<?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[Beeks Technology Blog - Medium]]></title>
        <description><![CDATA[Learn more about how Beeks solves challenging performance problems for the financial markets - Medium]]></description>
        <link>https://medium.com/beekstech?source=rss----7146a35a7721---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Beeks Technology Blog - Medium</title>
            <link>https://medium.com/beekstech?source=rss----7146a35a7721---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 20:32:52 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/beekstech" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Relative location of Packet Loss: The Theory Behind the Practice]]></title>
            <link>https://medium.com/beekstech/relative-location-of-packet-loss-the-theory-behind-the-practice-a4f52d77d548?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/a4f52d77d548</guid>
            <category><![CDATA[wireshark-tutorial]]></category>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[packet-capture]]></category>
            <category><![CDATA[wireshark]]></category>
            <category><![CDATA[packet-analysis]]></category>
            <dc:creator><![CDATA[Matt Cretney]]></dc:creator>
            <pubDate>Wed, 13 May 2026 10:28:22 GMT</pubDate>
            <atom:updated>2026-05-13T10:28:22.167Z</atom:updated>
            <content:encoded><![CDATA[<p>In our previous article, we walked through how to quickly spot TCP loss patterns in Wireshark — and how to tell whether the loss is happening <strong>before or after</strong> your capture point. That guide gave you <strong>a practical shortcut</strong> to get answers fast, using built-in Wireshark flags like tcp.analysis.retransmission, tcp.analysis.out_of_order, and tcp.analysis.duplicate_ack.</p><p>But if you’ve ever presented packet loss evidence to someone who swears their network is perfect — often backed by the claim that “there’s nothing in the logs” — you’ll know that shortcuts alone won’t always cut it.</p><p>This follow-up post explains <strong>why</strong> those interpretations are valid, and gives you the technical foundation to back them up when challenged. Knowing the TCP-level logic behind these flags means you won’t just detect problems — you’ll be able to <strong>defend your conclusions</strong> with confidence.</p><p>We’ll walk through simple examples to explain what each type of TCP behaviour actually means on the wire, and why it points to loss on one side of your capture point or the other. You will almost certainly encounter more complicated scenarios than this — selective acknowledgments (SACK) in particular — but if you’ve got a grounding in this simple example you will have a firm foundation for understanding anything more complicated.</p><p>Let’s get into it.</p><h3>Our Setup: Bob, Alice, Charlie, and Debbie</h3><p>To keep things simple, we’ll use a one-way TCP example:</p><ul><li><strong>Bob</strong> is sending data to <strong>Alice</strong></li><li>Somewhere between them, some packets are being lost</li><li><strong>Charlie</strong> and <strong>Debbie</strong> are two observers, positioned at different points along the network path</li></ul><p>In reality, you have to imagine that at the same time Alice will also be sending data to Bob. I’ve left that complexity out of the diagrams because if you can understand the simple situation, you’ll find it much easier to understand more complex cases. This simplification lets us explore how loss looks <strong>from different sides of the failure</strong>.</p><p>We’ve left out the full sequence and ACK numbers — you’ll see simplified references like DATA 1, DATA 2, and ACK 1, ACK 2 for clarity.</p><p>Let’s get into the meat of it…</p><h3>Example 1 — simple retransmission</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZT7I-bczoKAI8Pw7JiLoqA.png" /><figcaption>Example 1 — Bob is sending data to Alice over a TCP connection. Charlie and Debbie are observers in the middle. The loss occurs between Charlie and Debbie.</figcaption></figure><p>Note the behaviour of Bob’s retransmission timer in the example below. If Bob doesn’t see ACK2 within his retransmission timer window, he will assume it wasn’t delivered and will retransmit DATA2.</p><p>Charlie will see DATA2 as a retransmission (because he saw the original packet). Charlie knows that the loss is occurring <strong>closer to the destination IP address of the retransmitted packet (i.e. there was no loss between him and Bob)</strong>. Debbie will not have seen the retransmission at all, so in this specific situation she wouldn’t necessarily have any idea that the loss would have occurred.</p><h3>Example 2: Fast Retransmit Triggered by Duplicate ACKs</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rk0in5TRk7cAaPmzIR2jVA.png" /><figcaption>Example 2 — Bob is sending data to Alice. Charlie and Debbie are observers in the middle. The loss occurs between Charlie and Debbie but Charlie continues to send data after the loss, causing Alice to respond with duplicate acknowledgments.</figcaption></figure><p>Now suppose Bob sends DATA 2, which is lost. Unlike in the previous example, his retransmission timer (to receive the ACK back from Alice) never runs out, so he keeps sending DATA 3, DATA 4, etc. Each time Alice gets one of these out-of-sequence packets, she sends a <strong>duplicate ACK</strong> for ACK 2.</p><h4>Why is Alice sending duplicate acks?</h4><p>Because TCP is a <strong>reliable</strong>, <strong>ordered</strong> protocol, Alice can’t process DATA 3 until she gets DATA 2.</p><p>After seeing three duplicate ACKs, Bob triggers a <strong>retransmission</strong> of DATA 2.</p><p>So what does she do?</p><p>Every time she gets a new, out-of-order packet (DATA 3, DATA 4, etc.), she sends an <strong>ACK for the last packet she received in order</strong> — in this case, still ACK 2.</p><p>These repeated ACKs — all saying “I’m still waiting for DATA 2” — are what we call <strong>duplicate ACKs</strong>.</p><p>They’re her way of politely but firmly telling Bob:</p><blockquote><em>“I didn’t get what I needed. Still waiting for </em><em>DATA 2. Still waiting. Still waiting…”</em></blockquote><p>After Bob receives three of these duplicate ACKs, he assumes DATA 2 must have been lost, and <strong>retransmits it immediately</strong> without waiting for the normal timeout — this is called <strong>fast retransmit</strong>.</p><h4>What do the observers see for Example 2, and what does it tell us about how we should understand retransmissions and duplicate acknowledgments in a packet capture?</h4><p><strong>Charlie’s view:</strong></p><ul><li>Sees the original DATA 2</li><li>Sees the retransmission</li><li>Sees duplicate ACKs from Debbie — these strongly indicate that some of Bob’s later data reached Debbie, but she is missing earlier data.</li><li>Can again infer that <strong>loss is downstream</strong> of him — closer to the destination IP</li></ul><p><strong>Debbie’s view:</strong></p><ul><li>Sees no retransmission</li><li>Sees no original DATA 2</li><li>May or may not mark later packets as out of sequence — if she <strong>does</strong> track sequence numbers closely, she might mark DATA 3 and DATA 4 as out of sequence, which tells her <strong>loss is upstream</strong> — closer to the sender.</li><li>If she <strong>does not</strong> track sequence continuity, then all she sees is a stream of duplicate ACKs — which <strong>suggest</strong> loss, but don’t conclusively locate it unless she checks whether the missing packet ever crossed her capture point.</li></ul><p>In this example, the key to interpreting the duplicate ack messages is to understand whether you have a copy of the missing packet in your packet capture.</p><h3>Interpreting These Cases: Key Takeaways</h3><p>If we reflect on the examples, here’s what we can say with confidence:</p><p><strong>Retransmissions: Reliable and directional</strong><br>If you see a retransmission and you also saw the original packet, then the loss occurred <strong>after your capture point</strong>, closer to the destination IP of the packet.</p><p><strong>Out-of-order packets: Reliable if clearly defined</strong><br>If you see a packet arrive with a higher sequence number before earlier packets, and you did <strong>not</strong> see the earlier packet, then loss occurred <strong>before your capture point</strong>, closer to the source IP.<br>⚠️ Be careful to distinguish genuine retransmissions from out-of-order packets — don’t count the same event twice.</p><p><strong>Duplicate ACKs: More nuanced, but still useful</strong><br>You can infer location of loss <strong>from duplicate ACKs</strong> — but only if you check whether the missing data packet crossed your capture point.</p><ul><li><strong>If you saw the original data packet</strong>, then loss occurred <strong>closer to the receiver</strong> (source IP of the dup-ACK).</li><li><strong>If you didn’t see the original packet</strong>, then loss occurred <strong>closer to the sender</strong> (destination IP of the dup-ACK).</li></ul><p>Taking this one step further</p><p>The following tips will help you to follow this logic yourself in Wireshark:</p><ul><li>If you’re troubleshooting loss like this, use a general filter like tcp.analysis.retransmission or tcp.analysis.duplicate_ack to give you the ‘big picture’ of what is taking place in the packet capture across multiple TCP streams.</li><li>When you want to double check you’ve interpreted the direction of loss correctly, make sure that you are filtering on just a single TCP stream (‘Conversation Filter &gt; TCP’ context menu option or tcp.stream eq X display filter) — each TCP stream will have its own retransmission behaviour going on, and it’s confusing for most human beings to see them interleaved in the wireshark display.</li><li>Use a wireshark column layout that shows you sequence number, acknowledgment number and next sequence number when you’re troubleshooting.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BepDkjH-NvyM-b0EkFNavA.png" /><figcaption>Wirehsark column layout useful for understanding TCP retransmission behaviour — filter on a single stream and display the tcp.seq, tcp.nxtseq and tcp.ack columns, as well as either TCP port information or IP address information (something that will indicate to you the packet direction)</figcaption></figure><p>Finally, if you want a tool that can act as ‘Charlie’ or ‘Debbie’ and capture packets, invest in packet capture technology for your datacentre. <strong>And if you want real-time warnings when loss is occurring, trends for TCP behaviour, and to reduce the amount of time you spent retrieving and looking at packet captures, invest in a tool like </strong><a href="https://beeksgroup.com/services/beeks-analytics/"><strong>Beeks Analytics</strong></a><strong> which will do this diagnosis for you.</strong></p><p><em>This concludes our </em><a href="https://medium.com/beekstech"><em>Beeks Tech Blog series on Medium</em></a><em>. We hope that you’ve found the articles interesting. If you have further interest in the work that we do at Beeks, </em><a href="https://beeksgroup.com/"><em>visit our website</em></a><em> to find out more information or, for a deepdive, view our </em><a href="https://guides.beeksgroup.com/"><em>technical documentation site</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4f52d77d548" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/relative-location-of-packet-loss-the-theory-behind-the-practice-a4f52d77d548">Relative location of Packet Loss: The Theory Behind the Practice</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why is there packet loss — and where is it happening?]]></title>
            <link>https://medium.com/beekstech/why-is-there-packet-loss-and-where-is-it-happening-fdc513a37624?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/fdc513a37624</guid>
            <category><![CDATA[packet-loss]]></category>
            <category><![CDATA[packet-analysis]]></category>
            <category><![CDATA[packet-capture]]></category>
            <category><![CDATA[wireshark-tutorial]]></category>
            <category><![CDATA[wireshark]]></category>
            <dc:creator><![CDATA[Matt Cretney]]></dc:creator>
            <pubDate>Tue, 06 Jan 2026 17:02:39 GMT</pubDate>
            <atom:updated>2026-01-12T14:18:40.978Z</atom:updated>
            <content:encoded><![CDATA[<h3>Why is there packet loss — and where is it happening?</h3><p>If you’ve ever opened up a packet capture in Wireshark expecting a clear answer and instead found yourself buried in a sea of retransmissions, duplicate ACKs, and “Out-Of-Order” flags… you’re not alone.</p><p>Packet captures can be a goldmine of diagnostic information — but they can also be <strong>overwhelming</strong>, especially if you don’t look at them regularly. Even knowing that loss has occurred isn’t always the hardest part. The question that crops up again and again is: where is the loss happening? Is the problem before or after my capture point?</p><p>That question might seem irrelevant if you’re capturing at one of the TCP endpoints. But modern observability tools like <a href="https://beeksgroup.com/services/beeks-analytics/"><strong>Beeks Analytics</strong></a>, <strong>fmad.io</strong>, or <strong>cPacket</strong> allow you to capture packets <strong>midway through a network path</strong>. That opens up a powerful possibility: using TCP header behaviour to <strong>infer which side of the network path is dropping packets</strong>.</p><p>In this post, we’ll walk through how to tell — using just TCP header info easily accessible in Wireshark — whether loss is happening upstream or downstream of your capture point. We’ll also highlight a few caveats (like why not all retransmissions mean loss) and show how Beeks Analytics can take this work off your plate entirely with automated diagnostics. And hopefully this tip sheet will be a useful reference for you if you can’t find this page when you need it….</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eNeuSU5tFSqGz0rE9usjhA.jpeg" /><figcaption>Where is the Packet Loss? Wireshark Tip Sheet — see main text for the detail</figcaption></figure><h3>First of all — how do I know there is <em>suspected</em> loss in my packet capture?</h3><p>When you just need a quick gut-check for loss, you don’t have to trawl through sequence numbers by hand. Wireshark’s built-in “Expert Analysis” engine already does the heavy lifting for you. The fastest way is to drop this into the display-filter bar:</p><pre>tcp.analysis.retransmission or tcp.analysis.duplicate_ack</pre><h4>What about tcp.analysis.out_of_order or tcp.analysis.lost_segment?</h4><p>Those can <strong>help refine</strong> the story, but they’re not essential for a first-pass triage:</p><ul><li><strong>Out-of-order</strong> often signals reordering rather than loss — useful once you’re hunting root cause, noisy when you only want a “yes/no” answer.</li><li><strong>Lost-segment</strong> is a higher-level heuristic Wireshark derives from the same duplicate-ACK behaviour; you’ll usually see it <em>alongside</em> the two flags above anyway.</li></ul><h3>Now for the key quick diagnosis — where is the loss occurring?</h3><p>Broaden your <a href="https://www.youtube.com/watch?v=EAljq9LAkDo">display filter</a> to include suspect packets ( tcp.analysis.retransmission or tcp.analysis.duplicate_ack or tcp.analysis.out_of_order ), or jump through the packets using <a href="https://www.wireshark.org/docs/wsug_html_chunked/ChWorkFindPacketSection.html">Find Next</a>.</p><p>For any given suspect packets, you can interpret loss as follows:</p><h4><strong>tcp.analysis.retransmission</strong></h4><p>A packet was resent because the sender didn’t get an ACK.<br><strong>Quick diagnosis:</strong> Loss is likely closer to the <strong>destination IP</strong> of the flagged packet.<br>The packet left your capture point, but the receiver never acknowledged it — meaning it probably didn’t reach its destination.</p><h4><strong>tcp.analysis.out_of_order</strong></h4><p>A packet arrived “too late” to fit the normal sequence flow.<br><strong>Quick diagnosis:</strong> Loss (or reordering) is likely closer to the <strong>source IP</strong> of the flagged packet.<br>This could mean upstream loss — or just genuine reordering before it got to your probe. Still, it&#39;s a strong signal to investigate the upstream path.</p><h4><strong>tcp.analysis.duplicate_ack</strong></h4><p>The receiver saw a sequence gap and is sending repeated ACKs to signal missing data.<br><strong>Where the loss occurred depends:</strong></p><ul><li>If you <strong>saw</strong> the original data packet in your capture:<br><strong>Quick diagnosis:</strong> Loss is closer to the <strong>source IP of the duplicate ACK</strong> (i.e. the receiver side).</li><li>If you <strong>did not see</strong> the original packet:<br><strong>Quick diagnosis: </strong>Loss is closer to the <strong>destination IP of the duplicate ACK</strong> (i.e. the sender side).</li></ul><h3>Caveats</h3><p>I would definitely recommend that you also read our next article on this, so that you understand WHY the diagnostics above can be relied upon.</p><p>Also, before you jump to conclusions based on TCP analysis flags, keep these important cautions in mind:</p><p><strong>Capture location matters</strong><br>This logic only works reliably if your packet capture is taken at a single, consistent point in the network.<br>If you’re capturing traffic from multiple locations (e.g. multiple taps, mirrored VLANs, or asymmetric routes), you’ll likely see lots of false positives — especially for retransmissions.</p><p><strong>Out-of-order doesn’t always mean loss</strong><br>Out-of-order packets might be the result of <strong>genuine reordering</strong>, which is totally normal in networks with:</p><ul><li>Equal-cost multipath (ECMP) routing</li><li>Load balancers</li><li>NIC offload features</li><li>Asymmetrical paths</li></ul><p>Don’t treat every out-of-order flag as evidence of loss — look for patterns across time or use multiple views to confirm.</p><p><strong>Not all retransmissions are caused by actual loss</strong><br>Retransmissions can happen even when the original packet <em>did</em> arrive — just not fast enough to satisfy the sender. This is common when:</p><ul><li>The sender’s <strong>retransmission timeout (RTO)</strong> is too aggressive</li><li>The receiver is delaying ACKs</li></ul><p>If you see <strong>a consistent trickle of retransmissions</strong> throughout the trace, it might not be a network fault — it could be two different implementations of the TCP stack not working efficiently together (and can be common in low latency financial market situations where clients may be writing their own TCP stacks).</p><h3>Alternatives to looking at packet captures</h3><p>If, even after reading these guidelines, you still want to spend less time in packet captures and/or you just have too many TCP connections to manage performance by looking at packet captures, then you need a tool like Beeks Analytics which is continually monitoring your connections can tell you where loss is occurring in a simple chart. The chart below (from a Beeks Analytics dashboard) shows you whether loss is occurring outside your network or inside your network, for a given connection:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z-lAf_KsXgFYfp-yKGI6Qg.png" /><figcaption>Beeks Analytics dashboard excerpt, showing TCP loss within your network (inbound) and external to your network (outbound)</figcaption></figure><p>The higher values for outbound loss show that packet loss is affecting this particular connection, and the packet loss is occurring outside your network. All of this without digging into Wireshark captures. <a href="https://beeksgroup.com/services/beeks-analytics/"><strong>Find out more about Beeks Analytics.</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fdc513a37624" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/why-is-there-packet-loss-and-where-is-it-happening-fdc513a37624">Why is there packet loss — and where is it happening?</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Trade Smart, Stay Secure: Mastering Encryption and Observability in Financial Markets]]></title>
            <link>https://medium.com/beekstech/trade-smart-stay-secure-mastering-encryption-and-observability-in-financial-markets-90a9f5ac1f0c?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/90a9f5ac1f0c</guid>
            <category><![CDATA[observability]]></category>
            <category><![CDATA[encryption]]></category>
            <category><![CDATA[fix]]></category>
            <category><![CDATA[tls]]></category>
            <category><![CDATA[security]]></category>
            <dc:creator><![CDATA[Matt Cretney]]></dc:creator>
            <pubDate>Thu, 28 Nov 2024 18:05:25 GMT</pubDate>
            <atom:updated>2024-11-28T18:05:25.390Z</atom:updated>
            <content:encoded><![CDATA[<p>It might surprise some to learn that large swaths of financial market data in transit are still unencrypted. Despite advancements in cybersecurity and widespread encryption in other industries, financial trading infrastructure typically prioritizes speed over security. Every nanosecond counts in financial trading, but at what cost? The decision to prioritize speed over encryption is starting to be challenged by changing regulations, but this can cause unintended consequences for performance visibility and market transparency. In this article, we’ll delve into these challenges and what they mean for the future of financial markets.</p><p>To set the scene, let’s explore how encryption in financial markets evolved, why it has been a hot topic for performance, and the challenges it has introduced for observability.</p><figure><img alt="Conceptual image of global trading, low latency and security" src="https://cdn-images-1.medium.com/max/1024/1*LQc4lJJiW9KXH7v9z8y8Kw.png" /></figure><h3><strong>A Timeline of Encryption in Financial Markets</strong></h3><ul><li><strong>1990s</strong>: HTTPS encryption becomes widely adopted for web browsing, securing internet-based communications.</li><li><strong>1992</strong>: The Financial Information Exchange (FIX) protocol launches to standardize electronic trading. Early versions prioritize functionality over security.</li><li><strong>2000s</strong>: High-frequency trading (HFT) gains momentum, focusing on low-latency connections at the expense of encryption.</li><li><strong>2009</strong>: FIX protocol version 5.0 SP2 introduces encryption standards via TLS, but adoption remains uneven across trading systems.</li></ul><p>Prior to FIX 5.0 SP2, encryption support was not standardized, and implementations relied on network-level security measures (e.g., VPNs or proprietary solutions) rather than native encryption within the protocol. Despite the introduction of TLS support, adoption has been slow in many parts of the trading industry, primarily due to concerns about latency overhead and a perceived lack of necessity for encryption in low-latency environments like cross-connects.</p><h3><strong>The Current State: A Typical FIX Logon Example</strong></h3><p>Even today, many FIX logon messages transmit credentials in plaintext:</p><p>8=FIX.5.2|9=146|35=A|49=TRADER_A|56=BROKER_B|34=1|52=20241113–12:00:00|98=0|108=30|553=<strong>USERNAME</strong>|554=<strong>PASSWORD</strong>|10=004|</p><p>These credentials are accessible to anyone who views this message as it passes between TRADER_A and BROKER_B. This example underscores a significant vulnerability — one that becomes more critical as cyber threats escalate. In typical implementations, the password is not even <a href="https://www.okta.com/uk/blog/2019/03/what-are-salted-passwords-and-password-hashing/">hashed</a> and is simply viewable as plaintext on the wire. Usually, this traffic is going over a dedicated cross-connect to the trading venue. However, it still makes the compliance people at the firms and at the exchanges uncomfortable.</p><h3><strong>Why Hasn’t Encryption Taken Off for Trading?</strong></h3><p><strong>Encryption in Specific Corners</strong></p><p>Encryption initially found its place in customer-facing internet connectivity — like FX single and multi-bank portals — where not only orders but also rate distributions were encrypted. However, when high-frequency trading (HFT) over direct cross-connects emerged, priorities shifted.</p><p><strong>Low Latency vs. Security</strong></p><p>The HFT era brought an obsession with speed. Cross-connects — physical fiber links between trading counterparties — were deemed secure, making encryption seem unnecessary. Furthermore, encryption’s perceived latency overhead raised concerns. While TLS encryption overhead is minimal (typically a few microseconds), in ultra-low-latency environments, even that was considered too much.</p><p><strong>Observability Challenges with Encryption</strong></p><p>Encryption complicates observability because once data is encrypted, monitoring tools must decrypt it to make sense of it. For example, encrypted FIX messages obscure fields that are critical for performance and compliance monitoring, such as order IDs or timestamps.</p><h3><strong>What is the latency overhead of encrypting financial markets traffic? Is it still a genuine issue given newer hardware and technologies?</strong></h3><p>First of all, let’s define our terms here. Often when we characterise web browser performance, we think about encryption overhead as being the extra delay that is introduced when you first establish a new connection to a server. Encryption adds extra steps here, because the encryption protocol must first of all be negotiated. For financial markets trading, most connections are persistent so we are less concerned with the overhead on connection establishment, and more concerned about the overhead for each packet that needs to be encrypted (to be sent) or decrypted (to be received).</p><p>This encryption overhead is generally recognised to be between 1 and 10 microseconds per packet.</p><p>That will be acceptable for some financial market participants, but definitely not for all.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/696/1*v9UT4O2eEwogFilXxxxUXQ.png" /><figcaption>Typical latency sensitivity by type of trading firm</figcaption></figure><p>However, technologists at firms often have to implement infrastructure that can support multiple different use cases of trading style — so the lowest affordable latency usually wins. This usually means avoiding encryption.</p><h3><strong>What has changed to accelerate the adoption of encryption?</strong></h3><p>The narrative is shifting. Firm-wide cybersecurity policies increasingly mandate encryption, leaving trading desks with fewer options to seek exceptions. Simultaneously, governments and regulators are classifying financial infrastructure as critical, demanding higher security standards.</p><p>One specific widely-reported incident which may have focused minds was the <a href="https://en.wikipedia.org/wiki/Bangladesh_Bank_robbery">Bangladesh Bank heist</a> in 2016 also prompted some re-examination of security practice for ‘isolated’ systems that used dedicated infrastructure. Further impetus has been given by the operational resilience initiatives that regulators have pushed since 2020, culminating in the EU Digital Operational Resilience Act (DORA) in 2022.</p><h4>Initiative/Regulators with an Explicit Encryption Mandate (not an exhaustive list)</h4><ul><li><strong>Digital Operational Resilience Act (EU): </strong>Mandates encryption for data in transit and at rest ‘where appropriate’ to ensure confidentiality and integrity.</li><li><strong>Monetary Authority of Singapore (MAS):</strong> MAS Technology Risk Management guidelines require encryption for external/public networks and secure communication.</li><li><strong>Reserve Bank of India (RBI):</strong> Mandates encryption for sensitive data in transit and secure cryptographic key management.</li><li><strong>Hong Kong Monetary Authority (HKMA): </strong>Requires secure encryption protocols for critical data exchanges, including trading links.</li><li><strong>Securities and Exchange Board of India (SEBI): </strong>Mandates secure connections for trading systems, implying encryption.</li></ul><h4>Initiative/Regulators with an Indirect/Implicit Encryption Mandate (not an exhaustive list)</h4><ul><li><strong>Bank of England (BoE) / PRA / FCA (UK):</strong> Requires protecting critical services and secure data transmission, implying encryption.</li><li><strong>Federal Reserve / OCC / FDIC (USA):</strong> Emphasizes secure communication. Encryption required for sensitive data in FFIEC IT Examination Handbook (which supports these initiatives).</li><li><strong>Australian Prudential Regulation Authority (APRA): </strong>Encryption considered a baseline measure under APRA’s CPS 234 (Information Security).</li><li><strong>Basel Committee on Banking Supervision (BCBS):</strong> Recommends encryption as a best practice for protecting sensitive data.</li><li><strong>MiFID II (EU):</strong> Focuses on secure IT systems; encryption implied as part of robust system requirements.</li></ul><h3><strong>Different approaches to meeting these requirements</strong></h3><p>The Korea Exchange were one of the early adopters of encryption technology in its trading systems. Their EXTURE Plus trading system (launched in 2014) incorporated data encryption to protect data integrity and confidentiality.</p><p>Korea Exchange solved the issue in a way which still made analysis of financial timings possible. They did this by putting in place a messaging protocol which encrypted the exchange of credentials and encrypted the transaction details, but kept in plaintext fields that are useful for latency monitoring (message type, a unique message identifier, a member identifier, and the transaction time). This allowed a lot of latency monitoring to remain in place.</p><p>Deutsche Börse have followed suit, mandating encryption from October 2023. However, they allow market participants who are located in the Deutsche Börse colo to only require password encryption (which adds a latency penalty at logon time). All other market participants must use payload encryption, which encrypts the whole connection. Importantly for observability, unlike the Korea Exchange’s implementation, Deutsche Börse’s implementation of payload encryption encrypts all parts of the message, making it difficult for clients outside of their colo from analysing their wire data in order to understand their performance.</p><p>Other implementations of encryption are expected to be more likely to follow the Deutsche Börse implementation than the Korea Exchange one, because this is easier to implement using standard tools like stunnel, and does not change the underlying trading interface message definition (typically a more expensive change for clients).</p><p>Whether the regulators in other territories will be comfortable that local colo traffic remains for the most part unencrypted remains to be seen, and the appetite of exchanges to offer different types of encryption for different types of market participant will likely vary depending on how much of a presence high-frequency trading firms have at the exchange. Crypto exchanges in particular have tended to mandate blanket encryption for all connections, as this is more of a typical ‘internet’ design pattern that suited their early retail investor audience who were likely connecting to the exchange over untrusted networks like the public internet.</p><h3><strong>The Beeks Solution: Bridging Encryption and Observability</strong></h3><p>As encryption of order connections gathers pace, trading firms will need to review their observability options and ensure they are still fit for purpose. Beeks offers a number of solutions in this area, and we also work with exchanges to address the visibility blindspot that encryption leaves.</p><h4><em>Server-Side Observability of Encrypted Traffic (likely suitable for Exchanges, Trading Platform Providers, or Liquidity Providers)</em></h4><p>In this example, Beeks can process session keys provided by the SSL server for server-side decryption:</p><ul><li><strong>How It Works</strong>: Network packets are mirrored to the Beeks Analytics appliance from the client-side interface of an Application Controller. Decryption keys are provided to process encrypted streams.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RSV9OKu9EN65e40tRP6mCQ.png" /><figcaption>Beeks solution for server-side encrypted traffic analysis</figcaption></figure><p>By combining decryption and decoding, firms can monitor performance across encrypted and decrypted streams seamlessly. Clients can, if they prefer, use their packet aggregation network to perform the decryption functionality, if their aggregation layer supports this and they have other (e.g. security) tools.</p><h4><em>Client-Side Observability of Encrypted Traffic (likely suitable for Exchanges, Trading Platform Providers, or Liquidity Providers)</em></h4><ul><li><strong>How It Works</strong>: This assumes that the client can use a different application to perform the encryption (an SSL Proxy) from their Client Application which is sending the messages.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*csptJeqTansvMvC5a3JOSQ.png" /><figcaption>Beeks solution for client-side encrypted traffic analysis</figcaption></figure><p>The server which the SSL Proxy is running on can be fitted with an FPGA card which can mirror the packets to Beeks Analytics. The FPGA card enables minimally intrusive monitoring by securely tapping into application-level data streams without disrupting performance. By using an FPGA card rather than a software-agent based approach, you minimise the latency penalty of having an encrypted connection to your venues and counterparties.</p><h3><strong>Beeks role in supporting you with your traffic visibility challenges</strong></h3><p>Where exchanges mandate that you need to be in their colo to have full observability of your traffic to them, Beeks can provide the infrastructure necessary to host your trading systems, either directly to you or (in an increasing number of cases) in partnership with the exchanges using our Exchange Cloud offering.</p><p>And if you still require full visibility of encrypted traffic, Beeks has shown that encryption and observability no longer need to be at odds. Beeks’ innovative solutions ensure firms can protect their trading data while maintaining critical insights.</p><p>Ready to trade smarter and stay secure? Contact us to learn how Beeks can transform your trading infrastructure.</p><h3>Further reading</h3><p>- The 2021 <a href="https://www.f5.com/labs/articles/threat-intelligence/the-2021-tls-telemetry-report">TLS telemetry report</a> from F5 may be a little out of date but still provides an excellent overview of the difference between encryption protocols, cipher suites and ciphers.</p><p>- More on the Bangladesh Bank heist and its ensuing money laundering activities in the <a href="https://www.bbc.co.uk/programmes/w13xtvg9/episodes/downloads">Lazarus Project</a> podcast (also available on Spotify, Apple, and other podcast platforms)</p><p>-Deutsche Börse encryption <a href="https://www.xetra.com/xetra-en/newsroom/circulars-newsletters/xetra-circulars/T7-mandatory-interface-encryption-Updated-information-on-ETI-LF-payload-encryption-and-additional-support-for-TLS-1.3-3557850">official circular</a>, <a href="https://www.xetra.com/resource/blob/3411670/75e27f438c52e6ceb6fa882c5c4ac8d7/data/ETI-and-FIX-LF-Encryption-FAQ_en.pdf">FAQ</a>, customer focus call <a href="https://www.eurex.com/resource/blob/3477744/8c11f03edd87576714a8af5bdd0f1727/data/Focus%20Call%20Encryption%20Q&amp;A.pdf">write-up</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=90a9f5ac1f0c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/trade-smart-stay-secure-mastering-encryption-and-observability-in-financial-markets-90a9f5ac1f0c">Trade Smart, Stay Secure: Mastering Encryption and Observability in Financial Markets</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beeks Maximizes Storage Efficiency with Advanced Compression Techniques]]></title>
            <link>https://medium.com/beekstech/beeks-maximizes-storage-efficiency-with-advanced-compression-techniques-682d92397c7d?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/682d92397c7d</guid>
            <category><![CDATA[fpga]]></category>
            <category><![CDATA[compression]]></category>
            <category><![CDATA[hardware]]></category>
            <category><![CDATA[observability]]></category>
            <category><![CDATA[packet-capture]]></category>
            <dc:creator><![CDATA[Paul Sheppard]]></dc:creator>
            <pubDate>Thu, 05 Sep 2024 15:47:30 GMT</pubDate>
            <atom:updated>2024-09-09T15:21:35.797Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Maximising Storage Efficiency with Advanced Compression Techniques" src="https://cdn-images-1.medium.com/max/1024/1*iyfbPEC6vdJo3up4-kQDSA.png" /></figure><p><strong>Historical Approach</strong></p><p><a href="https://medium.com/beekstech/how-beeks-delivers-observability-for-financial-markets-04752a708ec9">Beeks Analytics</a> has traditionally utilized the AHA-37x range of cards to compress and decompress capture data during disk writes. These dedicated GZIP compression cards worked seamlessly with any disk type, providing a cost-effective solution that significantly increased storage capacity. An added benefit of compression was the reduction in I/O bandwidth, enabling support for higher input data rates.</p><p><strong>Product deprecation</strong></p><p>However, in 2021 the vendor deprecated the AHA-37xx product line, leaving us to look around for alternative solutions.</p><p>We needed to find a solution that could come close to achieving our goals. Our aim was to achieve 100 Gb/s compression/decompression throughput at a compression rate of around 66% while keeping hardware costs to a minimum. This would allow us to handle any current data stream in the industry. For comparison, our existing AHA-37xx solution would require 3 cards to achieve this goal.</p><p>Other factors that we needed to consider with any card-based solution were the card size and power consumption. We would prefer half-height cards, and to keep power consumption as low as possible (the AHA cards drew less than 50W).</p><p><strong>Potential Solutions</strong></p><p>We evaluated several fresh solutions, weighing their advantages and disadvantages. Our aim was to achieve 100 Gb/s as this would allow us to handle any current data stream in the industry:</p><figure><img alt="Comparison of compression methods — summary of the text from the following paragraphs." src="https://cdn-images-1.medium.com/max/827/1*_D3JaVNoqMEoWovMgPfy0Q.png" /></figure><p><strong>Software GZIP</strong></p><p>Using the CPU to compress data with the gzip compression algorithm is a cheap solution and easy to implement but is not efficient enough to do the job. It could be used in systems with a low data throughput but for the majority of our use cases would not be sufficient.</p><p>A CPU running at 100% usage is not efficient as it stresses the thermal management of the server and causes performance throttling. CPU SW based compression uses a significant amount of memory bandwidth which can affect all processes running on the server.</p><p>It is still useful to have a software GZIP solution available on the system to fallback to temporarily in the event of hardware failure.</p><p><strong>NVMe disks</strong></p><p>Although not a compression option, NVMe disks had reached a price-point whereby it might be possible to use these and benefit from their improved I/O speeds.</p><p>We carried out some tests on the newest NVMe disks and found that they could handle a high data throughput although not up to 100Gb/s. However, the limit here would be more around capacity as we would not be getting the benefits of compression.</p><p>The implications of this are that in some scenarios, we could utilise NVMe only systems without compression where the requirement to maintain a PCAP store is limited.</p><p><strong>GPU-based compression</strong></p><p>This looked like a viable candidate and would potentially be a cost-effective solution as GPU cards are commonly available and existing open-source compression code for GPU’s was available.</p><p>We settled on experimenting with NVIDIA cards using the nvcomp library and ran a series of tests against using an A2 card. Although we got the compression/decompression working with this card and libraries, we found the following issues:</p><ul><li>On the original AHA cards, we had separate compression and decompression cores. With nvcomp, resources are shared, meaning if we maxed out the card with hardware compression, any hardware decompression would affect performance.</li><li>The rates of compression on the GPU card we achieved were not sufficient for our requirements. We would have had to use a significantly more powerful card than the A2, with a significant increase in cost. And to achieve our 100 Gb/s goal, we would have needed more than 1 GPU card.</li><li>The available GPU cards were all full height which would impact our server hardware architecture plus would require 150W+ of power.</li><li>With no load-balancing available, we would have had to invest time and effort to implement this for use in a commercial product offering.</li></ul><p>It was fairly obvious that the GPU approach is possible, but would come at a significant cost and require significant infrastructure changes. This set of drawbacks was such that we moved on to looking at options for an FPGA based solution.</p><p><strong>FPGA-based compression</strong></p><p>Initially we had not considered this option due to the requirement for specialist programming expertise but knew that as a solution it would be fast enough to do the job we wanted. Once we had discarded the GPU approach, we carried out more in-depth research into options in this space and eventually partnered with <a href="http://www.eideticom.com/">Eideticom </a>to use their NoLoad solution to deliver the compression we required.</p><p>We used an Alveo U50 card in our initial development work. This card fitted our physical requirements of being half-height and utilising a maximum power of 75W, so is already a considerable improvement over the GPU card approach.</p><p>We then implemented compression using <a href="https://www.eideticom.com/">Eideticom’s</a> NoLoad engine initially configured with 6 compression accelerators and 8 decompression accelerators. Incorporating the compression was done by integrating our software with Eideticom’s provided userspace APIs (libnoload). Eideticom’s NoLoad API’s provided a flexible approach to implementation, allowing us to build a solution without needing extensive FPGA knowledge, and any queries we had were quickly resolved by their development team. After extensive testing, we found that this initial configuration maxed out at ~45.9Gb/s compression and ~72.5Gb/s decompression for our test datasets. Without much optimization this seemed to be on-target for our requirements, so we continued with further development of the FPGA solution and focused our efforts on achieving our required results.</p><p>Our testing also found that this the most power efficient compression implementation, using less than 30W which is normally the consumption of a set of fans in the server.</p><p><strong>Results Summary</strong></p><p>We continue to work with Eideticom to improve the solution for our requirements, but a brief summary of results that we have obtained to date is as follows</p><ul><li>Compression maximum throughput on a single card peaks at around 50.5 Gb/s</li><li>Decompression maximum throughput on a single card peaks at around 55.7 Gb/s</li></ul><p>These values can be modified for specific use cases by adjusting the ratio of compression/decompression cores on the FPGA, meaning we can also adapt the solution in firmware to suit the expected load.</p><p>The implication of these results is that for the majority of use cases, a single card for compression/decompression will suffice, but more processing power can be added when necessary to scale the performance.</p><p>For use cases such as handling <a href="https://www.opraplan.com/">OPRA</a>, where peaks approaching 100Gbps are possible, a 2 card solution would be required. This is an improvement on our previous AHA-37xx solution which would have required 3 cards. NoLoad provides load balancing capabilities across multiple cards making this an optimum solution for our requirements.</p><p>For improving disk space capacity, we have the compression rate set to achieve around 63% compression. This implies we can store at least 3x as much data on the same size disk as we would have been able to without compression. This is equivalent to what we achieved with the AHA-37x cards and can be improved further by adjusting compression rate settings.</p><p>We started our development using the Alveo U50 card, an FPGA-based dedicated accelerator card available in a half-height form factor from AMD. We have also tested using the Bittware IA-440i card which is available in both half-height and full-height form factor. The fact that we can use multiple vendors for these cards is excellent, as it means we can minimise lock-in to a specific hardware manufacturer.</p><p>In conclusion, we have found using FPGA-compression with NoLoad to be an excellent solution to the AHA-37xx deprecation problem. In addition, having an FPGA card as part of our solution opens up a range of opportunities for future improvements — keep your eye on the <a href="https://medium.com/beekstech">Beeks Technology Blog</a> and <a href="https://beeksgroup.com/">Beeks website</a> for further updates.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=682d92397c7d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/beeks-maximizes-storage-efficiency-with-advanced-compression-techniques-682d92397c7d">Beeks Maximizes Storage Efficiency with Advanced Compression Techniques</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding CPU topology and related performance gains]]></title>
            <link>https://medium.com/beekstech/understanding-cpu-topology-and-related-performance-gains-54d4d7635df5?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/54d4d7635df5</guid>
            <category><![CDATA[performance-testing]]></category>
            <category><![CDATA[cpu-performance]]></category>
            <category><![CDATA[cpu-utilization]]></category>
            <category><![CDATA[low-latency]]></category>
            <category><![CDATA[linux]]></category>
            <dc:creator><![CDATA[Mauro Rappa]]></dc:creator>
            <pubDate>Tue, 27 Aug 2024 14:28:14 GMT</pubDate>
            <atom:updated>2024-08-28T10:52:09.590Z</atom:updated>
            <content:encoded><![CDATA[<p>In a previous article on this blog, we looked at <a href="https://medium.com/beekstech/do-linux-containers-suffer-any-performance-penalty-compared-to-physical-servers-6ee9fd8c69c9">whether Linux containers suffer any performance penalty compared to physical servers</a> — a common problem that engineers and architects in financial services need to tackle.</p><p>In this article, we’re looking at another performance-related topic: CPU architecture and how this affects application performance.</p><p>Nowadays, servers are equipped with dozens or even hundreds of CPUs, organized in hierarchies and directly connected to different internal components such as RAM banks and PCI devices. This arrangement of CPU placement within a server is known as CPU topology. In this article, I will discuss the concept of CPU topology in a server and how to optimize it to improve the performance of a multithreaded program.</p><p>Let’s start by getting familiar with some concepts</p><h3>CPU architecture — history and context</h3><p>First, let’s consider some context so that we can clearly understand what we are trying to achieve.</p><p>It’s very common to refer to the CPU as the unitary brain of the computer. This is certainly right but it’s a bit reductive: with modern (I mean since 20years!) server architectures the internal complexity, hidden away behind a simple abstraction, hides considerable performance gains.</p><p>Another important aspect is the time required for the CPU to access the data located in different memory types.</p><p>The processor is extremely fast at crunching data; however, accessing the data is the main bottleneck. It is quite common for a processing unit to be in an idle state while waiting for data to arrive in its registers. The table below shows how long it takes to access information from different types of memory. Each step away from the CPU takes several orders of magnitude more time!</p><figure><img alt="CPU access time for data in cache versus on disk" src="https://cdn-images-1.medium.com/max/799/1*gZvlJR55s-1Nt79y3hv3sw.png" /><figcaption>CPU access time for data in cache versus on disk</figcaption></figure><p>Since the beginning of the information age, this problem was known and it has generally been addressed by adding a memory between the slow device (ie disk) and the CPU. Nowadays they are referred as cache memory and we generally count tree levels of them, named L1, L2 and L3.<br>L1 is the closest to the CPU, and operates at the same speed (or very close) — it’s a very small amount of memory but very fast.</p><p>L2 is located on the CPU itself and it is bigger (10x) than L1.<br>L3 is placed outside the CPU but is sited strategically between the RAM and the CPU.</p><p>Back in the 90s, IBM compatible PC had this cache memory on the motherboard close to the CPU socket, then the Pentium Pro had it in the CPU package but it was very expensive manufacturing it and a defect in it meant discarding also the whole processor. The Pentium Pro was a commercial failure due to its high price and its successor the Pentium II had the cache on the board together with the CPU but in a different chip (under the fancy plastic cover!). Finally with the Pentium III the cache went back in the CPU since the production cost and quality had improved over time.</p><p>Now let’s examine the latest AMD CPU, each core has its own L1/L2 cache and the L3 cache is shared between multiple adjacent cores and conveniently placed in the middle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/459/1*05VEAGy7YcqkxqxzHZ1Skg.png" /><figcaption>Current AMD CPU cache architecture</figcaption></figure><p>All different Epyc series have different arragments and you can read more about their 4th gen <a href="https://www.amd.com/system/files/documents/4th-gen-epyc-processor-architecture-white-paper.pdf">AMD processor architecture</a> in their white paper.</p><h3>Using Linux to understand your cache architecture</h3><p>With this command on a Linux terminal, you can see how much memory has cache layer:</p><pre># lscpu|grep cache<br>L1d cache: 32K<br>L1i cache: 32K<br>L2 cache: 1024K<br>L3 cache: 32768K</pre><p>The golden rule to make a multithread program run as quickly as possible with steady pace is to let all its threads run on processors which share L2 and L3 memory. Let’s work this through in a specific example.</p><p>A web application has 2 threads: one to authenticate and one to serve the client. Clearly these two threads will need to access to the same data (i.e. a session id), if this data is physically close to the CPU, the overall page will be served faster. If one thread is running on a CPU and the other on a ‘distant’ CPU (which does not share any cache with the prior), the common required data will have to travel across the server and take considerable more time.</p><p>The Linux kernel is aware of the CPU topology (see the <a href="https://github.com/torvalds/linux/blob/ee9a43b7cfe2d8a3520335fea7d8ce71b8cabd9d/include/linux/arch_topology.h#L73">kernel code</a> if you don’t believe me!) and the scheduler tries its best to accomodate it.</p><p>Now let’s get on a Linux console, you can see the full CPU topology with the full output from the utility <strong>lscpu :</strong></p><pre>#lscpu -p<br># CPU,Core,Socket,Node,,L1d,L1i,L2,L3<br>0,0,0,0,,0,0,0,0<br>1,1,0,0,,1,1,1,0<br>2,2,0,0,,2,2,2,0<br>3,3,0,0,,3,3,3,0<br>….<br>49,49,0,1,,49,49,49,6<br>50,50,0,1,,50,50,50,6<br>…..<br>64,0,0,0,,0,0,0,0<br></pre><p>On most Linux distributions you will probably also be able to install <a href="https://www.open-mpi.org/projects/hwloc/"><strong>hwloc</strong></a> package which provides <strong>lstopo</strong>, this utility displays the topology in a nice GUI format:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/441/1*CvzNq_PdZMMMFLPbS7wJmg.png" /><figcaption>Example hwloc graphical display of CPU architecture</figcaption></figure><p>You can also interrogate every single cpu via the pseudo filesystem /sys (in this example a single CPU with 64cores and 128 threads):</p><pre># cat /sys/devices/system/cpu/cpu0/topology/die_cpus_list<br>0–127<br># cat /sys/devices/system/cpu/cpu0/topology/core_siblings_list<br>0–127<br># cat /sys/devices/system/cpu/cpu0/topology/package_cpus_list<br>0–127<br># cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list<br>0,64<br>(newer kernels call it core_cpus_list)</pre><h3>Proving the benefits of OS tuning</h3><p>And now it’s show time ! I used a test suite called <a href="https://github.com/tsuna/contextswitch">contextswitch</a> and in particular the utility <strong>timectxsw</strong> which benchmarks the overhead of context switching between 2 processes.</p><p>I will compare a simple run and one pinning it to the siblings cpu:</p><pre># ./timectxsw<br>2000000 process context switches in 3445039218ns (1722.5ns/ctxsw)<br># taskset -c 0,64 ./timectxsw<br>2000000 process context switches in 2399610671ns (1199.8ns/ctxsw)</pre><p>The result are pretty clear, the second test run a whopping<strong> 30% faster</strong>.<br>I tried other cpu couples and, of course, they couldn’t run as fast.<br>We can use the <strong>perf</strong> utility to understand why there’s such a difference; perf will add its own overhead and for a such quick and specific utility won’t show the big increment as above… I suggest you to look only at the last line (execution time) and the cache-misses percentage : 3.6 % versus 0.12% !</p><p>Random CPU pair:</p><pre># perf stat -B -e cache-references,cache-misses,cycles,instructions,faults,migrations taskset -c 0,50 ./timectxsw<br>2000000 process context switches in 8803593111ns (4401.8ns/ctxsw)<br>Performance counter stats for &#39;taskset -c 0,50 ./timectxsw&#39;:<br>660,331,436 cache-references<br> 23,796,026 cache-misses # 3.604 % of all cache refs<br> 40,910,612,711 cycles<br> 29,170,999,981 instructions # 0.71 insn per cycle<br> 160 faults<br> 1 migrations<br>8.805717047 seconds time elapsed</pre><p>Pinned to the siblings CPU:</p><pre># perf stat -B -e cache-references,cache-misses,cycles,instructions,faults,migrations taskset -c 0,64 ./timectxsw<br>2000000 process context switches in 7977937524ns (3989.0ns/ctxsw)<br>Performance counter stats for &#39;taskset -c 0,64 ./timectxsw&#39;:<br>700,813,458 cache-references<br> 849,871 cache-misses # 0.121 % of all cache refs<br> 35,950,476,130 cycles<br> 19,387,156,865 instructions # 0.54 insn per cycle<br> 159 faults<br> 1 migrations<br>7.979957241 seconds time elapsed</pre><p>Now you may argue “Mauro, your program simply did 2 milions context switches, it’s not realistic!“.</p><p>I will show you another program to convince you it’s worth spending time tuning the OS to speed up your application.</p><p>We can have a look at this lovely c++ utility <a href="https://github.com/rigtorp/c2clat">c2clat</a>. By default it runs two threads on different processors and calculates how many clock ticks are needed for them to read each other memory, like a ping test. After all permutations between all available processors it prints a table like the one below:</p><pre>[root@vmx204 ~]# taskset -c 0,1,64,65 ./c2clat<br> CPU 0 1 64 65<br> 0 0 59 22 57<br> 1 59 0 57 22<br> 64 22 57 0 59<br> 65 57 22 59 0</pre><p>Column and rows are CPU cores, so the intersection between the column 0 and row 64 is <strong>22</strong>, the unit is nanoseconds, this means it takes <strong>22ns</strong> for two threads on the respective CPUs to communicate.</p><p>Instead of running of all 128 cores and generating a huge table, I pinned it to two couple of siblings 0–64 and 1–65, as you can see the lower values is only between siblings, while more than the double delay is noted between the other CPUs.</p><p>I also did the entire test and on my EPYC 9554P 64-Core Processor I noted the worse performance is <strong>243ns,</strong> 10 times slower :)</p><p>Modern CPUs implement simultaneous multithreading (SMT) technology which allows two threads to run on one core (Intel calls Hyperthreading). This ‘core’ provided by SMT is commonly referred to as a logical processor (at least in BIOS terminology). If you disable SMT , the kernel won’t find any siblings and the above optimizions are evaporated, in fact the cpu count seen by the kernel is halved and lstopo will list only one CPU sharing L2 cache:</p><pre># cat /sys/devices/system/cpu/cpu0/topology/thread_siblings_list<br>0<br># lstopo-no-graphics |head<br>Machine (251GB total)<br> Package L#0<br> Group0 L#0<br> NUMANode L#0 (P#0 125GB)<br> L3 L#0 (32MB)<br> L2 L#0 (1024KB) + L1d L#0 (32KB) + L1i L#0 (32KB) + Core L#0<br> PU L#0 (P#0)<br> L2 L#1 (1024KB) + L1d L#1 (32KB) + L1i L#1 (32KB) + Core L#1<br> PU L#1 (P#1)</pre><p>I won’t report its output for brevity, but if you run c2clat accross all processors, you won’t see any latency lower than 46ns; another sign that the logical processors are really interconnected.</p><h3>How Beeks helps you to take advantage of these performance benefits</h3><p>In <a href="https://beeksgroup.com/services/infrastructure-analytics/"><strong>Beeks Analytics</strong></a> we use this locality optimization to add an extra <strong>250k</strong> processed packets per second, which in case of multiple instances, which translates in <strong>1 Million more packets committed to the database</strong>. And if you’re deploying your application on the <a href="https://beeksgroup.com/services/trading-infrastructure/"><strong>Beeks cloud</strong></a> or in our precision engineered <a href="https://beeksgroup.com/services/trading-infrastructure/cloud-services/proximity-cloud/overview/">Proximity Cloud</a> or <a href="https://beeksgroup.com/services/trading-infrastructure/cloud-services/exchange-cloud/">Exchange Cloud</a> solutions you know that we have dedicated system administrators and performance experts who can help you make the most of the underlying operating system, hardware, and connectivity options.</p><p>In a future article I will discuss more in depth about NUMA and PCI ‘topology’ optimization to make sure it’s all coherently organised.</p><p>Note:</p><p>For my tests I used kernel 6.5.7 but I’m confident you can achieve similar results with recent kernel (unless the changelog show work around the scheduler).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54d4d7635df5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/understanding-cpu-topology-and-related-performance-gains-54d4d7635df5">Understanding CPU topology and related performance gains</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Open architecture for wire analytics]]></title>
            <link>https://medium.com/beekstech/open-architecture-for-wire-analytics-013c63d4c36d?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/013c63d4c36d</guid>
            <category><![CDATA[observability]]></category>
            <category><![CDATA[packet-capture]]></category>
            <category><![CDATA[analytics-platforms]]></category>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[fintech]]></category>
            <dc:creator><![CDATA[Tim Billinge]]></dc:creator>
            <pubDate>Tue, 16 Jul 2024 12:32:24 GMT</pubDate>
            <atom:updated>2024-09-18T13:37:04.048Z</atom:updated>
            <content:encoded><![CDATA[<p>In our <a href="https://medium.com/beekstech/how-beeks-delivers-observability-for-financial-markets-04752a708ec9">last blog post</a>, we delved into the importance of wire data for measuring performance in financial markets and talked about extracting wire data via packet captures. We also discussed why it was useful to go beyond traditional packet capture by decoding the data and providing in-depth analyses.</p><p><strong>Today’s post will take a look ‘under the hood’ to see how Beeks Analytics achieves precision data capture for replay and analytics.</strong></p><h3>Rich history of innovation</h3><p>Beeks Analytics has a rich heritage.</p><figure><img alt="Beeks acquires Velocimetrics in 2020, launches Analytics as a Service in 2020, launches Analytics as part of Proximity Cloud and Exchange Cloud in 2022." src="https://cdn-images-1.medium.com/max/707/1*577VADK_ubXLpZp1WwJqaQ.png" /><figcaption>Beeks Analytics timeline</figcaption></figure><p>In 2020 Beeks acquired Velocimetrics, an innovative fintech founded in 2009. Velocimetrics incorporated technologies and expertise from another innovator, TS-Associates, established in 1999. Where Velocimetrics’ main focus was summarising complex network and application events in easy-to-understand business terms, TS-Associates focused on high performance engineered solutions. The combination with Beeks then brought global scale and the ability to host services on behalf of our clients.</p><p>The result? A Beeks Analytics architecture that delivers unparalleled performance monitoring and analysis for financial markets.</p><p>We’ll take a look at the components in Beeks Analytics architecture in the next post in this series. Today, let’s talk about the value it will add to your organisation.</p><h3>Achieving superior observability</h3><p>Beeks is the only cloud provider that’s finely tuned to the core requirements of capital markets. Beeks Analytics extracts high-quality wire information, measured precisely at the boundaries of your operational areas, to deliver insights that support a trusted, favoured, high-speed trading environment.</p><p>We have the following architectural goals in mind as we develop the Beeks Analytics architecture:</p><ul><li>Enable highly performant dropless distributed packet capture.</li><li>Decode any public financial market protocol, and provide tools that make it easy to decode proprietary protocols.</li><li>Make the architecture as open as possible, because wire data shouldn’t live in a silo — it should be easy to integrate with other data.</li><li>Make the architecture as modular as possible.</li><li>Prioritise scalability. You shouldn’t need a small mountain of Beeks Analytics hardware to perform your analytics needs. We commit to always stay ahead of the growing volume of data in financial markets, and offer consolidation opportunities to reduce the monitoring hardware footprint in datacentres.</li></ul><p>Let’s cover each of these in a bit more depth.</p><h4>The ultimate in distributed packet capture</h4><p>Beeks Analytics delivers high-volume and granular capture of wire data. We can capture 100Gbps to 400Gbps without dropping packets. This is important with the growth in financial market volumes.</p><p>We provide this high-performance capture whilst at the same time providing nanosecond-precision timestamping. Or we can read the timestamps provided by switch vendors, which allows us to provide maximum possible microburst measurement. Microbursts are a particular feature of financial markets, where one unexpected price movement or piece of news can trigger multiple overlapping bursts of data. If you don’t size your network to accomodate these, you’ll experience delays or, even worse, data loss.</p><p>The high precision packet capture also includes analysis of network quality metrics such as TCP loss, TCP round-trip or TCP zero window size.</p><h4>Advanced data decoding</h4><p>To enable ingestion of any type of data, Beeks Analytics has a vast range of off-the-shelf decoders for the majority of the world’s financial markets and common underlying protocols. If you need a custom decoder, you can create your own using our <a href="https://guides.beeksgroup.com/BKDA007/Advanced-Configurable-Decoder-(ACD)TM.html">Advanced Configurable Decoder (ACD)™</a> or SDK.</p><p>Decoding data unlocks value added metrics, such as:</p><ul><li><a href="https://guides.beeksgroup.com/glossary/Gap-Detection.html">Gap detection</a></li><li>App to Wire latency for every message</li><li>Statistics breakdown per message property e.g. Fill Ratio, Latency per counterparty, latency per symbol</li><li>Round-trip measurements (e.g. Tick to Order, Order to Ack, Order to Fill) or order tracking, using timings as close to your customer (or counterparty) as possible — providing real-world performance measurement</li></ul><h4>Open architecture</h4><p>At Beeks, we prioritise an open architecture. Beeks Analytics is available as an appliance, a cloud service, or via cross-connect, and if required, you can run the software on your own hardware. This highly-flexible approach frees you from the commercial and operational burden of having many unused appliances cluttering your data centers. We also allow flexible licencing, for example:</p><ul><li>Licencing a subset of cores on a machine, allowing for future growth without needing expensive and wasteful hardware replacement.</li><li>Seamless transfer of software licences from an old machine to a new machine — we want to earn revenue for the quality of our service and our rapid innovation, rather than from charging our customers twice (or more!) for the same monitoring capability.</li></ul><h4>Scalability</h4><p>Beeks Analytics’ architecture is scalable far beyond typical expectations. Our VMX-Capture stack scales per CPU and can be load balanced by an FPGA, leveraging Moore’s law to provide greater visibility at a lower price. This ensures that you can stay ahead of increasing market data and trade messaging volumes cost-effectively.</p><h4>Clean and modular</h4><p>Beeks Analytics is modular, meaning you can install only the software you need. From a basic packet capture solution to a full many-to-many correlation analysis supported by our extensive range of decoders, Beeks can meet your requirement.</p><h3>Open Architecture in more detail</h3><p>The diagram below is a high-level view of the main components and workflows in Beeks Analytics. If you want more information, the <a href="https://guides.beeksgroup.com/">Beeks technical documentation site</a> has more details.</p><figure><img alt="Beeks Analytics architecture — shows the key components, and the processing of data received from the Network, and then stored for packet capture retrieval or summarised for user analytics." src="https://cdn-images-1.medium.com/max/1024/1*zlvL7R1m_mYrd6S9kyKjHw.png" /><figcaption>Beeks Analytics Open Architecture</figcaption></figure><h4>VMX-Capture</h4><ul><li>Responsible for data capture and decoding.</li><li>Packet capture via a set of FPGA cards that deliver high-performance capture and compression, that include a Capture Card that provides buffering, timestamping, and traffic separation, and an optional Compression Card provides compression (when writing captures to disk)/decompression (when reading packet captures).</li></ul><p>At its simplest, you can use VMX-Capture as a packet capture appliance: writing packets to disk and making them available for search and retrieval. However, VMX-Capture does so much more than this. It can also decode data, generate invaluable aggregated statistics, and egress data via our Core Data Feed.</p><h4>VMX-Analysis</h4><ul><li>Responsible for analysing data from VMX-Capture and passing aggregated data to VMX-Explorer for visualisation.</li><li>Supports higher-level configuration of flows and business objects, summarising information about your network in business terms. This includes configuring complex message correlation logic with VEx, Beeks Analytics’ rich expression language.</li></ul><p>The Analysis Server offers a flexible framework that enables you to code and plug in your own processing units (PUs) to add or change service behaviours.</p><h4>mdPlay</h4><ul><li>An optional standalone replay-only version of the main VMX-Capture component.</li><li>Replays real data captured previously by Beeks Analytics to enable you to reproduce conditions that occurred during poor system performance or to optimise performance and test scaling.</li><li>Replays from any packet capture, not just those recorded by Beeks Analytics.</li></ul><p>For more on mdPlay, refer to our <a href="https://medium.com/beekstech/mdplay-multicast-market-data-replay-from-packet-capture-3fe54ab4d450">market data replay</a> article.</p><h4>VMX-Explorer</h4><ul><li>A web-based user interface that uses Grafana to visualise analytics provided by Beeks Analytics.</li><li>Beeks Analytics provides a dedicated data source for Grafana.</li><li>VMX-Explorer includes off-the-shelf dashboards for common analysis needs, and you have complete freedom to create your own queries and visualisations.</li></ul><p>Read more about VMX-Explorer <a href="https://guides.beeksgroup.com/BKDA004/">here</a>.</p><h4>Proximity Cloud and Exchange Cloud portals</h4><p>Beeks Analytics metrics are also available in the Beeks portals for Proximity Cloud and Exchange Cloud services. Beeks portals enable you to manage your infrastructure and monitor it through the same interface.</p><h3>What’s next?</h3><p>In our next posts in this series, we’ll look at a challenging case study that really shows the power of Beeks Analytics, as well as exploring in depth the different types of latency detection, and the complex correlation between different messages in a flow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=013c63d4c36d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/open-architecture-for-wire-analytics-013c63d4c36d">Open architecture for wire analytics</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Beeks delivers Observability for Financial Markets]]></title>
            <link>https://medium.com/beekstech/how-beeks-delivers-observability-for-financial-markets-04752a708ec9?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/04752a708ec9</guid>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[cloud-services]]></category>
            <category><![CDATA[market-data]]></category>
            <category><![CDATA[packet-capture]]></category>
            <category><![CDATA[observability]]></category>
            <dc:creator><![CDATA[Matt Cretney]]></dc:creator>
            <pubDate>Wed, 03 Apr 2024 17:11:40 GMT</pubDate>
            <atom:updated>2024-09-18T13:53:13.498Z</atom:updated>
            <content:encoded><![CDATA[<p>Our motto at Beeks is simple: Build. Connect. Analyse. In this article, I talk more about the ‘Analyse’ part of what we do.</p><p>You’ll have understood from <a href="https://beeksgroup.com/">reading about Beeks</a> how we offer bare metal servers, virtual servers, and managed networking for clients. We specialise in offering these systems at particular locations where there is a lot of financial market liquidity. We also specialise in the security and high performance that financial clients need.</p><p>It’s easy for some people to picture a computer running in a datacentre, and you can think of connectivity as being about enabling computers to talk to each other. But the analyse bit of what we do is a bit harder to picture.</p><figure><img alt="The Gilfoyle character from the HBO series staring at a computer monitor in a homespun ‘datacentre’ (actually running in a garage)" src="https://cdn-images-1.medium.com/max/1024/1*h9gYPbXyP6KYASLZBkaKcA.png" /><figcaption>A computer running in a datacentre, as many people might picture it. Note that Beeks datacentres do not actually look like this. Source: Silicon Valley TV series, HBO</figcaption></figure><p>Why do clients need the analysis part of what we do? Well, the simple answers are <em>performance </em>and<em> service</em>.</p><p>People using Beeks servers typically want to be the fastest — in financial markets, speed often means profit. So measuring your performance (often, it’s as simple as how fast you are) is critical.</p><p>Service is incredibly important, because in financial markets you frequently have firms using Beeks servers who are offering a service to their end customers. For example, you might be a software firm providing a trading platform. If your clients get a poor experience of your platform, they are likely to move away to a competitor, so we need to ensure that the service we provide is exemplary.</p><p>Beeks Analytics (previously known as Velocimetrics) allows our users to quantify and report on both performance and service, and it does this using <em>wire-level data</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/299/1*7gp3y5S8jAciMUDAMrafGw.png" /><figcaption>Beeks Analytics icon</figcaption></figure><h3>The truth is on the wire</h3><p>What is wire-level data?</p><p>Well, imagine you are using an electronic trading platform and you see a promising price to buy US Dollars so you send an order to a bank. If you don’t get the price that you had hoped for, the bank will say that your order was slow to reach them. The provider of the electronic trading platform swears to you that the order left their system on time. This can happen again and again. How do you know who to believe?</p><p>Wire-level data avoids this issue. It allows you to get a copy of the data that is being sent from the wire that connects your trading system to the bank (don’t worry too much for this article how it comes from the wire). Critically, it doesn’t just give you a copy of your messages as they travel through this wire, but it records them for you with highly accurate timestamps.</p><p>Our hypothetical electronic trading platform user can now use this wire-level data to solve the argument between the bank and his trading platform.</p><p>If you are especially mistrustful of the trading platform provider, you might choose to insist that the time source for the wire data is independent (as much as possible) from the time source that they use for their trading platform monitoring.</p><figure><img alt="Left-hand image shows a switchboard operator with old-fashioned copper cabling into a manual telephone switchboard." src="https://cdn-images-1.medium.com/max/1024/1*tBMIU34GjrfzCv8D8QQwQg.jpeg" /><figcaption>‘Wire’ is a general term. Think of the right-hand image, showing fiber optic cabling, more than the left-hand one. Photo credits: <a href="https://unsplash.com/@museumsvictoria?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Museums Victoria</a> on <a href="https://unsplash.com/photos/grayscale-photo-of-woman-using-headphones-TVe0IEdsVc8?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a> (left image), <a href="https://unsplash.com/@larskienle?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Lars Kienle</a> on <a href="https://unsplash.com/photos/person-wearing-silver-diamond-ring-uaOoPv0nNLQ?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a> (right image)</figcaption></figure><h3>From packet capture to advanced trading analytics</h3><p>What we’ve described in the example above is a good example of packet capture. This is where you’re storing packets (in modern computing, messages are sent across networks in one or more ‘packets’), and looking at the packets when you’ve got a particular problem that you want to investigate.</p><p>Beeks Analytics provides this packet capture and timestamping capability, and it can cope with extremely high message volumes when doing so — as much as 200Gbps capture in a single appliance.</p><p>But for packet capture to be useful, you need to suspect that there’s a problem in the first place. In the bad old days, that would be from a user reporting a problem — not great service.</p><p>Where Beeks Analytics comes into its own is that it saves the amount of time you need to spend digging into packet captures by storing statistics that show you when a problem occurred. More importantly, it also allows you to anticipate and prevent where the next problem is coming from. Let’s start to talk about how it does that.</p><h3>Decoding and analysing packets from the wire</h3><p>Moving beyond just storing the packets and the bandwidth information associated with them, Beeks Analytics decodes them as well. What do we mean by decoding? Well, each packet contains a hierarchy of one or more different network protocols. To take one example, the widely-used (in finance) FIX protocol operates on top of the TCP protocol, which itself operates on top of the IP protocol, which itself normally operates on top of the Ethernet protocol. Different information needed for analytics will exist at each level of the protocol hierarchy. For example, to know the symbol on the exchange that is being traded we would need to look at the FIX protocol, but to understand which computer on the local network it originated from, we would need to look at the addresses in the Ethernet protocol.</p><p>If we can decode information from the message, then we can provide value-added information such as:</p><ul><li>Is a host that I’m communicating with struggling to process the volume of data being sent to it? (this would be indicated by a reducing TCP window size)</li><li>Is there a problem where packets are not being delivered first time, and need to be retransmitted? (this is indicated by TCP retransmissions or duplicate acknowledgments)</li><li>Are there UDP messages that have gone missing, and where are they going missing? (this is indicated by gaps in the sequence numbers of the UDP packets that are observed — this is known as <em>gap detection</em> and is particularly important where you are consuming market data)</li><li>Are there delays occurring to any FIX messages that you are sending to or receiving from your trading venues or counterparties?</li></ul><blockquote>By monitoring gaps and latency at multiple points in the network, we can be definitive about where the issues are occurring, and we allow you to be clear about the service level that you’re offering to your clients.</blockquote><p>All of this monitoring is made possible by the wide range of decoders that we make available as part of the Beeks Analytics offering.</p><h3>Beeks Analytics has over 100 decoders for different financial markets protocols</h3><h4>These decoders also cover underlying network protocols and enterprise middleware used within the financial markets.</h4><p>We also make it easy to decode your own custom protocols, without necessarily needing to write your own decoder code</p><ul><li>if your protocol uses Simple Binary Encoding, our standard SBE decoder can use the message schema file as its definition of the fields that need to be decoded.</li><li>If your protocol is UDP based, then our AutoMD™ decoder will automatically detect the sequence number to allow gap detection to work out of the box, without knowing any other details of the message specification. More on gap detection in a future article.</li><li>If neither of the above are suitable, the Advanced Configurable Decoder (ACD)™ allows you to define decoders in JSON configuration. The advantage of ACD is that, once a protocol is implemented using this configuration, there is no release, retest, and reinstall cycle for minor changes in the protocol’s specs. Rather than needing development (sometimes by third-parties) against an SDK to respond to internal messaging changes, you can simply update the JSON configuration.</li></ul><p>We’ve barely scratched the surface of what Beeks Analytics is capable of — you can find more about how it works under the hood in the <a href="https://medium.com/beekstech/open-architecture-for-wire-analytics-013c63d4c36d">next article in the series</a>. Or, for more in depth information, check out our <a href="https://guides.beeksgroup.com/">technical documents</a>, and the Analytics Concepts Guide in particular, to understand how you can gain better observability through Beeks Analytics.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=04752a708ec9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/how-beeks-delivers-observability-for-financial-markets-04752a708ec9">How Beeks delivers Observability for Financial Markets</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Do Linux Containers suffer any performance penalty compared to physical servers?]]></title>
            <link>https://medium.com/beekstech/do-linux-containers-suffer-any-performance-penalty-compared-to-physical-servers-6ee9fd8c69c9?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/6ee9fd8c69c9</guid>
            <category><![CDATA[financial-markets]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[performance-testing]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[containers]]></category>
            <dc:creator><![CDATA[Mauro Rappa]]></dc:creator>
            <pubDate>Tue, 27 Feb 2024 13:50:35 GMT</pubDate>
            <atom:updated>2024-02-27T13:50:35.801Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mDtqKwHTwHkbPjpqZAfXEA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@carrier_lost?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Ian Taylor</a> on <a href="https://unsplash.com/photos/blue-and-red-cargo-ship-on-sea-during-daytime-jOqJbvo1P9g?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a></figcaption></figure><h3>Building a Trading System using Linux Containers</h3><h4>Good idea or bad?</h4><p>In this article, we wanted to talk a bit about how we help our clients with challenging performance issues facing financial markets. Beeks works with many of our clients to implement Kubernetes on the bare metal servers that we provide in colocation facilities and as part of our ‘deploy anywhere’ <a href="https://beeksgroup.com/services/trading-infrastructure/cloud-services/proximity-cloud/">Proximity Cloud</a> solution. One of the topics that often comes up early in these discussions is whether the containers (which are the fundamental building blocks of a Kubernetes deployment) impose a performance overhead that would impact latency sensitive applications.</p><h3>Background on virtualization technologies</h3><p>Any virtualization technology adds a layer of abstraction which introduces a performance penalty. This was certainly the case for the early virtualization techniques which were emulating in software a default set of underlying hardware. Initially Qemu, Xen and Vmware had to emulate a physical server; later they started to introduce para-virtualized drivers. At around this point, Intel and AMD introduced a set of CPU instructions to help the hypervisors to translate memory region assigned to different virtual machines. These instruction sets provide faster access to memory for a virtual machine than if this mapping had to be done in software. The overhead was greatly reduced but you could still see a light performance drop from virtualisation. In our experience, if everything is optimized, you can reduce the overhead to around 10%.</p><p>In parallel to the traditional hypervisor-based virtualization technologies, there have always been alternative approaches available to isolating workloads on an individual host. Sun Solaris had something called Zones (in 2005!), BSD had Jails and Linux had Cgroups back in 2006 (developed by OpenVZ). In contrast to the hypervisor approach (virtualizing hardware), these technologies focused instead on carefully isolating workloads. Using these technologies, you lose the freedom to run binaries intended for different operating systems, but in theory you can isolate multiple workloads without facing any tangible performance penalty.</p><h3>Linux Containers</h3><p>Ok, after this long introduction, what is a linux container? It addition to the aforementioned Cgroup it uses namespaces, which arrived slightly later in the Linux kernel. Both technologies provide separation. CGroup is dedicated to CPU separation, namespaces support separation of other resources such as PIDs, mountpoints, date/time configs and network stacks.</p><p>We can see therefore that a container is not virtualization technology but an isolation technology. This means that, with no hypervisor intermediating the containers access to hardware resources, in theory containers should in theory be able to maintain the same performance levels as physical hardware. However, in practice there have often been doubts that this is the case.</p><h3>So, in short, is there any extra layer in the kernel to provide such abstraction?</h3><h4>The short answer is no.</h4><p>Remember the main task of a kernel is to allow multiple processes to share common hardware, in particular to allow them to effortlessly use memory, network, and disk.</p><p>Every process (called task in kernel terminology) is represented by a structure called task_struct, which is a massive structure of <a href="https://github.com/torvalds/linux/blob/e2ca6ba6ba0152361aa4fcbf6067db71b2c7a770/include/linux/sched.h#L737">hundreds of lines of code</a>.<br>Inside this struct there are multiple linked data types like the <a href="https://github.com/torvalds/linux/blob/e2ca6ba6ba0152361aa4fcbf6067db71b2c7a770/include/linux/sched.h#L800">Cgroups</a> and the <a href="https://github.com/torvalds/linux/blob/e2ca6ba6ba0152361aa4fcbf6067db71b2c7a770/include/linux/sched.h#L1095">namespaces</a>.</p><p>This means every process regardless whether it’s running on the host operating system or in a container is managed exactly in the same way. So there is no overhead in running a container since its processes are allocated exactly as they would be for any process on the base OS. The only difference is that it may have different namespaces (basically, a different list).</p><p>To clarify with an example, think about running all your applications (a browser, Office, Slack and terminals) in single desktop or organize them in two desktops ‘web’ and ‘productivity’. Your computer will execute them as usual but they are organized in a more convenient way. Container technology organize workloads in a more convenient way but it still leverage on the classic kernel; no extra layer is needed.</p><h3>Testing theory with a quick real-world test</h3><h4>So that is what we can understand from reviewing the Linux source code.</h4><p>Now let’s look at a real-world example to demonstrate that there’s no overhead from running within a container. We will look at a comparison between a network test on physical and the same on a container setup.<br>We’ll use two servers with 25Gb network cards connected to the same switch, the software I choose is iperf which can simulate almost any traffic, in my test I will use multicast traffic which is widely used in financial systems.</p><p>On one physical server iperf will be listening for incoming connections:<br>iperf -s -u -B 226.94.1.1 -i 10</p><p>On the other physical server I will start iperf and it will start streaming traffic with a 1Gbits bandwidth:<br>iperf -c 226.94.1.1 -u -T 1 -t 30 -i 1 -b 1G</p><p>the result are presented below:</p><pre>[ ID] Interval Transfer Bandwidth Jitter Lost/Total Datagrams<br>[ 3] 0.0–10.0 sec 1.11 GBytes 957 Mbits/sec 0.022 ms 0/813812 (0%)<br>[ 3] 10.0–20.0 sec 1.11 GBytes 957 Mbits/sec 0.019 ms 0/813805 (0%)<br>[ 3] 20.0–30.0 sec 1.11 GBytes 957 Mbits/sec 0.023 ms 0/813805 (0%)<br>[ 3] 0.0–30.0 sec 3.34 GBytes 957 Mbits/sec 0.013 ms 0/2441492 (0%)</pre><p>The same test was run using a container image with the same version of Iperf:</p><p>docker run -it — net=host beeksgroup/iperf -s -u -B 226.94.1.1 -i 10<strong><br></strong>docker run -it — net=host beeksgroup/iperf -c 226.94.1.1 -u -T 1 -t 30 -i 1 -b 1G</p><p>We can also use podman in this example if you want an open source alterantive to Docker.</p><p>Results on the containers are as follow:</p><pre>[ ID] Interval Transfer Bandwidth Jitter Lost/Total Datagrams <br>[ 2] 0.00–10.00 sec 1.11 GBytes 957 Mbits/sec 0.013 ms 0/813800 (0%) <br>[ 2] 10.00–20.00 sec 1.11 GBytes 957 Mbits/sec 0.023 ms 0/813805 (0%) <br>[ 2] 20.00–30.00 sec 1.11 GBytes 957 Mbits/sec 0.013 ms 0/813805 (0%) <br>[ 2] 0.00–30.00 sec 3.34 GBytes 957 Mbits/sec 0.027 ms 0/2441478 (0%)</pre><p>As you can see the bandwidth was exactly the same and the jitter was very similar across all the tests. So the observed performance matched the theory in this case — which is always nice to find!</p><p>So we’ve proven that in our test containers operating on bare metal hosts can be as performant as deployments of applications directly onto the bare metal.</p><h3>From simple deployments to Kubernetes for capital markets</h3><p>It’s important to note that this simple example focuses on traffic between containers. Our tests were also relatively simple in terms of the traffic profile — running different profiles or a more elaborate container setup orchestrated using Kubernetes might reveal more edge cases where perhaps the overhead of the network layer abstraction imposed by the Kubernetes container network infrastructure (CNI) layer. As container deployments get more complex, you also start to find that latency can be introduced by the load balancing technologies that are used to direct incoming traffic to the appropriate container.</p><p>Beeks aims to give customers who want to run co-located workloads a full choice of container, pure bare metal, or virtual machine — bringing the flexibility of the cloud to front-office capital markets workloads. In future articles we’ll look at more of the kinds of performance challenges and myths that <a href="https://beeksgroup.com/">Beeks</a> helps our internal and external customers with.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6ee9fd8c69c9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/do-linux-containers-suffer-any-performance-penalty-compared-to-physical-servers-6ee9fd8c69c9">Do Linux Containers suffer any performance penalty compared to physical servers?</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Multicast Market Data Replay]]></title>
            <link>https://medium.com/beekstech/mdplay-multicast-market-data-replay-from-packet-capture-3fe54ab4d450?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/3fe54ab4d450</guid>
            <category><![CDATA[algorithmic-trading]]></category>
            <category><![CDATA[market-data]]></category>
            <category><![CDATA[packet-capture]]></category>
            <category><![CDATA[financial-markets]]></category>
            <category><![CDATA[performance-testing]]></category>
            <dc:creator><![CDATA[Alex Simpson]]></dc:creator>
            <pubDate>Tue, 13 Feb 2024 14:01:56 GMT</pubDate>
            <atom:updated>2024-02-14T11:45:10.814Z</atom:updated>
            <content:encoded><![CDATA[<h4>Testing precise timed replay from packet captures</h4><p><a href="https://www.yumpu.com/en/document/read/68611606/e-forex-january-2024">eForex</a> recently published a timely article about the Beeks <a href="https://beeksgroup.com/services/infrastructure-analytics/market-data-replay/">mdPlay</a> service, so we felt for today’s blog update it would be worth doing a deeper dive into how this technology works, and why customers find it so valuable.</p><p>Firstly, it’s important to understand that high volume market data from an exchange is usually distributed using the UDP multicast protocol. Christoph Lameter did a good <a href="https://openfabrics.org/images/eventpresos/2016presentations/114multifinance.pdf">overview</a> in 2016 of why multicast became the technology of choice here.</p><h3><strong>But why replay market data?</strong></h3><h4>Algo performance validation, backtesting, stress testing, and regulations</h4><figure><img alt="Market data replay image" src="https://cdn-images-1.medium.com/max/1024/1*ce4gEeWwRa0wHI6eBfcC7A.jpeg" /></figure><p>But <strong>why</strong> would clients want to replay this data? Well, if you have a trading algorithm you might want to understand how particular market conditions caused certain <a href="https://en.wikipedia.org/wiki/2010_flash_crash">trading decisions</a> to be made. Or you might want to compare how different tweaks to an algorithm react to the same market data inputs. Often this type of functional testing is performed by ‘playing back’ market data logs that your software feed handler (or your trading application itself) has collected, into your application.</p><p>Another more recent need for accurate market data replay comes from the MIFID II regulations which have now been in force for over six years. This mandates testing of algorithmic systems (and associated controls and procedures) at double historic peak volumes (as identified in the specific regulations on <a href="https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32017R0589#d1e759-417-1">stress testing</a> and on <a href="https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32017R0584#d1e982-350-1">trading venue capacity</a>). As a result, we had a clear requirement that mdPlay should support easily scaling up (or scaling down) market data replays, as well as allowing multiple different packet captures to be replayed concurrently (so that peaks from multiple venues, or recorded with different packet capture servers, could be combined in pre-production). This is good practice even for venues outside of the scope of MIFID II. It is especially topical given the challenges that firms face in scaling up their infrastructure to accommodate the <a href="https://www.opraplan.com/document-library">OPRA bandwidth increase</a> caused by the February 2024 increase from 48-lines to 96-lines.</p><h3><strong>The downsides of feed handler market data replay</strong></h3><h4>It does not replicate microbursts and network behaviour</h4><p><a href="https://beeksgroup.com/services/infrastructure-analytics/market-data-replay/">mdPlay</a> takes a different approach. Rather than playing back from logs, it plays back from independent packet captures of the data, and rather than playing directly into an application, it transmits the traffic onto the network. This ensures a higher fidelity playback than the feed handler replay approach — you can understand the interplay of your physical network infrastructure and your application.</p><p>The other key aim that we had for mdPlay was for the replay to have nanosecond accuracy. Financial market data is characterised by <a href="https://www.bing.com/ck/a?!&amp;&amp;p=1e0105c5bf74a06bJmltdHM9MTcwNzM1MDQwMCZpZ3VpZD0zNDdiMjNlMS0wNDJmLTZhNzktMmE4NS0zMTA4MDU3NzZiNDMmaW5zaWQ9NTI0Mg&amp;ptn=3&amp;ver=2&amp;hsh=3&amp;fclid=347b23e1-042f-6a79-2a85-310805776b43&amp;psq=financial+markets+microbursts&amp;u=a1aHR0cHM6Ly93d3cuYXJpc3RhLmNvbS9hc3NldHMvZGF0YS9wZGYvVGVjaEJ1bGxldGlucy9BcmlzdGFNaWNyb2J1cnN0cy5wZGY&amp;ntb=1">microbursts</a> . A financial firm needs to accurately reproduce microbursts because they need to be able to make sure that their algorithms (whether they’re operating software running on a standard CPU, or they’re embedded in an FPGA) can handle these bursts, as well as being able to reliably recreate any issues. For these types of firms, mdPlay needs to be able to offer replays that will quickly pump out a very high bandwidth pulse of data without it being spread over a wider range of time. It must also make it look just like the original pulse that may have caused the original issue.</p><p>When we developed mdPlay, <a href="https://beeksgroup.com/">Beeks</a> looked at three different techniques to ensure we picked the technology that could provide the highest fidelity replay. We assessed fidelity by comparing the replay results with the original market data packet captures.</p><h3><strong>Replay Test Methodology</strong></h3><p>To conduct these tests, we initially captured a busy hour of market data traffic from one of the world’s largest exchanges. The particular exchange was chosen because we knew it to be representative of the type of data our clients are likely to wish to replay. The capture technique employed achieved this with 10 nanosecond tick precision.</p><p>We then transmitted this data using each of the three different tools, and re-captured it (for this we used a specific type of switch known to introduce very little jitter). We also synchronised the clocks on the capture side and the re-transmit side as accurately as possible, using either <a href="https://beeksgroup.com/services/low-latency-network/microwave-and-ptp/">PTP</a> or (where feasible) a PPS signal. Having performed these tasks, we were then able to compare the original capture with the new one.</p><p>For each packet in each capture, we calculated the timestamp delta from the first packet in the capture (the “absolute” timestamp), and the timestamp delta to the previous packet in the capture (“the frame-to-frame delta”). We then compared these figures to see how close the original capture was to the new one. Where a difference in the absolute timestamp delta was revealed, this demonstrated drift in the replay mechanism. Additionally, we interpreted a difference in the frame-to-frame delta timestamp to indicate the amount of jitter in the replay. Here’s what we found….</p><h3><strong>Replay Test Technologies</strong></h3><p>We looked at three technologies for our testing:</p><ul><li><a href="https://github.com/appneta/tcpreplay">Tcpreplay</a> a GPLv3-licenced Unix tool (despite the name, this can also replay UDP). Does not require dedicated replay hardware.</li><li>Two commercially available replay libraries, provided by <a href="https://semiengineering.com/the-rise-of-smartnics/">smart NIC</a> manufacturers. Both these offerings require that manufacturer’s Smart NICs.</li></ul><h3><strong>Replay Test Results</strong></h3><h4><em>The clear winner…</em></h4><p>The best results that we found were with one of the commercially available replay libraries. With the ‘premium’ platform the degree of accuracy achieved significantly surpassed that of the other two offerings. It was able to achieve +/-20 nanosecond replay accuracy, with all packets retransmitted.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*k6k8XakQYRL51r1YQaXNPA.png" /></figure><p>This particular replay mechanism operated dedicated replay hardware and even when challenged with transmitting a big buffer of packets simultaneously, it was able to accurately replay the packets using its on-board clock. In doing so, the premium technology eliminated the possible situation where the replay mechanism receives a packet and is waiting for the operating system (OS) to get to a certain time before it can receive the next packet on the card. This is because the packet’s already with the card and the card does the retransmission in hardware.</p><h4><strong>Synchronised market data replay</strong></h4><p>Also, this was the only platform able to synchronise replays on different ports accurately. In fact, the other platforms didn’t offer any features in their standard tool kits to do this, it had to be done by hand. For clients operating algo trading systems in ultra-time sensitive environments, the ability to accurately synchronise multiple replays on different ports can be very useful as it ensures that packets are put on to the wire in the correct order.</p><h3><strong>Runner-Up Performance</strong></h3><p>Comparatively the second commercially available replay mechanism we tested (we’ll refer to it as the ‘standard’ offering), whilst able to accurately retransmit the vast majority of packets did experience occasional jitter and delay of up to 700 microseconds. However, this was limited to a very small fraction of the total number of packets being transmitted. 99.355% of all packets were replayed within 1100 nanoseconds compared to original capture file.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*cjuvFzXMe7hKmYL6PopHzQ.png" /></figure><p>Like the other commercial replay technology, this platform also required specific replay hardware, but with this particular tool the packets <em>were</em> waiting in software for the right time to be released. (However, it should be said that once it reached that time the card was able to retransmit the packets very quickly as optimisation techniques avoided the need for the packets to travel though many of the OS layers.)</p><h3><strong>Tcpreplay accuracy</strong></h3><h4><em>An open source tool — best price/performance trade-off for retail traders, not suitable for trading firms</em></h4><p>The final replay mechanism tested was an open source tool that does not require dedicated replay hardware. Because Tcpreplay retransmitted files through a regular interface (versus employing optimisation techniques to circumnavigate certain OS layers) the retransmission accuracy suffered.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*oiG_6Ow4sHx1OUi11ODyFg.png" /></figure><p>Furthermore, whilst the other platforms were able to handle nanosecond precision capture, Tcpreplay was limited to being able to access the previously captured data with microsecond precision.</p><p>An interesting feature of the open-source platform (clearly visible in the chart above) was the drift it introduced over time. The errors in each packet accumulated so there ended up being a constant drift. In one of the tests we ran, the retransmission of an hour’s traffic took an hour and a minute to replay.</p><p>In summary, there was a decline in accuracy (whilst mild) experienced between the first and second replay mechanism we tested and then a more substantial differential evidenced between these commercial offerings and the open source tool. However, it’s also important to consider the cost / accuracy trade-off these platforms present. For certain firms for whom nanosecond accuracy isn’t essential this balance may easily swing in the direction of the open source mechanism. Certainly for individual retail traders, it may well be that Tcpreplay (or even <a href="https://towardsdatascience.com/market-tick-data-replay-9719866eb64c">bespoke software replay</a>) is likely to be sufficient.</p><h3><strong>Wrapping a service around the best technology</strong></h3><h4>100Gbps capture and analytics</h4><p>We don’t have time in this article to go into any more depth around the other features that we built into the <a href="https://beeksgroup.com/services/infrastructure-analytics/market-data-replay/">market data replay</a> (mdPlay) product to make it easy to use for our financial clients. Briefly speaking, these include scenario management, an easily manageable library of captures, and integration with our <a href="https://beeksgroup.com/services/infrastructure-analytics/">analytics technology</a> to both record the packet captures at high speed (we are <a href="https://beeksgroup.com/news/unique-modular-cost-effective-data-analytics-for-100gbps-launched-by-beeks-analytics-formerly-velocimetrics-2/">100Gbps packet capture</a> capable) and provide visibility of the quality (gaps, <a href="https://beeksgroup.com/news/beeks-wins-memx-contract-following-competitive-review/">latency</a>) of the incoming feed.</p><p>I welcome any questions you have on these findings and would be more than happy to discuss with you which replay mechanism may prove most appropriate in meeting your firm’s requirements.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3fe54ab4d450" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/mdplay-multicast-market-data-replay-from-packet-capture-3fe54ab4d450">Multicast Market Data Replay</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Process of Rebranding the Beeks Portal]]></title>
            <link>https://medium.com/beekstech/the-process-of-rebranding-the-beeks-portal-8c45ef277497?source=rss----7146a35a7721---4</link>
            <guid isPermaLink="false">https://medium.com/p/8c45ef277497</guid>
            <category><![CDATA[stock-exchange]]></category>
            <category><![CDATA[rebranding]]></category>
            <category><![CDATA[ui-design]]></category>
            <category><![CDATA[private-label]]></category>
            <category><![CDATA[ui-design-process]]></category>
            <dc:creator><![CDATA[Ross Williamson]]></dc:creator>
            <pubDate>Tue, 30 Jan 2024 14:09:09 GMT</pubDate>
            <atom:updated>2024-01-30T14:04:18.567Z</atom:updated>
            <content:encoded><![CDATA[<p><em>We wanted to focus the second article of our technology blog on some of the user interface design work that we do at Beeks. The Beeks Portal is the web-based user interface which clients use to manage their infrastructure which Beeks hosts for them. It allows them to self-service their compute in colo: from power cycling servers to redeploying operating systems to scaling up additional virtual machines. The portal is one of the many ways in which Beeks simplifies the challenges of co-located compute and connectivity for capital markets.</em></p><h3>Introduction</h3><p>In early 2023 <a href="https://beeksgroup.com/">Beeks</a> decided to offer <em>Portal</em> as a private labelled product to our customers allowing them to rebrand elements of the user-interface to fit with their own brands. Up until then, client customisation had been minimal. Beeks’ own brand was used throughout <em>Portal</em>. This article delves into the motivations behind the decision to introduce the option of rebranding and provides a comprehensive overview of the implementation process it entailed.</p><h3>What is Private Labelling?</h3><p>Private labelled software offers unparalleled customisation, enabling clients to seamlessly rebrand it with their own unique identity. Typically available through subscription models, this approach grants customers the rights to utilise and tailor the front-end components of the software. Private labelling permits clients to incorporate their distinctive branding elements, thereby allowing them to present the software to end-users as a natural extension of their own brand.</p><p><strong>Why Private Label with Beeks?</strong></p><p>When the need arises for software to fulfil specific client requirements, an organisation has the option of either developing an entirely new product or identifying an existing solution that can be rebranded as its own.</p><p>Purchasing software from a vendor such as Beeks means you can reduce costs and achieve more rapid time to market. You can also reduce maintenance costs as well as the need for hiring new developers, infrastructure, and security.</p><p>For technology vendors working in the capital markets, private labelling the Beeks portal allows them to offer their clients control of the infrastructure underlying their applications.</p><p>For stock exchanges or other trading platforms, private labelling the Beeks Portal makes connecting to the exchange easier and quicker, and reduces the number of vendors that someone trading on the exchange needs to deal with in order to get connected.</p><h3>How We Private Labelled Portal</h3><h4>Identify Page Elements</h4><p>The first stage in the process was to identify the individual page elements which required customisation in order to move away from the Beeks colour scheme. We determined that the following elements would need to be customisable for each customer brand:</p><ul><li>logo</li><li>favicon</li><li>button colours</li><li>email addresses used for contacting technical support</li><li>URLs for Terms &amp; Conditions and Privacy Policy</li></ul><p><em>Portal</em> uses a language file which contains most of the application’s UI text. We identified elements in the language file which would need updating including many instances of the name “Beeks” which needed removing.</p><p>Language file examples:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iZLdqVlTVJTgxa0ASJteXQ.png" /></figure><h4>Beeks Colour Palette</h4><p><em>Portal</em> was originally designed using Beeks’ own brand colours. The Beeks colour palette features some strong, vibrant colours but these were not compatible with our clients’ own branding.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OCl9BZzT-yu1ua5hgZcTTg.png" /><figcaption><strong>Beeks’ Original Colour Scheme</strong></figcaption></figure><p>The key was to design a colour scheme which would work for all of our clients’ brands irrespective of the clients’ colours. For example, some of our customers use green as their primary brand colour while another customer uses purple — the new colour palette had to be compatible with all brands. It also had to be able to work with all potential future clients’ brands.</p><h4>Introduce a Neutral Colour Scheme</h4><p>To ensure the new colour scheme wouldn’t clash with our customers’ brands, we opted for a neutral grey, black and white palette which is easy on the eye and compatible with all potential colour schemes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2PT4nZQ0moHpinDqyZQxdQ.png" /><figcaption><strong>Beeks’ rebranded colour scheme</strong></figcaption></figure><p>We incorporated each client’s primary brand colour on specific page elements to make them stand out and draw the user’s attention. We also introduced a secondary colour which would be a lighter version of the primary brand colour, used for things such as highlighting page elements on hover.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DFagIJVGS6foljBlR-g8rQ.png" /><figcaption><strong>Client A’s</strong> <strong>rebranded colour scheme</strong></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mwGJnuu2sgqIIz5hWKYxEQ.png" /><figcaption><strong>Client B’s</strong> <strong>rebranded colour scheme</strong></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H9yI5uTH6vgmxAXxDXxqwQ.png" /><figcaption><strong>Client C’s</strong> <strong>rebranded colour scheme</strong></figcaption></figure><p>Example usage:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QkvF4_6kFPpE5py0wI4M3A.png" /></figure><h3>Examples</h3><p>Below are some examples of the <em>Portal</em> pages before and after the rebranding process. The <em>before</em> pages feature Beeks’ original branding and the <em>after</em> pages showcase the corresponding redesigned <em>neutral</em> versions. Furthermore, there are examples of clients’ brands in use.</p><h4>Portal Login Page</h4><p>The original login page design utilised Beeks’ three main brand colours and the company logo. On the rebranded version, the only page element that requires a colour change is the <em>Log In</em> button.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ArKOlYV-pGIEFApVUwePzw.png" /></figure><h4>Analytics Page</h4><p>As well using the Beeks’ brand colours, the <em>Analytics</em> page also featured a set of strong greens and blues for the various metrics on display. These colours were problematic for clashing with clients’ brands and were therefore removed during rebranding.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*72mXLBA7LsXgaVMfjjlCwg.png" /></figure><h4>Rack Elevation Page­­­</h4><p>The <em>Rack Elevation</em> page uses some additional desaturated blue colours. Again, these colours were chosen for their ability to be used by all client brands.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FVE9rKejDHk5LNmtsmyaDw.png" /></figure><h3>Conclusion</h3><p>Most of our customers still use the Beeks brand, but the ability to offer their own branding has been a key enabler for some of our customers, in particular for clients taking the Exchange Cloud version of our product.</p><p>While rebranding, we incorporated many design improvements into the <em>Portal</em> UI. Our product is now more appealing to our existing customers while also being more attractive to potential new customers.</p><p>As well as being able to offer our clients the option to rebrand our application, we have successfully achieved a user interface that exudes a cleaner, more contemporary aesthetic, while simultaneously enhancing its usability.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c45ef277497" width="1" height="1" alt=""><hr><p><a href="https://medium.com/beekstech/the-process-of-rebranding-the-beeks-portal-8c45ef277497">The Process of Rebranding the Beeks Portal</a> was originally published in <a href="https://medium.com/beekstech">Beeks Technology Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>