabric efficiency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bHyT3b828_f0vV9DiW-Faw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IegN-F2osHK_eni3QGKJcA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I5GFVGDxNqc3bPRPbe1Wxw.png" /></figure><h3>In-Band Management Network: Seamless Connectivity</h3><p>The in-band network provides connectivity for both in-cluster services and external services, linking all compute nodes and management nodes (general-purpose x86 servers).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*evzWkyGWdd10EsabhTPt3Q.png" /></figure><h3>Out-of-Band Management Network: Comprehensive Control</h3><p>Asterfusion’s 1G/10G/25G enterprise switches, which also support Leaf-Spine architecture, offer robust out-of-band management capabilities, ensuring control over all superPOD components.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ESQojTbECTudLl_aqdXAaQ.png" /></figure><p>Asterfusion’s RoCEv2-enabled Ethernet solution for NVIDIA DGX H100-based superPODs redefines the possibilities for AI and machine learning infrastructure. By offering a scalable, efficient, and highly performant alternative to traditional InfiniBand-based architectures, Asterfusion empowers organizations to unlock the full potential of their AI workloads.</p><h3>For more About AI Network traffic and AI Data center Network</h3><h3><a href="https://cloudswit.ch/blogs/ai-data-center-network-traffic/">Unveiling AI Data Center Network Traffic</a></h3><h3><a href="https://cloudswit.ch/blogs/future-of-ai-data-center-network-infrastructure/&#39;">Embrace the Future of AI Data Center Network Infrastructure with Asterfusion</a></h3><h3><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/">A Comparative Analysis of InfiniBand and RoCE for AI Data Centers</a></h3><h3><a href="https://cloudswit.ch/blogs/optimize-aigc-network-for-llm-training/">Cost Effective AIGC Network Solution by Asterfusion ROCEv2 Ready Switch</a></h3><p>Or please contact lixian@asterfusion.com for more details</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=010ac7ddc1cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[100G Data Center Switches: Driving the Future through Open Networking]]></title>
            <link>https://medium.com/@lixian_58397/100g-data-center-switches-driving-the-future-through-open-networking-1865a6cf9a6b?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/1865a6cf9a6b</guid>
            <category><![CDATA[hpc-data-center]]></category>
            <category><![CDATA[hpc]]></category>
            <category><![CDATA[open-networking]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Thu, 22 Aug 2024 02:33:57 GMT</pubDate>
            <atom:updated>2024-08-22T02:33:57.047Z</atom:updated>
            <content:encoded><![CDATA[<h3>Table of Contents</h3><ul><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#what-is-100-g-switch">What is 100G Switch?</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#how-to-choose-a-100-g-data-center-switch">How to choose a 100G Data Center Switch?</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#asterfusion-100-g-data-center-switches-powered-by-open-network">Asterfusion 100G Data Center Switches powered by open network</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#the-features-of-100-g-data-center-switches">The features of 100G Data Center Switches</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#hardware-features">Hardware features</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#marvell-teralynx-7-unmatched-low-latency-power">Marvell Teralynx 7: Unmatched Low-Latency Power</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#versatile-interface-options-for-asterfusion-100-g-data-center-switches">Versatile Interface Options for Asterfusion 100G Data Center Switches</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#software-features-pre-installed-enterprise-grade-so-ni-c-nos">Software Features: Pre-installed Enterprise-grade SONiC NOS</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#ro-c-ev-2-enables-lossless-transmission-in-data-center-networks">RoCEv2 Enables Lossless Transmission in Data Center Networks</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#evpn-multihoming-optimizes-network-load-balancing">EVPN Multihoming Optimizes Network Load Balancing</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#field-test-ultra-low-latency-ethernet-switch-vs-infiniband-switch">Field test : Ultra-Low Latency Ethernet Switch VS. Infiniband Switch</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#ro-c-ev-2-switches-test-result-in-hpc-network">RoCEv2 Switches Test Result in HPC Network</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#test-result">Test Result</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#for-more-text-in-hpc-and-ai">For more text in HPC and AI</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#48-25-g-8-100-g-qsfp-28-uplinks-data-center-leaf-tor-switch">48*25G+8*100G QSFP28 uplinks Data Center Leaf (TOR) Switch</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#speedy-little-light-beams-the-magic-of-100-g-optical-modules-in-data-centers">Speedy Little Light Beams: The Magic of 100G Optical Modules in Data Centers</a></li><li><a href="https://cloudswit.ch/blogs/100g-data-center-switches-of-open-networking/#references">References:</a></li></ul><h3>What is 100G Switch?</h3><p>The 100G Ethernet Switch is a high-performance network device engineered to provide throughput of up to 100 gigabits per second (Gbps). Acting as a central switch within a network, it interconnects multiple devices, enabling swift and efficient data exchange. By leveraging 100G Ethernet technology, these switches offer substantially increased transmission capacity with minimized latency, significantly boosting data transfer speeds in data centers. This capability makes them particularly well-suited for managing the growing demands of contemporary high-bandwidth applications.</p><h3>How to choose a 100G Data Center Switch?</h3><p>When selecting a 100G data center switch, it’s essential to consider several pivotal factors to ensure the device not only meets your current needs but also scales for future demands. Here’s a more dynamic look at what to keep in mind:</p><p><strong>Performance</strong>: The heart of any 100G data center switch is its performance. The chipset and CPU are the driving forces behind its ability to process data, manage high-speed traffic, and handle vast volumes without bottlenecks. Opt for switches equipped with cutting-edge ASICs (Application-Specific Integrated Circuits) that promise high throughput and low latency. Ensure the CPU can adeptly manage control plane tasks, allowing the switch to efficiently handle both routine and complex operations in a dynamic data center environment.</p><p><strong>Hardware Design</strong>:Hardware design plays a crucial role in both maintenance ease and switch reliability. Look for hot-swappable components like power supplies and fans to minimize downtime during maintenance or upgrades. Switches also need robust thermal capacity to handle high heat loads, making the design of internal cooling channels vital. Proper fan system design and layout ensure optimal airflow, reducing hotspots and keeping temperatures in check. You can even choose the direction of fan airflow for data center switches.</p><p>Plus, the number of ports directly impacts scalability — high port density switches support more connections, cutting down the need for additional switches.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PMp-5NqM4A9OoWLTsw01YA.png" /></figure><p><strong>Software</strong>: Software capabilities define the functionality and manageability of your switch. Look for a switch with a robust network operating system that delivers comprehensive management tools, automation, and telemetry features. As we advance into the AI era, the demand for higher bandwidth, lower latency, and stable data transmission is crucial for large-scale model training. With the rise of open networking, SONiC stands out as a flexible, scalable network OS with a robust ecosystem, widely supported and utilized in the open-source community. It’s an excellent choice when considering data center switches.</p><h3>Asterfusion 100G Data Center Switches powered by open network</h3><p>Asterfusion’s CX-N series switches are designed for ultra-low latency, offering industry-leading forwarding speeds. Equipped with enterprise-grade SONiC NOS, they unlock advanced features for data centers. The series includes two 100G switches: the CX532P-N, featuring 32*100G ports, and the CX564P-N, boasting 64*100G ports. These switches cater to a variety of user needs, ensuring flexibility and performance in demanding environments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MptLHdfUOaUfp5Qlxdfx3A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j5oISrA5X5X2BTXSfennmg.png" /></figure><h3>The features of 100G Data Center Switches</h3><h4>Hardware features</h4><p>Marvell Teralynx 7: Unmatched Low-Latency Power</p><p>The Asterfusion CX532P-N and CX564P-N switches deliver lightning-fast performance powered by the ultra-efficient Marvell Teralynx 7 chipset.This cutting-edge chipset minimizes forwarding latency by combining a high-speed hardware acceleration engine with a non-blocking switching architecture, ensuring that packets move through the switch with minimal delay for exceptional network speed and efficiency.</p><p>With a mere 520ns latency for the CX532P-N and an even faster 400ns for the CX564P-N, these switches excel in environments where every nanosecond counts. Perfect for high-frequency trading, financial data centers, and cutting-edge AI or machine learning applications, they deliver the real-time performance that mission-critical operations demand.</p><p>At the end of the article there are forwarding latency related test reports.</p><p>Versatile Interface Options for Asterfusion 100G Data Center Switches</p><p>The CX532P-N and CX564P-N 100G data center switches offer versatile port options. The 100G ports can backwards compatible to 40G or break out into four 25G or four 10G, giving network administrators flexible configuration choices.</p><ul><li><strong>CX532P-N</strong>: 32x 100G ports, ideal for small data centers with tight budgets.</li><li><strong>CX564P-N</strong>: 64x 100G ports, perfect for large data centers needing high-density connectivity and scalability.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*shrhpzIJ5vjHSban675eYQ.png" /><figcaption><em>CX564P-N</em></figcaption></figure><p>Both switches include an RJ45 management port, RJ45 console port, and USB2.0 port for easy configuration and monitoring. What’s more,they also feature two additional 10G ports for management tasks like<a href="https://cloudswit.ch/blogs/what-is-int%ef%bc%88in-band-network-telemetry%ef%bc%89/"> INT Telemetry</a>.</p><h4>Software Features: Pre-installed Enterprise-grade SONiC NOS</h4><p>Asterfusion 100G switches are equipped with an Enterprise <a href="https://sonicfoundation.dev/">SONiC</a> NOS, finely tuned for data center environments. Compared to the community version, it provides advanced features, enhanced stability, and dedicated after-sales support. The device includes the KLISH command line, allowing traditional network engineers to transition smoothly to open networking, reducing learning costs. Since 2017, Asterfusion’s R&amp;D team of over 100 experts has focused on perfecting the enterprise version of SONiC. Over the years, this version has accumulated valuable experience and found success in AI, education,telecommunications, and cloud providers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uRguJUuTsv9JKcqE3IwNAQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0CKlvvZZrcFGovcbXqJWAg.png" /></figure><p>RoCEv2 Enables Lossless Transmission in Data Center Networks</p><p>The Asterfusion CX-N series supports RoCEv2.</p><p>The CX532P-N and CX564P-N switches bring the power of Remote Direct Memory Access (RDMA) over Ethernet to your network, tackling server-side data computation latency and freeing up the CPU for critical tasks like running applications and processing large datasets. This technology boosts bandwidth while slashing latency, jitter, and CPU load.</p><p>With Priority Flow Control (PFC), RoCE flows get their own priority lanes, while Explicit Congestion Notification (ECN) smartly predicts and prevents network congestion. Enhanced Transmission Selection (ETS) ensures consistent service quality for similar traffic types, fully guaranteeing bandwidth, and Data Center Bridging Exchange (DCBX) streamlines the configuration of DCB parameters for RoCE devices.</p><p>Plus, RoCEv2 delivers ultra-low-latency, lossless networking at less than one-third the price of infiniband, giving customers exceptional performance without breaking the bank.</p><p>For more information on ROCE and Infiniband and which one you should choose, refer to <a href="https://cloudswit.ch/blogs/roce-for-hpc-test-data-and-deploy-on-sonic/">https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/</a></p><p><strong>EVPN Multihoming Optimizes Network Load Balancing</strong></p><p>The CX532P-N and CX564P-N integrate advanced technologies like EVPN Multi-homing for redundancy and load balancing, BGP EVPN for managing network information, and VXLAN for large-scale network expansion through tunneling. In practical scenarios, these switches can offload virtual networks and distributed network function gateways from the computing space, directly hosting them on the devices. This enhances server computing power, simplifies the creation of virtual nodes, and maximizes CPU efficiency, resulting in an efficient, reliable, and scalable data center network. For more details, please refer to <a href="https://cloudswit.ch/blogs/asterfusion-enterprise-sonic-nos-evpn-multi-homing/#why-evpn-multi-homing">https://cloudswit.ch/blogs/asterfusion-enterprise-sonic-nos-evpn-multi-homing/#why-evpn-multi-homing</a></p><h3>Field test : Ultra-Low Latency Ethernet Switch VS. Infiniband Switch</h3><p>The following is a comparison test in a distributed storage network scenario, and the devices participating in the test are all 32x100G specifications. Comparing the test results, we can clearly see that the overall performance of Asterfusion’s CX532P-N is better than Mellanox’s SB7700, and the latency level is lower.</p><p>In this storage scenario test, the CX532P-N switch outperforms the Infiniband switch (SB7700) in various features, and the CX-N series, which is a general-purpose Ethernet hardware, is only about 50% of the price of the dedicated device.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l86jlkCHcz29DFuPVFUusw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ytnvf924NM8fl9-MXYYD6Q.png" /></figure><h3>RoCEv2 Switches Test Result in HPC Network</h3><p>We build an <a href="https://cloudswit.ch/blogs/asterfusion-low-latency-switch-vesus-infiniband-switch/">HPC</a> test network using an <strong>Asterfusion CX564P-N</strong> ultra-low latency switch and a <strong>Mellanox InifiBand</strong> switch to compare their performance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Txc48Jm7C2ricXH5bB7idA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HNSgy0HM4rQP31QA2jT_fw.png" /></figure><h4>Test Result</h4><p>Layer 2 network, RoCEv2 enabled on CX532P-N, HPC node (Mellanox NICs) optimized for RDMA</p><p><strong>E2E forwarding performance test</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KqtW46bBrQkmymOmIYRgDQ.png" /></figure><p><strong>HPC application performance test (WRF)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B6srYl21gqVkA9QxyCLAKg.png" /><figcaption>For more text in HPC and AI</figcaption></figure><p>Click here to view additional test data on Asterfusion data center switches in AI: <a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/">https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/</a></p><p>Click here to view additional test data on Asterfusion data center switches in HPC scenarios:<a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#ro-c-ev-2-switches-test-result-in-aigc-network">https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#ro-c-ev-2-switches-test-result-in-aigc-network</a>‘</p><h3>48*25G+8*100G QSFP28 uplinks Data Center Leaf (TOR) Switch</h3><p>We’ve already explored the spine switch, so now let’s dive into the impressive leaf switch: the CX308P-N. This powerhouse features 8 upstream 100G ports and 48 downstream 25G ports, making it a versatile choice for your network needs.</p><p>For more details. please contact lixian@asterfusion.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1865a6cf9a6b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[5G O-RAN Mobile xHaul Transport Solution]]></title>
            <link>https://medium.com/@lixian_58397/5g-o-ran-mobile-xhaul-transport-solution-b05674d11e14?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/b05674d11e14</guid>
            <category><![CDATA[5g]]></category>
            <category><![CDATA[5g-technology]]></category>
            <category><![CDATA[5g-network]]></category>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Fri, 26 Jul 2024 09:36:04 GMT</pubDate>
            <atom:updated>2024-07-26T09:36:04.904Z</atom:updated>
            <content:encoded><![CDATA[<h3>O-RAN Overview</h3><p>Open Radio Access Networks (O-RAN) is transforming the telecommunication industry. It changes traditional appliance-based RAN deployment to a more open, interoperable, standard-based, and virtualized radio access network. Open RAN deployments remove vendor lock-in and enable telecom service providers to take advantage of virtualization, utilize telecom edge for innovative apps and services, and save costs.</p><p>Asterfusion has been committed to building the open network since its inception. We promise to deliver the leading connectivity solutions and services through productizing open software and hardware solutions for service providers and new experiences for their customers.</p><h3>Mobile xHaul Transport</h3><p>The current Open RAN architecture defined by <a href="https://www.o-ran.org/specifications">O-RAN Alliance</a> provides the functional separation of RAN components into a Open Radio Unit (RU), Open Distributed Unit (O-DU) and Open Centralized Unit (O-CU). Ethernet switches are needed for xHaul (FrontHaul, MidHaul, BackHaul) transmission to facilitate efficient and flexible communication between different components of O-RAN.</p><p>Asterfusion provides comprehensive telecom solution to enable the interconnection of various network elements within the O-RAN architecture.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UObojw4V3fp8ABFs.png" /></figure><h3>Key Benefits</h3><p>Comprehensive Mobile Transport Solution</p><ul><li>High Precision Time Synchronization FrontHaul</li><li>Low Latency MidHaul/BackHaul Transmission</li></ul><p>Time Synchronization Protocol Support</p><ul><li>IEEE1588v2 (PTP)</li><li>Synchronous Ethernet</li></ul><p>Enterprise ready SONiC NOS</p><ul><li>CPRI-to-eCPRI conversion</li><li>QoS and EVPN to support Network Slicing</li></ul><h3>Mobile In-Building</h3><p>The current coverage provided by traditional cells will not support ever-growing demands for cellular services in heavy populated areas such as large office buildings, university campuses, hospitals and stadiums. An overlay of small cells must be added to the existing deployment of macro-cells in dense areas.</p><p><a href="https://cloudswit.ch/product-category/sonic-access_aggregation_switch/">Asterfusion CX-M Series</a> is a popular Campus Access and Aggregation switch for O-RAN small-cell deployment at in-building locations because of its support of PTP, SyncE, PoE+, and line rates of 1G, 10G, 25G and 100G ports.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3uFsT9yD7TOM7ZIC.png" /></figure><h3>Key Benefits</h3><ul><li>Rich L2/L3 Features</li><li>PTP and SyncE sychronization support</li><li>Active PoE up to 60W per port</li></ul><p>For more details. Please contact: lixian@asterfusion.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b05674d11e14" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Asterfusion 32-Port 100G L3 Low Latency Leaf/Spine Switch for Data Center, Enterprise, AI, HPC…]]></title>
            <link>https://medium.com/@lixian_58397/asterfusion-32-port-100g-l3-low-latency-leaf-spine-switch-for-data-center-enterprise-ai-hpc-ccd0a0120691?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/ccd0a0120691</guid>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Fri, 14 Jun 2024 08:55:34 GMT</pubDate>
            <atom:updated>2024-06-14T08:55:34.440Z</atom:updated>
            <content:encoded><![CDATA[<h3>Asterfusion 32-Port 100G L3 Low Latency Leaf/Spine Switch for Data Center, Enterprise, AI, HPC ,Ceph and NVME</h3><p>The Asterfusion CX532-N is a 32 x 100G QSFP28 data center switch designed for leaf spine architecture. It offers low latency and boasts a line-rate L2/L3 up to 3.2Tbps switching performance by powerful Marvell Teralynx chip. This switch supports ROCEv2 and EVPN Multi-homing, with Enterprise SONiC NOS pre-installed. With its cost-effective nature, it stands out as the most affordable and efficient 100G open network switch for data center fabric.</p><h3>32-Port 100G L3 Low Latency Leaf/Spine Switch for Data Center, Enterprise, AI, HPC ,Ceph and NVME</h3><ul><li>10/25G for servers, 40/50/100G uplinks for spine switches.</li><li>The compact 1U switch can be deployed as a leaf or top-of-rack (ToR) switch supporting 10/25GbE to servers with 40/50/100GbE, or as a spine switch supporting 40/50/100GbE spine interconnects.</li><li>Each QSFP28 port can be conﬁgured as 40/100GbE, or as 4x 10/25GbE or 2x 50GbE via breakout cables.</li><li>4+1 hot pluggable built-in fans for superior cooling, reducing noise pollution.</li><li>1+1 hot pluggable PSUs can be removed without shutting down the system.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*fSeZpEm8t7DXwCbn" /></figure><p>Asterfusion CX532P-N</p><p>Astefusion SONiC NOS Support RoCE, BGP-EVPN, VXLAN, MLAG, PFC, ECN, DCBX, INT</p><h3>L2/L3 Features</h3><ul><li>VLAN: 4K VLAN / Access mode / Trunk mode / VLAN-Based MAC learning LAG / LACP / STP / MSTP / 9K Jumbo frame</li><li>LLDP: Neighborhood discovery&amp; aging / Layer2/3 mode / Send &amp;receive / Send / Receive / Custom TLV</li><li>IPv4 /IPv6: Forwarding/ IPv6 ND Proxy</li><li>ARP: Static ARP / Dynamic ARP / Ageing / Renew / Free ARP / ARP Proxy</li><li>Routing: Static routing / BGP / MP-BGP / OSPF Multipath: ECMP / Max support 252 channels / Elastic hash</li><li>DHCP: DHCPv4 / DHCPv6 / DHCPv4/v6 server / DHCPv4/v6 relay</li><li>Multi VRF: 256 VRF</li></ul><h3>High Reliability &amp; Rich QoS</h3><ul><li>System: Container</li><li><a href="https://cloudswit.ch/blogs/what-is-laglacp-mlag/">MC-LAG</a>: Physical link-based cross-chassis LAG</li><li><a href="https://cloudswit.ch/blogs/asterfusion-enterprise-sonic-nos-evpn-multi-homing/">EVPN Multi-homing</a></li><li>BFD: BGP protocol association / OSPF protocol association</li><li>Network quality analysis: SLA / Track / ICMP-echo</li><li>Basic QOS: Multi-Queue scheduling / Flow classification / DSCP mapping queue / Speed limit strategy / Bandwidth guarantee / Tail drop / WRED</li><li>Advancing QOS:PFC / ECN / ETS / DCBXQCN / DCQCN / DCTCP</li><li><a href="https://cloudswit.ch/blogs/roce-rdma-over-converged-ethernet-for-high-efficiency-network-performance/">RoCEv2</a></li></ul><h3>Virtualization</h3><ul><li><a href="https://cloudswit.ch/blogs/why-vxlan-is-necessary-for-cloud-data-centers/">VXLAN</a>: VXLAN Encapsulate / Decapsulate / VTEP / Layer-2 forwarding / Layer-3 gateway / Distributed gateway / ARP inhibition</li><li><a href="https://cloudswit.ch/blogs/asterfusion-sonic-based-evpn-vxlan-solution/">BGP EVPN</a> :VXLAN L2/L3 tunnel automatically established / Dynamic propagation of virtual network routes / Distributed gateway / Symmetric IRB / ARP inhibition / VM migration awareness</li><li>PICFATM: PICFATM VRF mode / IPv6 mode / Distributed routing / VM migration awareness</li></ul><h3>Security &amp; Easy Network Maintenance</h3><ul><li>ACL: ACL for IPv4/v6 / DSCP mapping, marking, classify / IACL / EACL</li><li>Control plane: protection IPv4 CoPP / IPv6 CoPP / Policy speed limited</li><li>AAA: TACACS+</li><li>Management: CLI / WEBUI / Controller/DevOps Ansible / Python/<a href="https://cloudswit.ch/blogs/what-onie-means-to-bare-metal-switch/">ONIE</a> install / ZTP / SONiC upgrade / Warm reboot / Fast reboot / NTP</li><li>Monitoring: SPAN / ERSPAN / SNMP / Syslog / REST API / gRPC / CRM/</li><li>Telemetry: <a href="https://cloudswit.ch/blogs/what-is-int%ef%bc%88in-band-network-telemetry%ef%bc%89/">INT</a></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ze0TNJbDpj3vn-v3" /></figure><h3>01 / Ultra Low Latency Cloud Data Center Switch by using Marvell Teralynx</h3><p>-For latency-sensitive cloud data center applications</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*J9EFt58k3gIW5oxO" /></figure><p>The Asterfusion CX532P-N offers industry-leading</p><h3>02 / Asterfusion SONiC NOS Support RDMA over Converged Ethernet (ROCE)</h3><ul><li>Zero Packet Loss RoCE provides efficient lossless transport data center network</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_23bdCHUFiOK9KhU" /></figure><p>The Asterfusion CX532P-N supports</p><h3>03 / Asterfusion SONiC NOS Support VXLAN &amp; BGP EVPN</h3><p>-Implements a layer-2 virtual network structure dynamically</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1aAqdJVzZUGfLqby" /></figure><p>EVPN-VXLAN expands the potential number of Layer 2 domains and realizes services such as multi-tenant hosting and virtual machine migration.</p><h3>04 / Asterfusion SONiC NOS Supports EVPN-Multihoming</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HPbTPC81iD04EEzj" /></figure><p>AsterNOS 3.1-R0314P06.2, supports EVPN multihoming,adresses several customer challenges. Firstly, EVPN multihoming allows for the connection of a customer site to multiple Provider Edge (PE) devices, ensuring redundancy &amp;high availability through multiple traffic paths. It also facilitates traffic load balancing across multiple PE devices. Moreever, it eliminates the need for complex protocols like STP and VRRP, thereby simplifying network design.</p><h3>05 / Asterfusion SONiC NOS Support In-band Network Telemetry</h3><p>-For intelligent cloud data center network operations</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pU-jfixWWe10dCi4" /></figure><p>The Asterfusion CX532P-N switch supports</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vOIAriVyp4qjpWey" /></figure><h3>Datasheet</h3><p>Please visit <a href="https://help.cloudswit.ch/portal/en/home">https://help.cloudswit.ch/portal/en/home</a></p><h3>Warranty:</h3><p>The Asterfusion CX532P-N switch is backed by a 2 year limited hardware warranty. Multiple extended support options, including advanced replacement and 24×7 support services, are available. Contact us for complete details.</p><p>For more details, please contact: <a href="/service/mailto:lixian@asterfusion.com">lixian@asterfusion.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ccd0a0120691" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Asterfusion Announces its 800G Ethernet Switch for AI/ML/HPC/Data Center]]></title>
            <link>https://medium.com/@lixian_58397/asterfusion-announces-its-800g-ethernet-switch-for-ai-ml-hpc-data-center-8cd465891d2c?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/8cd465891d2c</guid>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Sun, 28 Apr 2024 09:50:59 GMT</pubDate>
            <atom:updated>2024-04-28T09:50:59.463Z</atom:updated>
            <content:encoded><![CDATA[<p>Asterfusion is excited to announce the upcoming release of their revolutionary high-performance 800GbE (51.2T) switch, the CX864E-N. This cutting-edge switch is designed to meet the demanding requirements of artificial intelligence (AI), machine learning (ML), and high-performance computing. By providing high-density and high-performance solutions, Asterfusion aims to cater to the ever-evolving needs of data center networks.</p><p>Set to be released in Q3 2024, the CX864E-N switch boasts 64 powerful OSFP 800GbE ports, delivering a remarkable 51.2 Tbps Ethernet-based switching fabric solution. What sets this switch apart is its sleek and compact 2U form factor, while still offering power and fan tray redundancy for optimum performance and reliability, ensuring next-generation high-performance networking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/694/1*pBpZQe1n3T9KhKBTjceXsA.png" /></figure><p>The outstanding CX864E-N switch is built on the Marvell Teralynx 10 Ethernet switch chip. This remarkable chip is based on 5nm architecture and features an impressive 51.2 Tbps capacity and industry-leading low latency. As a single chip, Teralynx 10 is able to replace four 12.8T switches (equivalent to 32 400GbE switches), resulting in an incredible 80% reduction in power consumption without compromising capacity. In addition, Teralynx 10 supports extensive real-time network telemetry, including P4 in-band network telemetry (INT), enabling predictive analytics, faster problem resolution, and automation. This versatility makes it ideal for a variety of applications, including leaf and spine configurations in next-generation data center networks, as well as AI/ML and high-performance computing (HPC) fabrics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/679/1*OME5F1VDTVu2CgBZ_0oD6w.png" /></figure><p>Asterfusion CX864E-N is optimized for:</p><ul><li>AI/ML and big data analytics</li><li>Hyperscale data centers and cloud computing</li><li>High Performance Computing (HPC)</li><li>Network backbone (800GbE data center leaf/backbone)</li></ul><p>The CX864E-N, as well as Asterfusion’s entire <a href="https://cloudswit.ch/product-category/asterfusion-ultra-low-latency-sonic-based-data-center-switch-by-marvell-teralynx/">CX-N network switch</a> portfolio, supports <a href="https://sonicfoundation.dev/">SONiC Community OS</a> as well as the<a href="https://cloudswit.ch/product/sonic-enterprise-distribution/"> Asterfusion Enterprise SONiC Distribution</a> — AsterNOS. In addition, Asterfusion CX864E-N integrates the Open Network Installation Environment (<a href="https://cloudswit.ch/blogs/what-onie-means-to-bare-metal-switch/">ONIE</a>) installer into the CX864E-N to support other open and third-party network operating systems.</p><h3>Asterfusion 800G Hardware Highlight:</h3><ul><li>OSFP800 switch ports, each supporting 1 x 800 GbE (100G PAM4), or via breakout cables 2 x 400G GbE,4x 200 GbE, or 8 x 100 GbE.</li><li>OSFP800 switch ports also support 1 x 400 GbE (50G PAM4) and via breakout cables 2 x 200 GbE,4x 100 GbE, or 8 x 50 GbE.</li><li>Extra two 10G SFP+ ports for network telemetry management on front panel</li><li>Up to 24 W power budget per OSFP800 port.</li><li>Incorporates Marvell Teralynx10 switch series silicon.</li><li><em>Offering the industry’s lowest single-hop latency.</em></li><li><em>Proven, robust 112G Serdes, the industry’s lowest bit error rate (BER).</em></li><li><em>Comprehensive data center feature set: including IP forwarding, tunneling, rich QoS and robust RDMA.</em></li><li><em>Programmable forwarding: offers permutable flex-forwarding to enable operators to program new packet forwarding protocols as network requirements evolve, with no compromise to throughput, latency or power.</em></li><li><em>Extensive real-time network telemetry, including P4 in-band network telemetry (INT).</em></li><li><em>Advanced shared buffering</em></li><li>2 RU form factor</li><li>SyncE and PTP support</li><li>BMC module with serial-over-LAN support</li><li>Supports hot/cold aisles with front-to-back airflow</li><li>Hot-swappable, load-sharing, redundant 3200 W AC PSUs.</li><li>3+1 hot-swappable fans</li><li>Hardware switch pre-loaded with Open Network Install Environment (ONIE)</li><li>Hardware switch pre-load with Asterfusion Enterprise SONiC Distribution (AsterNOS)</li></ul><p>Overall, the reveal of the CX864E-N switch from Asterfusion showcases their commitment to innovation and addressing the ever-expanding demands of the data center network landscape. The enhanced performance, efficiency, and advanced features of the CX864E-N make it a valuable asset for industries reliant on AI, ML, and HPC.</p><p>For more details. please contact email: lixian@asterfusion.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8cd465891d2c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is PTP and How does it Work?]]></title>
            <link>https://medium.com/@lixian_58397/what-is-ptp-and-how-does-it-work-06558649db11?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/06558649db11</guid>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Fri, 29 Mar 2024 09:09:13 GMT</pubDate>
            <atom:updated>2024-03-29T09:09:13.931Z</atom:updated>
            <content:encoded><![CDATA[<h3>Table of Contents</h3><ul><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#why-need-ptp%EF%BC%88-precision-time-protocol">Why Need PTP（Precision Time Protocol)?</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#what-is-ptp-precision-time-protocol">What is PTP (Precision Time Protocol)?</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#how-does-pt-pv-2-work">How does PTPv2 Work?</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#ptp-related-basic-concepts">PTP Related Basic Concepts</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#ptp-domain">PTP Domain</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#ptp-port">PTP Port</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#clock-nodes">Clock Nodes</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#grandmaster-clock">Grandmaster Clock</a></li><li><a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/#asterfusion-so-ni-c-based-open-network-switches-with-ptp-support">Asterfusion SONiC-based Open Network Switches with PTP Support</a></li></ul><p>This article provides a comprehensive overview of the PTP protocol and explains how the Asterfusion SONiC-based open network switch supports PTP in both hardware and software.</p><h3>Why Need PTP（Precision Time Protocol)?</h3><p>Ethernet technology is widely used in telecom-grade networks due to its openness, affordability, and ease of use. Over the years, data transmission speeds have significantly increased from 10 Mbps to 800 Gbps. One of the key advantages of Ethernet is its “plug-and-play” feature, allowing Ethernet terminals to effortlessly connect to an IP network and access services at any time. However, for a truly reliable telecommunications-grade network, synchronization of the IP network is crucial to ensure quality for real-time and data services. Currently, the synchronization requirements for telecommunication-level networks are very strict. The backbone network delay should ideally be controlled within 50ms. Unfortunately, the existing <a href="https://cloudswit.ch/blogs/what-is-ntp-ptp-sntp/">NTP</a> (Network Time Protocol) and <a href="https://cloudswit.ch/blogs/what-is-ntp-ptp-sntp/">SNTP</a> (Simple Network Time Protocol) protocols fail to achieve the desired synchronization accuracy or convergence speed.</p><h3>What is PTP (Precision Time Protocol)?</h3><p>The IEEE1588 PTP protocol utilizes <a href="https://cloudswit.ch/blogs/what-is-ntp-ptp-sntp/">NTP</a> technology and is known for its simple configuration, quick convergence, and minimal consumption of network bandwidth and resources. The IEEE1588 standard, also known as the IEEE1588 Precision Clock Synchronization Protocol or PTP (Precision Timing Protocol), aims to periodically correct and synchronize the clocks of all network nodes through a synchronization signal. This synchronization enables precise synchronization in Ethernet-based distributed systems and can be applied to any multicast network. The IEEE1588 protocol has evolved to version 2, which features bi-directional channels, nanosecond-level precision, cost-effectiveness, and adaptability to various access environments.</p><h3>How does PTPv2 Work?</h3><p>IEEE1588 PTPv2 uses a relative time synchronization mechanism. One participant is selected as the master time clock and it will send synchronization messages to the slaves. The master will send synchronization messages to the network and all the slaves calculate the time delay.</p><p>The principle of time synchronization is shown below</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/519/1*UpRdPzoudrX_2G65-K-DSg.png" /></figure><p>The time offset (Offset) and transmission delay (Delay) are calculated from the clock based on T1, T2, T3, T4, i.e.:</p><p>T2 — T1 = Offset + Delay</p><p>T4 — T3 = Delay — Offset</p><p>From this we can calculate:</p><p>Delay = (T4 — T3 + T2 — T1)/2</p><p>Offset = (T2 — T1 — T4 + T3)/2</p><p>Depending on the Offset, the slave can adjust its own clock for precise time synchronization.</p><p>The sophisticated network synchronization protocol defined by the IEEE1588 PTPv2 standard achieves a high degree of synchronization in the network. No special synchronization communication is required when allocating control tasks, thereby achieving the effect of separating communication time patterns from application execution time patterns. Due to the highly precise synchronization operation, data transmission time fluctuations inherent in Ethernet technology are reduced to an acceptable range without affecting control accuracy.</p><h3>PTP Related Basic Concepts</h3><h3>PTP Domain</h3><p>A network that applies the PTP protocol is called a PTP domain.A PTP domain has one and only one synchronized clock, and all devices within the domain are synchronized to that clock.</p><h3>PTP Port</h3><p>A port on a device that runs the PTP protocol is called a PTP port. As shown in Figure 2, there are three roles of PTP ports:</p><ul><li>Master Port: The port that publishes the synchronized time, which can exist on BC or OC;</li><li>Slave Port: the port that receives the synchronization time, which can exist on the BC or OC;</li><li>Passive Port: a port that neither receives nor publishes synchronization time, and exists on the BC only</li></ul><h3>Clock Nodes</h3><ul><li>OC (Ordinary Clock): The clock of only one PTP communication port is an ordinary clock.</li><li>BC (Boundary Clock): A clock with more than one PTP communication port.</li><li>TC (Transparent Clock): In contrast to BC/OC, which requires time synchronization with other clock nodes, TC does not synchronize with other clock nodes. TC has multiple PTP ports, but it only forwards PTP protocol messages between these ports and corrects them for forwarding delays, and does not synchronize the time through any of the ports.</li></ul><h3>Grandmaster Clock</h3><p>As shown in Figure 2, all clock nodes in the PTP domain are organized according to a certain hierarchical structure. The reference time of the entire domain is the Grandmaster Clock, which is the highest-level clock.</p><p>Grandmaster Clock is the highest level clock, and its time will eventually be synchronized to the entire PTP domain through the interaction of PTP protocol messages between clock nodes, so it is also called a clock source. As shown in Figure 2, this is a typical schematic diagram of the deployment relationship between the master clock and the slave clock.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*Q52v4pOOMdLK5kpehC1wfQ.png" /></figure><h3>Asterfusion SONiC-based Open Network Switches with PTP Support</h3><p>Currently, the Asterfusion<a href="https://cloudswit.ch/asterfusion-1g-to-400g-bare-metal-switches-product/"> CX series open network switch</a> products offer support for the IEEE1588 PTPv2 standard through both hardware and software. When implemented using hardware, which means the time stamp function is carried out through hardware, we can achieve accuracy up to the nanosecond level. Similarly, when implemented using software, we can usually achieve nanosecond-level accuracy as well.</p><p>In terms of hardware, we have integrated a high-precision clock module into the machine to facilitate IEEE1588 and Synchronous Ethernet (SyncE) implementation. The hardware port offers full support for PTP and can be used as boundary clock（BC) or Ttransparent clock (TC). Furthermore, the module can be flexibly disassembled based on customer requirements.</p><p>On the software side, Asterfusion leads the community in implementing the PTP function in <a href="https://sonicfoundation.dev/">SONiC</a> and optimizing its performance on our <a href="https://cloudswit.ch/product/sonic-enterprise-distribution/">Enterprise SONiC Distribution</a> –<a href="https://cloudswit.ch/product/sonic-enterprise-distribution/">AsterNOS</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/896/1*SpjsU6Dnky2NVYrJV79yiQ.png" /></figure><p>For more details, Please visit: <a href="https://cloudswit.ch/blogs/what-is-ptp-and-how-does-it-work/">What Is PTP And How Does It Work? — Asterfusion Data Technologies (cloudswit.ch)</a></p><p>Or contact email: lixian@asterfusion.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=06558649db11" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Comparative Analysis of InfiniBand and RoCE for AI Data Centers]]></title>
            <link>https://medium.com/@lixian_58397/a-comparative-analysis-of-infiniband-and-roce-for-ai-data-centers-78a99c39881c?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/78a99c39881c</guid>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Tue, 06 Feb 2024 08:31:37 GMT</pubDate>
            <atom:updated>2024-02-06T08:31:37.405Z</atom:updated>
            <content:encoded><![CDATA[<h3>Table of Contents</h3><ul><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#understanding-of-infini-band">Understanding of InfiniBand</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#what-is-infini-band">What is InfiniBand?</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#infini-bands-network-architecture">InfiniBand’s network architecture</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#infini-band-has-the-following-advantages">InfiniBand has the following advantages</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#which-manufacturers-on-the-market-support-infini-band">Which manufacturers on the market support InfiniBand?</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#infini-band-s-shortcomings-and-limitations">InfiniBand‘s Shortcomings and Limitations</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#but-what-is-better-than-infini-band">But What is better than InfiniBand?</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#what-is-roce">What is ROCE?</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#for-more-about-roce-i-warp-infini-band-please-read">For more about ROCE /iWARP/InfiniBand</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#advantages-of-ro-c-ev-2">Advantages of RoCEv2</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#comparison-of-infini-band-and-ro-ce-v-2">Comparison of InfiniBand and RoCE v2</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#nvidia-infini-band-vs-asterfusion-ro-c-ev-2-switchs-onsite-test-data">NVIDIA InfiniBand Vs Asterfusion RoCEv2 switch’s Onsite Test Data</a></li><li><a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/#one-click-configuration-of-asterfusion-roce-ready-switch-network-technology-white-paper">One-click configuration of Asterfusion ROCE-ready switch network technology white paper</a></li></ul><p>In the increasingly fierce competition between InfiniBand and Ethernet for market dominance in AI backend networks, manufacturers are battling to provide a high-performance network infrastructure. This is crucial to support the operation of AIGC. To meet the demands of AI cluster computing, the industry has put forward two main network solutions: InfiniBand and RDMA. In this comparison, we will delve into these two technologies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/926/1*VLQfB7IlYRi1PGmmVAYnmQ.png" /></figure><h3>Understanding of InfiniBand</h3><h3>What is InfiniBand?</h3><p>InfiniBand (abbreviated as IB), is a communication standard for high-performance computing that offers remarkable throughput and minimal latency for computer-to-computer data interconnections. It is used both as a direct or switched interconnect between servers and storage systems, as well as an interconnect between storage systems. With the exponential growth of artificial intelligence, InfiniBand has become the preferred network interconnect technology for GPU servers.</p><h3>InfiniBand’s network architecture</h3><p>InfiniBand is a communications link for data flow between processors and I/O devices, supporting up to 64,000 addressable devices. InfiniBand Architecture (IBA) is an industry-standard specification that defines a point-to-point switching input/output framework for interconnecting servers, communications infrastructure, storage devices, and embedded systems.</p><p>InfiniBand is ubiquitous, low-latency, high-bandwidth, and low-management cost, and is ideally suited for connecting multiple data streams (clustering, communications, storage, management) in a single connection with thousands of interconnected nodes. The smallest complete IBA unit is a subnet, and multiple subnets are connected through routers to form a large IBA network.</p><p>An InfiniBand system consists of channel adapters, switches, routers, cables, and connectors. the CAs are categorized as host channel adapters (HCAs) and target channel adapters (TCAs). IBA switches are similar in principle to other standard network switches but must meet the high performance and low cost requirements of InfiniBand. The HCA is the device point at which IB end nodes, such as servers or storage devices, connect to the IB network. the TCA is a special form of channel adapter used primarily in embedded environments, such as storage devices.</p><p>The HCA is the device point where IB end nodes (such as servers or storage devices) connect to the IB network. TCA is a special form of channel adapter used primarily in embedded environments such as storage devices.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/595/1*4E7KgfqDVCRoPXSiy21N3g.png" /></figure><h3>InfiniBand has the following advantages</h3><ol><li>Low Latency: InfiniBand networks are known for their extremely low latency. RDMA zero-copy networks reduce operating system overhead, allowing data to move quickly through the network.</li><li>High Bandwidth: InfiniBand networks offer high bandwidth data transfer capabilities. It usually supports tens of Gb/s or even higher bandwidth, high bandwidth makes it possible to exchange data between nodes at high speed.</li><li>High throughput: Due to the characteristics of low latency and high bandwidth, IB networks are able to realize high throughput data transmission. It supports parallel transmission of large-scale data streams, while reducing intermediate processing and copy operations</li><li>Scalability: It supports multi-level topologies, such as global interconnection networks, tree structure and flat structure, which can be flexibly configured and expanded according to application requirements and scale.</li></ol><h3>Which manufacturers on the market support InfiniBand?</h3><ul><li><a href="https://en.wikipedia.org/wiki/Mellanox_Technologies">Mellanox</a>: Mellanox is basically the only manufacturer of IB switches in the world. NVIDIA successfully acquired it in 2020 and now also offers NVIDIA Quantum InfiniBand switches.</li><li>HP, DELL, etc.: These companies are mainly OEMs, which means they do not directly produce IB switches, but purchase or obtain authorization from manufacturers such as Mellanox, and then conduct re-sells.</li></ul><h3>InfiniBand‘s shortcomings and limitations:</h3><p>Since InfiniBand (IB) boasts numerous ultimate advantages, it is intriguing to wonder why the industry is increasingly expressing hopes for Ethernet to become the preferred technology for artificial intelligence workloads amidst the explosive development of artificial intelligence. As time goes by, it has been revealed that the industry has also discovered several fatal shortcomings of InfiniBand:</p><ol><li>Very High Cost: The main disadvantage of InfiniBand is its high price compared to Ethernet. Using InfiniBand to build a private network requires designated IB network cards and switches, the cost of which is generally five to ten times that of ordinary network equipment. As a result, InfiniBand is only seriously considered for implementation in high-end industries such as finance and futures trading.</li><li><strong>Elevated O&amp;M Expenses</strong>: InfiniBand, being a dedicated network technology, cannot leverage the user’s existing IP network infrastructure and accumulated knowledge in terms of operation and maintenance. Enterprises opting for InfiniBand must hire specialized personnel to handle the specific operation and maintenance requirements. However, due to the limited market space occupied by InfiniBand (less than 1% of Ethernet), there is a severe shortage of experienced operation and maintenance personnel in the industry. Network failures may go unrepaired for extended periods of time, leading to exorbitant operational expenses (OPEX).</li><li>Vendor lock-in: The presence of vendor lock-in is a major disadvantage of InfiniBand switches. These switches are manufactured by specific vendors and use proprietary protocols, which makes it difficult for them to communicate with other network devices using Ethernet over IP. This closed architecture also creates risks for businesses that need scalable extensions in the future, as being tied to a single vendor can lead to uncontrollable risks.</li><li><strong>Long lead time</strong>: The long lead time of InfiniBand switches is another concern. Their arrival is unpredictable, which increases project risk and hampers business expansion.</li><li><strong>Slow upgrade</strong>: The upgrade process of the network depends on individual vendors’ product releases, preventing unified upgrades across the industry. This slowness in upgrades can hinder the overall progress of the network.</li></ol><h3>But What is better than InfiniBand?</h3><p>Compared with InfiniBand, RoCE provides greater versatility and better cost performance. Not only can it build high-performance RDMA networks, it is also compatible with traditional Ethernet.</p><h3>What is ROCE?</h3><p>RoCE (RDMA over Converged Ethernet) is a cutting-edge network protocol that helps transfer memory data directly between computers without the involvement of the CPU or operating system. This not only frees up memory bandwidth and CPU cycles, but also enhances system performance, enabling faster communication between nodes, lower latency, and higher throughput. RDMA technology is currently widely used in various high-performance scenarios such as supercomputing, AI training, and storage.</p><p>For more details. please visit: <a href="https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/">https://cloudswit.ch/blogs/how-to-choose-from-infiniband-and-roce-for-aigc/</a></p><p>Or email: lixian@asterfusion.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=78a99c39881c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[On-site Test Result: Asterfusion RoCEv2-enabled SONiC Switches Vs. InfiniBand]]></title>
            <link>https://medium.com/@lixian_58397/on-site-test-result-asterfusion-rocev2-enabled-sonic-switches-vs-infiniband-d58b1f3e29c1?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/d58b1f3e29c1</guid>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Fri, 26 Jan 2024 08:42:29 GMT</pubDate>
            <atom:updated>2024-01-26T08:42:29.061Z</atom:updated>
            <content:encoded><![CDATA[<h3>Table of Contents</h3><ul><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#what-is-infini-band%EF%BC%9F">What is InfiniBand？</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#challenges-of-infini-band-network-deployments">Challenges of InfiniBand Network Deployments</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#what-is-ro-ce-and-ro-c-ev-2">What is RoCE and RoCEv2?</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#ro-c-ev-2-switches-test-result-in-aigc-network">RoCEv2 Switches Test Result in AIGC Network</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#nccl-test%EF%BC%88-bandwidth-latency%EF%BC%89">NCCL-Test（Bandwidth &amp; Latency）</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#ro-c-ev-2-switches-test-result-in-hpc-network">RoCEv2 Switches Test Result in HPC Network</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#ro-c-ev-2-switches-test-result-in-distributed-storage-network">RoCEv2 Switches Test Result in Distributed Storage Network</a></li><li><a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/#conclusion-asterfusion-ro-c-ev-2-low-latency-so-ni-c-switches-are-fully-capable-of-replacing-ib-switches">Conclusion: Asterfusion RoCEv2 low latency SONiC switches are fully capable of replacing IB switches</a></li></ul><p>In early deployments of AI and ML, NVIDIA’s InfiniBand networking solution was the preferred choice because of its excellent low <a href="https://cloudswit.ch/blogs/how-to-build-ultra-low-latency-data-center-networks/">network latency</a>. Over time, however, networking vendors and hyperscalers are actively exploring Ethernet alternatives to cater to the demands of rapidly expanding AI workloads. Currently, RoCEv2 replacing InfiniBand to carry RDMA traffic has become a common solution in the industry. Let’s explore the following article！</p><h3>What is InfiniBand？</h3><p>InfiniBand is a network communication protocol that facilitates the movement of data and messages by creating a dedicated, protected channel directly between nodes through a switch. RDMA and send/receive offloads managed and performed by InfiniBand adapters. One end of the adapter connects to the CPU via PCIe, and the other end connects to the InfiniBand subnet via an InfiniBand network port. This provides significant advantages over other network communication protocols, including higher bandwidth, lower latency, and enhanced scalability.</p><h3>Challenges of InfiniBand Network Deployments</h3><ul><li><strong>Vendor lock-in: </strong>Only one vendor has mature IB products &amp; solutions, which are expensive.</li><li><strong>Compatibility: </strong>InfiniBand uses a unique protocol, not TCP/IP; Specialized network, separate cabling and switches</li><li><strong>Availability:</strong> IB switches generally have a long delivery time</li><li><strong>Service: </strong>O&amp;M depends on the manufacturer, which makes it difficult to locate faults and takes a long time to solve problems</li><li><strong>Expansion and upgrading: </strong>Depends on the progress of the vendor’s product releases</li></ul><h3>What is RoCE and RoCEv2?</h3><p>Among these alternatives, <a href="https://cloudswit.ch/blogs/roce-rdma-over-converged-ethernet-for-high-efficiency-network-performance/#what-is-ro-ce%EF%BC%9F">RDMA</a> (Remote Direct Memory Access) emerges as a groundbreaking technology that enhances data throughput and reduces latency significantly. It initially deployed on the InfiniBand, but has now been successfully integrated into Ethernet infrastructure.</p><p>Currently, the prevailing solution for high-performance networks involves constructing an RDMA-supported network based on the RoCEv2 (RDMA over Converged Ethernet version 2) protocol. This innovative approach makes use of two crucial technologies, Priority Flow Control (PFC) and Explicit Congestion Notification (ECN). The utilization of this technology offers several advantages, including cost-effectiveness, impeccable scalability, and the elimination of vendor lock-in.</p><p><em>For more about ROCE please read:</em> <a href="https://cloudswit.ch/blogs/roce-rdma-over-converged-ethernet-for-high-efficiency-network-performance/#what-is-ro-ce%EF%BC%9F">https://cloudswit.ch/blogs/roce-rdma-over-converged-ethernet-for-high-efficiency-network-performance/#what-is-ro-ce%EF%BC%9F</a></p><p>Next, we’ll take a look at how Asterfusion ROCEv2 low latency SONiC switches perform against InfiniBand switches or other Ethernet switches in AIGC, distributed storage and HPC network.</p><h3>RoCEv2 Switches Test Result in AIGC Network</h3><p>We built an <a href="https://cloudswit.ch/blogs/optimize-aigc-network-for-llm-training/">AIGC network</a> test environment using the <a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/"><strong>Asterfusion CX664D-N</strong></a> Ultra Low Latency Switch and another brand RoCEv2 supported switch（Switch A） to compare NCCL performance.</p><p>NCCL is short for NVIDIA Collective Communications Library. It is an open source tool from NVIDIA for testing collection communication. It can be used to test whether the collection communication is normal, and to stress test the collection communication rate.</p><p><a href="https://github.com/NVIDIA/nccl-tests">https://github.com/NVIDIA/nccl-tests</a></p><p><a href="https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/overview.html">Overview of NCCL - NCCL 2.32.3 documentation</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/977/1*Bxfm25P3-t6GnOmSRNHkKA.png" /></figure><p><strong>ItemModelSpecsQuantityRemarkSwitch</strong>AsterfusionCX664D-N (64-port 200G)<br>Switch A (48-port 200G,as a test control)\11<strong>Server</strong>X86_64CPU: Intel Xeon Silver 4214Memory: 256G2Need 200G NIC and corresponding driver<strong>Optical Module</strong>100GQSFP284\<strong>Fiber</strong>Multi-mode\2\<strong>NIC</strong>\MT28800 Family CX-5 Lx2Switch to Ethernet Mode<strong>OS</strong>Ubuntu20.04\\Kernel version &gt; 5.10<strong>Mellanox Driver</strong>MLNX_OFED-5.0\\NIC driver adapted to host kernel<strong>Switch OS</strong>AsterNOS SONiC 201911.R0314P02\\\<strong>NCCL</strong>V2.13\\\<strong>OpenMPI</strong>V4.0.3\\\</p><h3>NCCL-Test（Bandwidth &amp; Latency）</h3><p>Layer 2 network, RoCEv2 enabled on CX664P-N Switch, GPU server NICs optimized for RDMA.</p><p><strong>Bandwidth （GB/s）GPU0GPU1GPU2GPU3GPU4GPU5GPU6GPU7</strong>Switch A GPU1–25.396.6111.317.222.012.451.772.28Switch A GPU3–45.977.1711.217.762.42.812.052.66Asterfusion GPU1–26.027.2211.247.82.392.851.872.66</p><p><strong>Latency (μs)GPU0GPU1GPU2GPU3GPU4GPU5GPU6GPU7</strong>Switch A GPU1–222.0421.8121.8422.3122.9322.4222.922.75Switch A GPU3–421.6321.6521.9821.7122.4622.5123.123.05Asterfusion GPU1–222.5721.7921.7821.6222.7922.6323.0123.16</p><p><strong>Mellanox CX-5<br>(Direct connection)CX664D-NSwitch ALatency (ns)</strong>1400ns480ns580ns<strong>Bandwidth (Gb/s)</strong>99.498.1498.14</p><h3><a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/">64-Port 200G QSFP56 Low Latency Data Center Switch, Enterprise SONiC Ready, Marvell Teralynx: CX664P-N</a></h3><p>More about Asterfusion AIGC Networking Solutions: <a href="https://cloudswit.ch/blogs/optimize-aigc-network-for-llm-training/">Cost Effective AIGC Network Solution by Asterfusion ROCEv2 Ready Switch</a></p><p>For more please visit <a href="https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/">https://cloudswit.ch/blogs/sonic-switches-rocev2-test-in-aigc-storage-hpc/</a></p><p>Or contact email: lixian@asterfusion.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d58b1f3e29c1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#MWC24]]></title>
            <link>https://medium.com/@lixian_58397/mwc24-a8e8f5d312b1?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/a8e8f5d312b1</guid>
            <category><![CDATA[mwc]]></category>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Tue, 19 Dec 2023 10:18:52 GMT</pubDate>
            <atom:updated>2023-12-19T10:18:52.486Z</atom:updated>
            <content:encoded><![CDATA[<p>Headed to <a href="https://twitter.com/hashtag/MWC24?src=hashtag_click">#MWC24</a> this year? Asterfusion’s experts are ready to discuss all your network needs, Starting to Explore Our Full Series Open Network Switches from Data Center to Campus. Please contact me directly to schedule a meeting with our experts.</p><p>Email: lixian@asterfusion.com</p><p>TG: <a href="https://t.me/Candy5347">https://t.me/Candy5347</a></p><p>#MWCBarcelona</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a8e8f5d312b1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Asterfusion’s Distributed Storage Network Solution Boosts Public Cloud for World’s Leading ISP]]></title>
            <link>https://medium.com/@lixian_58397/asterfusions-distributed-storage-network-solution-boosts-public-cloud-for-world-s-leading-isp-df5ab4997b03?source=rss-c8413ba2a07e------2</link>
            <guid isPermaLink="false">https://medium.com/p/df5ab4997b03</guid>
            <category><![CDATA[open-networking]]></category>
            <category><![CDATA[data-center]]></category>
            <category><![CDATA[open-source]]></category>
            <dc:creator><![CDATA[Candy798]]></dc:creator>
            <pubDate>Fri, 15 Dec 2023 09:18:13 GMT</pubDate>
            <atom:updated>2023-12-15T09:18:13.324Z</atom:updated>
            <content:encoded><![CDATA[<h3>Table of Contents</h3><ul><li><a href="https://cloudswit.ch/blogs/lossless-storage-network-for-public-cloud/#one-model-of-200-g-whitebox-switch-builds-a-large-scale-storage-network">One Model of 200G Whitebox Switch Builds a Large-Scale Storage Network</a></li><li><a href="https://cloudswit.ch/blogs/lossless-storage-network-for-public-cloud/#what-is-rdma-why-packet-loss-rate-is-essential">What is RDMA &amp; Why Packet Loss Rate is Essential?</a></li><li><a href="https://cloudswit.ch/blogs/lossless-storage-network-for-public-cloud/#what-is-nv-me-o-f-in-storage-networks">What is NVMe-oF in Storage Networks?</a></li><li><a href="https://cloudswit.ch/blogs/lossless-storage-network-for-public-cloud/#build-lossless-data-center-fabric-with-easy-ro-ce-configuration">Build Lossless Data Center Fabric with “Easy RoCE Configuration”</a></li></ul><p>With the acceleration of digitization, the size of the cloud service market has been increasing in recent years. Big Data, NoSQL, AIGC, as well as Internet gaming, 3D rendering…all of the above require better IOPS and network latency performance in storage network.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/450/1*80I7frAG5sP-NWfDr_XUwA.png" /></figure><p><a href="https://www.chinatelecomglobal.com/"><strong>China Telecom</strong></a> is a leading global ISP with more than 2 million public cloud customers. In the near future, Asterfusion low-latency Ethernet switches（CX664D-N）will be deployed at scale to provide a resilient and reliable network.</p><h4><a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/">64-Port 200G QSFP56 Low Latency Data Center Switch, Enterprise SONiC Ready, Marvell Teralynx: CX664P-N</a></h4><h3>One Model of <a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/">200G </a><a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/">Whitebox</a><a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/"> Switch</a> Builds a Large-Scale Storage Network</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/929/1*eSwRCrysnAY0vxtDlxKx-g.png" /></figure><p>Based on standard CLOS, Asterfusion’s distributed storage network solution shortens the communication paths between servers, reducing latency and improving application and service performance. Besides, simple network architecture is easy for O&amp;M.</p><p>The <a href="https://cloudswit.ch/product/64-port-200g-data-center-switch-enterprise-sonic-teralynx/">Asterfusion CX664P-N is data center switch</a> designed for leaf/spine architecture with Enterprise SONiC pre-installed. It provides high density 200G/100G compatible ports for future network expansion, as well as two 10G ports for extended access.</p><p>CX664P-N offers low latency and boasts a line-rate L2/L3 up to 12.8Tbps switching performance with Marvell Teralynx ASIC, supports <a href="https://cloudswit.ch/blogs/roce-rdma-over-converged-ethernet-for-high-efficiency-network-performance/"><strong>ROCEv2</strong></a> and <a href="https://cloudswit.ch/blogs/asterfusion-enterprise-sonic-nos-evpn-multi-homing/"><strong>EVPN Multi-homing</strong></a>, and it stands out as an affordable and efficient open network switch for data center fabric.</p><h3>What is RDMA &amp; Why Packet Loss Rate is Essential?</h3><p>Remote Direct Memory Access (RDMA) allows data to be transferred without the intervention of the CPU cache, which is especially suitable for SAN and other large-scale computing cluster networks like AI/ML/HPC. It is estimated that using RDMA instead of TCP/IP increases the I/O speed of SSD storage by about 50 times.</p><p>However, network packet loss has a huge impact on data transfer performance as shown in the <a href="https://www.semanticscholar.org/paper/Congestion-Control-for-Large-Scale-RDMA-Deployments-Zhu-Eran/07367703f587dbc3313cc613289c4330cebe5c8c">figure</a>: 0.1% packet loss will result in a sharp drop in RDMA throughput; 2% packet loss will cause throughput to drop to zero.</p><p>Since InfiniBand networks are very expensive, it is logical for vendors to consider less costly ways to transport RDMA traffic. RoCEv2 replacing InfiniBand with IPv4/IPv6 headers and UDP headers on the Ethernet link layer. IP and UDP improve network scalability, support networking across three layers, and support load balancing, but RoCEv2 still requires lossless networking for optimal performance.</p><h3>What is NVMe-oF in Storage Networks?</h3><p><a href="https://nvmexpress.org/">NVMe (Non-Volatile Memory Express)</a> is a protocol designed specifically for accessing storage devices, such as solid-state drives (SSDs), over PCIe. NVMe offers significantly faster data transfer speeds and lower latency compared to traditional storage protocols like SATA or SAS.</p><p>When NVMe and RoCEv2 are combined, they provide a highly performant and scalable solution for storage networks. This combination allows for direct access to NVMe storage devices, bypassing traditional storage protocols and network stack layers. As a result, it significantly reduces latency and improves throughput, enabling faster and more efficient data transfers.</p><h3>Build Lossless Data Center Fabric with “Easy RoCE Configuration”</h3><p>Equipped with the <a href="https://cloudswit.ch/product/sonic-enterprise-distribution/">powerful SONiC distribution(AsterNOS)</a>, Asterfusion CX-N series switches support DCB protocol groups such as PFC, PFC WatchDog, ECN to reduce network packet loss and congestion. With a forwarding delay of at least 400ns per port, they ensure efficient data transmission. (Check out <a href="https://cloudswit.ch/wp-content/uploads/2023/04/6-741x1024.png">Test Results </a>on distributed storage network between Asterfusion’s low latency switch &amp; InfiniBand switch.)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/901/1*fIU8g9EUrMHZwzp9xgQG0w.png" /></figure><p>Typically, O&amp;M engineers need to modify the PFC and ECN separately to configure or cancel the lossless network. For reducing the complexity of O&amp;M, AsterNOS provides encapsulated set of commands-“<strong>Easy RoCE Configuration”.</strong></p><p>Not only in distributed storage networks, Asterfusion’s low-latency data center switches have also been deployed at scale in AIGC, HPC and other scenarios.</p><p>Please visit cloudswit.ch and mail to lixian@asterfusion.com for more.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=df5ab4997b03" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>