<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Douglas Gourlay on Medium]]></title>
        <description><![CDATA[Stories by Douglas Gourlay on Medium]]></description>
        <link>https://medium.com/@loopback0?source=rss-4a729eb24d9f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*IMtpvhxjoqYsXt_3JNl2Xg.jpeg</url>
            <title>Stories by Douglas Gourlay on Medium</title>
            <link>https://medium.com/@loopback0?source=rss-4a729eb24d9f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 00:56:09 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@loopback0/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Why do Ethernet switch ports matter?]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@loopback0/why-do-ethernet-switch-ports-matter-b4fd3b6b3c8e?source=rss-4a729eb24d9f------2"><img src="https://cdn-images-1.medium.com/max/768/1*6WCedA8OPIJu2Wd-Ef8R4w.jpeg" width="768"></a></p><p class="medium-feed-snippet">So yesterday morning on a flight from San Diego to San Jose I wrote my first real post here on Medium. It was fun to write again, it has&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@loopback0/why-do-ethernet-switch-ports-matter-b4fd3b6b3c8e?source=rss-4a729eb24d9f------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@loopback0/why-do-ethernet-switch-ports-matter-b4fd3b6b3c8e?source=rss-4a729eb24d9f------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4fd3b6b3c8e</guid>
            <category><![CDATA[arista]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[ethernet]]></category>
            <category><![CDATA[cisco]]></category>
            <category><![CDATA[networking]]></category>
            <dc:creator><![CDATA[Douglas Gourlay]]></dc:creator>
            <pubDate>Wed, 06 Feb 2019 05:23:27 GMT</pubDate>
            <atom:updated>2019-02-06T05:23:27.226Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[In the beginning — Intent Based Networks]]></title>
            <link>https://medium.com/@loopback0/in-the-beginning-intent-based-networks-b46ad80f8144?source=rss-4a729eb24d9f------2</link>
            <guid isPermaLink="false">https://medium.com/p/b46ad80f8144</guid>
            <category><![CDATA[dns]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[networking]]></category>
            <dc:creator><![CDATA[Douglas Gourlay]]></dc:creator>
            <pubDate>Tue, 05 Feb 2019 16:42:50 GMT</pubDate>
            <atom:updated>2019-02-05T16:55:58.179Z</atom:updated>
            <content:encoded><![CDATA[<h3>In the beginning — Intent Based Networks</h3><p>In 2003 I was fortunate to be the product lead for the development of the Cisco Nexus line of Ethernet switches — weird because they didn’t ship until 2008 and I didn’t even figure out what to name them until 2005 or 2006 — things get a bit fuzzy with time (I often joke that it came to me in the shower, but it was actually on a list from a naming study we funded.)</p><p>One of the key things we were doing was developing a new operating system for our Ethernet switches that would, hopefully, deliver the level of reliability and quality that were deemed essential in the data center environment — this was critical because we felt in this timeframe that the consolidation of the FibreChannel network onto an Ethernet substrate was a critical requirement — I think time has proven us wrong on this one, but it seemed important in 2003. (Well, technically it seemed really important to my boss and as such of some import to me. Side note, my boss at the time had just come back into Cisco from a FibreChannel switching company Cisco acquired.)</p><p>In the course of this process of defining a new operating system I viewed this as a chance to ‘do things differently’ than we did before, change up a few things, shake the tree, and any other appropriate analogy for disrupting the status quo you want to throw at it. You see, in the most base example I simply felt that we shut switches and routers down ‘wrong’. I use the air quotes because this used to infuriate some of the engineers I would work with. I would ask leading interrogatory questions like, “does our switch reload ‘right’” and the typical response was something along the lines of “well, you typed ‘reload’ and the box rebooted — so yes, it rebooted ‘right’”. This of course elicited the empathetic response of, “but is that what our customers want?”</p><blockquote>Is that what our customers want?</blockquote><blockquote>Do our customers know what they want?</blockquote><blockquote>If I built what my customers asked for I would have invented a faster horse — Henry Ford.</blockquote><p>I contended that what our customers really want, at some uber-meta level is a network that is highly reliable and maximizes uptime. I would then wax eloquent into some rehearsed diatribe about how what customers really wanted when they typed reload was a switch/router to exit the network as gracefully as possible, making its pleasantries and gently waving goodbye with a Miss America-style wave as it gently left the network as a John Williams track faded softly out of audible range.</p><p>Or in network engineer speak: set route metrics to infinity, hold down until traffic counters drain to close to 0, send LACP markers, adjust STP weights, snapshot the number of routes, adjacencies, and link state, and then execute the reload. (the idea basically being that you can pre-converge a network topology around a planned or human-initiated outage when 90–95% of the outages we have are caused intentionally by the judicious use of the command ‘reload power now’ or the slightly more immediate ‘tripping over the power cord emergency change control’.)</p><p>I made getting this feature done a personal crusade of mine. For the next 6 years at Cisco. <strong>I got nowhere</strong>.</p><p>I joined Arista in the summer of 2009, full of renewed vigor, full of excitement for what we could build, and I found a welcome ear in Ken Duda — the head of SW engineering. For the first time in many years I found someone who ‘got it’ and realized that the operating models of the network were an expression of system quality — and they could be drastically improved.</p><p>Ken and I would get together along with another two willing culprits in our goal of changing the way networks operated — Andre and Mark — and we expanded on the concept, we recognized that the way a switch booted was wrong too. It is often just as disruptive to cold boot a connected network element and have it insert itself forcefully into the network — <em>VTP bomb anyone?</em></p><p>It was around 2011 that, as we continued to evolve our customer story, the concept of location-based system profiles started to take shape: the behaviors of a leaf switch were profoundly different than a spine — the configuration complexity, the type of services enabled, the actual goal of the network element is fundamentally different. Yes, they both forward packets — but the spine switch connects other switches where the leaf switch primarily connects hosts. As such they have a different set of controls used to signal reachability to their primary consumers of this information.</p><p>There was one stand-out meeting where a lot of this coalesced — Ken and I had been chatting about test driven development — where our engineers write a test plan and their spec, then write their software. When their test plan passes they know their software is working. In some organizations one person writes the test plan while another writes the functional software. (note to product managers and aspiring product people — read the functional spec, but don’t trust it. If you read the functional spec, <em>study</em> the test plan — that is your source of truth. Publish road maps based on the test plan, not the concept commit).</p><p>Back to this meeting: It occurred to us, like a ton of mayonnaise falling from the sky, that there is not prescriptive test plan for the network. I mean, we write these multi-hundred line configurations, but how do we evaluate if the configuration actually delivered what we wanted? How do we and where do we express what it is we want our network to do? Where do we define what this collection of cables, hardware, and network software is supposed to achieve?</p><p>This is where we, I guess invented, but at least coined the term ‘Intent Based Networking’. We realized that what we wanted was a way for our clients to be able to state what outcome they wanted — then in real time — evaluate their cabling, their hardware selection, and their software configurations against this statement and verify if the collection of variables under the control of the network team is delivering that which was expected.</p><p>A common example I would use is —</p><blockquote>If I take a leaf switch and I connect 4x40Gb uplinks to a single spine switch (when I have four available) will it work? (the answer is yes).</blockquote><blockquote>Will I get any errors? (nope, none).</blockquote><blockquote>Any indication that this is stupid? (nada!).</blockquote><blockquote>Is it dumb? (exceedingly).</blockquote><p>So we literally have these network systems that let you make colossally dumb decisions that increase the operational risk of the entire organization in such a way that the next routine maintenance or change control or error on a spine switch could cause the complete isolation of ~40 servers which means ~400 VMs, or ~2000–4000 containers or some mission critical data warehouse or transactional database — and we didn’t even warn you.</p><p>Solving for this improvement in User Experience — where the user is the network operations team — was our goal — it was another expression of system quality — it was a way of showcasing an empathetic relationship with our clients and potential clients — it was simply, a way to make their jobs easier and more pleasant. This was the initial goal of Intent Based Networking.</p><p>In its first incarnation and prototypes it was pretty rudimentary. We envisioned a system where you would create a group of switches with a name like ‘DataCenter1’ and then assign roles ‘Leaf’ and ‘Spine’ for instance and for the switches labeled ‘Leaf’ you would define an expected connectivity pattern like ‘uplink-pattern 40/40/40/40’ this would define that this leaf should have 4x40Gb uplinks each connecting to a separate spine switch. The Network Name ‘DataCenter1’ would also have an associated pattern such as ‘network-pattern L3LS’ — a Layer-3 Leaf/Spine network. With these four lines of ‘code’ you define a networks expected outcome:</p><blockquote>network DataCenter1</blockquote><blockquote>network-pattern L3LS</blockquote><blockquote>node-role Leaf</blockquote><blockquote>uplink-pattern 40/40/40/40</blockquote><p>Also, with these four lines of code we can evaluate if a system is ‘Green’ — operating properly within the defined patterns, ‘Yellow’ — operating and passing traffic, but not meeting the expectations of the pattern, or ‘Red’ — service is unavailable.</p><p>Since that time, about eight years ago I presented this topic at a few Network Field Day events, did a few webinars, and watched a handful of companies take this concept and add a lot more capability to it — extending it to include automation, provisioning, analytics, multi-vendor awareness and so on. I then watched industry giants usurp the term and try to make it their own, but somehow losing in the translation that the goal was to improve the networks <em>for our customers </em>— and not just create the opportunity for a sales team to have another chat with a client or to cram some overpriced analytics package into their enterprise.</p><p>I still like the concept of Intent Based Networking in its purist form — a method to define the outcome you want, and then evaluate your infrastructure choices — a network expression of test driven development or extreme engineering depending which book you read in the late 1990s.</p><p>It is my belief that Intent Based systems can be stand-alone but are absolutely augmented by integration with provisioning, automation, analytics, and will probably have a future that involves AI/ML driven operations as well — side note and foreshadowing a future project: I think our antiquated QoS models are ripe for a dynamic and real-time model that reacts based on a data-driven model informed through rich time-series analytics.</p><p>This collection of capability probably needs a name other than Intent Based Networking — not sure what that is yet, but I am not a fan of perverting our original intent into some macro-vendor-acronym-congolomerate that means nothing other than I spend more for something I won’t use to most clients (pun intended — <em>ok, this is getting fun</em>).</p><p>While it is neat to see something that you are the progenitor of take on a life of its own and start getting mass awareness and contribution to it; however, I would caution the vendors who are using the term ‘Intent Based Networking’ to pay attention to the core goal we had at its inception — improving the operating environments for network teams. If you find that your interpretation of Intent Based Networking is more of a selling strategy, a messaging pillar, or ‘wow thats popular let’s add it to our marketing playbook and brand because it sounds cool’ — then you are missing <strong>our intent</strong>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b46ad80f8144" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Vendor Responsibility and Product Safety]]></title>
            <link>https://medium.com/@loopback0/vendor-responsibility-and-product-safety-146e0b9c2c3e?source=rss-4a729eb24d9f------2</link>
            <guid isPermaLink="false">https://medium.com/p/146e0b9c2c3e</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[it-security]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[government]]></category>
            <category><![CDATA[vendors]]></category>
            <dc:creator><![CDATA[Douglas Gourlay]]></dc:creator>
            <pubDate>Thu, 09 Jun 2016 16:03:45 GMT</pubDate>
            <atom:updated>2016-06-09T16:03:45.229Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Vendor Responsibility and Product Safety</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/1*lq-D_WvUpKX3gM_IzBIZhw.jpeg" /></figure><p>For twenty years now I’ve been building and operating IT products: switches, routers, secure enclaves, etc. The overwhelming majority of vendors I know have a goal of building great products that meet or exceed customer expectations in performance, reliability, quality, etc. With the trend towards the consumerization of IT and an increasingly competitive marketplace I’ve witnessed an increased focus on time-to-market. There have been many examples cited where there is a significant first-mover advantage and a <a href="http://www.newyorker.com/tech/elements/in-silicon-valley-now-its-almost-always-winner-takes-all">winner/fast-mover-takes-all type of outcome</a>, whether we are discussing car sharing, home sharing, tablets, or the latest enterprise technology. This unerring focus on speed may be having some unintended consequences.</p><p>When you hurry a product schedule along you aim for the MVP — the <em>minimum viable product</em>. In short, what is the least we can do that will fulfill the customer’s expectation and hopefully enable this product to sell, capture some market share, and then based on real-world customer feedback add incremental capabilities to the offering to further expand the market and adoption rate. It is a model that has been getting a lot of press as it is more agile and nimble than the take-your-time, ‘nail it and scale it’ model. The costs, though, may be not only in quality and testing, but as we have seen more recently, cyber security is often being ignored in the overall product design and architecture. There have been too many reports over the past year of companies who should have known better shipping products with gross flaws and then ignoring the feedback from trusted security researchers.</p><p><strong>The Regulatory Risks of Insecure Products</strong></p><p>My worry is that if we, as an industry, continue to ship bad products to consumers we could find ourselves in the same situation as other industries: regulated. Regulation is a dirty word to many. It implies process, auditors, lengthy policy documentation, lawyers, and everything that spells disaster to a small and nimble business. If you’ve ever sold an IT product to the US Government, especially the DoD, compared with selling it to a medium business, you’ll know what I am talking about. With the Internet of Things a myriad of critical infrastructure systems are now getting connected to Internet and cloud-based management systems; the risk of vendor missteps now goes up, and enters the same level of product safety responsibility and liability we have historically seen in industries such as financial services, automotive manufacturing, and pharmaceutical development.</p><p>I had the pleasure of educating an attorney the other week about the impact of human error on networks. We talked about how with a few bad routes you can crash financial trading systems across major markets; how heart monitors and drug dispensaries can be disabled and black-holed, how every user/machine/application at a bank could be effectively disabled. These were for accidental outages. When we think about purposeful attacks — exploiting vulnerable and legacy systems — the risk gets much higher: the more connected we are and our systems are, the broader the surface we need to protect is, the greater the potential impact of any compromise.</p><p>Let me be very clear about my opinion here: I do not want to see government regulation of IT product development. I also would prefer to not see self-regulation via committee-based standards bodies dominated by big vendors preserving market-share. Either would dramatically stifle the technology industry’s ability to bring new products to market, delight customers, and continue to innovate.</p><p>If we, as an industry, do not step up and start being better citizens and better partners to our customers I am fairly confident we will see some form of regulation within the next 3–5 years.</p><p><strong>So what can we do to forestall this?</strong></p><p>The obvious answer is to build products that are inherently secure, but our actual responsibility is greater than that. We also have an emerging responsibility to train and educate the users of our products on their proper usage. As many vendors shift from one-time enterprise site licenses to ongoing recurring revenue in SaaS models our responsibility to provide not only a secure product but a secure service increases further. When I wrote about the <a href="https://skyportblog.com/2016/05/13/five-necessary-improvements-to-the-swift-security-model/">SWIFT breach</a>, I was not casting aspersions at their product, or its security model — although like any product there is always room to improve — I was instead noting that they did not provide prescriptive and deployable guidance to the least-common denominator of their user community about how to properly secure their product and service.</p><p>Vendors have a new, higher level, of responsibility:</p><p>1) We have a responsibility to provide our end-users, operators, and administrators prescriptive guidance on how to secure our products and services. This is not a point in time guidance, but needs to be treated as a living document. It should also include audit and evaluation criteria so the customer can validate their implementation against the reference model. Kudos to Microsoft for doing this with their <a href="http://aka.ms/ESAE">Active Directory security guidance</a> — this helps everyone.</p><p>2) Guidance should always strive to be vendor agnostic, although can and should include any specific SW/FW/HW versions and specific configurations that have been tested and verified. The more prescriptive and exacting the guidance is the better. One example is to not just say ‘segment this application’ but instead provide a list of ports and protocols used and explicitly state which ones need to be allowed in/out for the application to properly function. Even better is stating where each port/protocol needs to go to.</p><p>3) A method for feedback from and engagement with security researchers — this should also have a named and responsible party for implementing the feedback in a timely manner. These “bug bounty” type programs should always accompany an internally implemented and documented secure coding process. Organizations should be prepared to share the documentation of their process and customers should routinely to see it.</p><p>4) It would be nice for there to be a consistent format that a vendor could document the ports and protocols required to go to their product and why each of them is open and necessary. If this could be done in a consistent format and stored in a central repository it would be rather nice for simplifying deployment of in-band security protections and also monitoring for unauthorized traffic types.</p><p>5) Signature signed images and secure software distribution mechanisms for updates and patches to critical components such as crypto libraries and for CVEs.</p><p>For each of the above it is equally important to have a test so the organization can verify, “Ok, I thought we did X, how do I know for a fact I am doing X properly?” Some common examples that jump to mind are managing ACL lists in switches and routers but never performing garbage collection, or name chasing where you accept an FQDN but only resolve it once and cache it until reboot.</p><p>Am sure there is much more that should be included here — would love to invite any feedback or discussion on what could probably evolve towards an ‘<em>Enterprise IT Bill of Rights’</em> that we would hope vendors would subscribe to: we have to do better.</p><p><strong>Next Steps: What Happens With Continued Weak Security and Vendor Inaction</strong></p><p>In conclusion, let’s talk for a second about what happens if we don’t step up our game: In the 1950’s automobile manufacturers had incredible freedom and flexibility to do what they wanted, each make and model could vary greatly year-by-year to innovate and differentiate from each other. It was a golden-era of mechanical and industrial innovation.</p><p>However, there were cars with no seat belts, cars that caught fire when rear-ended, cars that had sharp instruments that literally impaled people upon impact. In 1966 the US enacted the<a href="https://en.wikipedia.org/wiki/National_Traffic_and_Motor_Vehicle_Safety_Act"> National Traffic and Motor Vehicle Safety Act</a> initiating a federal regulatory agency and practices for vehicle design. There are now literally thousands of pages and hundreds of separate regulations that an automotive manufacturer must adhere to from safety, fuel economy, hazardous material usage, down to consistent operations such as requiring turn signals, and rear-view mirrors. There was a subsequent uproar from the manufacturers when airbags were mandated and fuel economy requirements tightened. There is no doubt that the regulations improved the product safety, but I wonder if the industry had established their own standards, self-policed and self-regulated, would we have had need of these government regulations and the taxpayer burden they carry.</p><p>An example where an industry self-policed is the payment card industry with the adoption of the PCI standards for systems accepting credit/debit cards. Not only are there fairly clear guidelines on how to deploy the systems and protect them, there is also an integrated audit process and guidelines for self-assessing the security posture of the environment. The challenge, as seen in the recent <a href="http://www.businessinsurance.com/article/20160602/NEWS06/160609935/chubb-p-f-changs-federal-insurance-co-cybersecurity-by-chubb-credit?tags=%7C68%7C83%7C302%7C299">PF Chang’s ruling</a>, is that self-regulation does not always equate insurance indemnification.</p><p>The products we build, as an industry, run power grids, financial trading systems, life support, and air traffic control. When these systems are at stake our customers deserve more than a ‘minimum viable product.’ Our customers deserve our best efforts in security as well as agility.</p><p>dg — @dgourlay</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=146e0b9c2c3e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>