Runt Packet – Arista Networks and Data Centre Switching

Greg has asked Doug Gourlay from Arista Networks to get his feet under the workbench and talk about Arista Networks. A reasonably new vendor to community, Arista have a unique approach to building Ethernet switches that are software driven for enhanced data centre management and better operation.

Arista Networks have interested me for some time because their hardware architecture is very different from traditional network equipment, with a focus on their EoS software platform being consistent across the entire product line. We didn’t have time to talk in more detail but either contact Arista to learn more about their product, or get in contact by leaving a comment or use the contact form, and I can schedule more time.

You can check out Arista Networks or follow Doug on Twitter at @dgourlay.

Feedback

Follow the Packet Pushers on Twitter (@packetpushers | Greg @etherealmind | @ecbanks) and send your queries and comments about the show to [email protected].  

We want to hear from you!

Greg Ferro
Greg Ferro is a Network Engineer/Architect, mostly focussed on Data Centre, Security Infrastructure, and recently Virtualization. He has over 20 years in IT, in wide range of employers working as a freelance consultant including Finance, Service Providers and Online Companies. He is CCIE#6920 and has a few ideas about the world, but not enough to really count. He is a host on the Packet Pushers Podcast, blogger at EtherealMind.com and on Twitter @etherealmind and Google Plus.
Greg Ferro
Greg Ferro
Greg Ferro

Latest posts by Greg Ferro (see all)

  • http://www.aristanetworks.com Douglas Gourlay

    Thanks Greg, fun times! btw- my Twitter ID is @dgourlay

  • http://perlmonkey.blogspot.com/ Russell Heilling

    Very interesting subject choice again. As a long time service provider switch bitch anything that makes managing l2 domains easier is good with me :)

    I found the section on early innovations from Cisco especially interesting. When many of the technologies mention launched, EIGRP, VTP, ISL, etc. They did lead the field and drove voluntary market uptake by innovation. At some point this seemed to change and these proprietary technologies became a burden and reduced the choice of the user by locking us in to a certain vendor.

    I am all for a vendor that makes my life easier but am very wary of proprietary protocols. The immediate benefits are tempting but can often turn to a world of pain in the future.

    I will watch Arista with interest.

  • Alexei Monastyrnyi

    Guys,
    I have being listening to your podcasts regularly. Good choice of topics and people on the shows. The one with Ivan Pepelnjak was such a thrill. :-) You know BTW that he denounced his active CCIE status in 2008, don't you. I remember his post about that on IOS Hints and Tricks a while ago. Anyway, it would be great to hear from others like Jeff Doyle (he was a Juniper spokesman for many years if I remember right) or Wendell Odom. Bruce Caslow and Val Pavlichenko could be a great choice as well since they have provided a technical foundation to build Cisco 360 program for R/S track.

    And it is probably time to start numbering those runt packets of yours, they are getting out of order lately in my MP3 player. :-)

    Again, thanks very much for your great efforts in pushing those shows across. Keep it up!

    Cheers,
    Alexei Monastyrnyi

  • smccl

    Enjoyed this one a lot and would like to here more from Arista, as well as, other vendors like Extreme. The lesser known but with potentially differentiating technology that most might not know about. Basically, why would I buy you other than price.. how active are you in the standards communities like VEPA, DCB, TRILL, etc. How well do you integrate with other core data centre products like VMware, Xen, NOC software, arrays by Netapp|EMC and the like, etc. I would have enjoyed hearing more about Arista's vSwitch implementation for instance.

    Keeping it technical is important for me as an admin and listener since competitive podcasts have no shortage of marketing. It's mostly a given that your listeners known Cisco but what about the other guys. Thanks and enjoy the podcast.

  • http://google.com/profiles/philashman Phil Ashman

    Startups are always going to nimbler than the big companies and will bring to market better engineered product. However in the grand scheme of things the bigger faster boxes are aiming at a very small niche of the market. Nevertheless, it is these companies that can "hopefully" light a fire up the backside of the big dogs if they get too comfortable lying on the porch.

    On the technical side, one thing I was trying to understand was how they port their software between the different silicon they purchase from Dune, Broadcom etc…..I'm assuming there is a separate hardware abstraction layer to the OS for each CPU? I remember when Dave Cutler was helping build MS NT from the ground up, the HAL was what gave NT the ability to be more easily ported between the Alpha, MIPS and Intel architectures. So from year to year, as they purchase different silicon, they need to change the HAL, but I'm assuming the main kernel of the OS will layer on top and therefore be backwards compatible with the older CPUs. The openness of the OS sounds very refreshing!!!

    Not sure, but this might help avoid those situations where you buy a set of blades only to found out that some of the new features coming a year and half down the road require new blades and new silicon.

    • http://www.aristanetworks.com Douglas Gourlay

      We moved all of the device drivers for each pieve of silicon into user space from the kernel by moving PCI interrupts from the kernel to selected user space with a driver we wrote that works across all of our silicon vendors.

      Then each chipset driver becomes patch-able, re-start-able, etc. Also it allows us to load just the necessary silicon driver at boot-time after we detect the specific hardware platform we are booting on. This way we have a single image that can run on all of our switches and in a VM as well.

      Some work still to do on fleshing out our VM emulator, vEOS. But right now it boots, runs the whole O/S, and gives you CLI/shell/etc. I would like to get it to where it can emulate a specific device look/feel for syntax checking, and over time CP packet forwarding operations and such.

  • Anonymous

    I knew I had heard Doug talk about silicon and Innovation in the past…..

    Doug on Merchant Silicon:

    So does this mean that Doug owes the former CTO of Nortel an apology? I would be curious to know what made Doug change his point of view here (other than who writes his checks).

    http://blogs.cisco.com/datacenter/comments/on_mer

    Doug On Innovation:

    http://www.youtube.com/watch?v=xsByhardYZ4

    "Here today to introduce the Nexus 7000 series, and INNOVATION platform for Cisco, purpose built from a clean sheet deign for the data center of the future."

    "…..change the way we build datacenters forever….."

    This doesn't sound like the Pentium II datacenter switch you want us to believe now……

    • http://www.aristanetworks.com Douglas Gourlay

      Dear Anonymous,
      (it is always fun to write that…err, not.)

      I wrote about this eleven months ago or so, here is the post. http://www.douglasgourlay.com/blog/2009/10/on-mer

      I completely agree with you that taking a single time-slice view of someone's statements and contrasting to a more evolved opinion will show disparity. I would hopefully argue convincingly that this is growth, a broadened perspective, etc. Besides, I can say with absolute certainty that I wanted to invest in new silicon for the N7k and was told categorically that we were not to do so.

      dg

  • http://www.definethecloud.net Joe Onisick

    Greg,

    Playing devil's advocate here, would you have been as receptive to that coversation had it been a Cisco PM describing a new line of Cisco switches with the exact same technology/concepts?

    Joe

    • http://etherealmind.com Greg Ferro

      Yes, I think so. For example, I have spoken with Omar Sultan regarding FabricPath in a previous show to achieve the same enlightenment on how the vendor approaches challenges, and Cisco is using FabricPath to provide a superset of features.

      The difference is that Cisco is too big to change or create innovation, and startups have almost completely dependent on new technologies to differentiate from the larger players.

      I also think that Cisco is struggling to deliver what they have, and are not able to consider delivering new features. That's probably the result of moving development to India where things tend to slow down and become mired in bureaucracy.

  • makeupyourmind

    I read both articles and actually was at the competitor switching vendor when Doug started touting his custom silicon pitch vs merchant silicon. That competitor did both as I agree there is some custom silicone and some merchant. But to do a complete turnaround pitching merchant silicone due to a new employer is a bit over the edge in terms of credibility. I don't think this retraction would have taken place if he was still at Cisco.

    Custom silicone is always the secret sauce which creates strong differentiators. An all merchant silicone solution is useless after sometime once everybody else catches up buying the same pieces. Merchant silicon also provides no control over your destiny as components are discontinued causing a major supply chain headache. You are always catching up on features or spending time tweaking with replacements.

    So with the same analogy of merchant silicon now, if Tesla designs a sports car buying the same engine that Ferrari puts in, there is no guarantee it can do better since the entire system has to perform around it. Nothing like designing your own components. Eventually the merchant siliconers end up as wannabe, cheap, me-too roadkill products.

    When Doug is back at Cisco, I am sure he will eat crow all over again or perhaps the ghastly remains of the one he just ate. ;-)

  • http://www.aristanetworks.com Arista Fan

    Cisco has validated Arista and Doug’s points by going with merchant silicon on the Nexus 3000 (and Arista 7050 is still half the power!), clearly proving that all the Nexus 5000 family and all the great ASIC silicon could’nt achieve the latency they needed and they turned to Broadcom.
    Now if they only had modern software such as EOS to go along with it (as opposed to IOS, NXOS, IOS-XE, Cat OS) all derived from 30 year old IOS. Software foundations must change and have not for decades.

    Technology and opinions needs to keep up with the changes and time. You stay with old , you get old, and out-innovated Cisco! This is why the stock is taking a beating, gross margins down and sales flat. Cisco has not understood the market segmentation and migration to modern scalable two tier designs.

    Look forward to the Packet Pushers Webcast from Arista’s best technologists on May 20 2011

  • http://www.aristanetworks.com Douglas Gourlay

    I guess the announcement of the Nexus 3000 really is a nice validation of what Arista has been evangelizing. I just cannot see how any vendor, even Cisco, can keep up with the pace of innovation in the semiconductor market in their current multi-BU polarizing model. I you do not rev your silicon every 2 years and use advanced processes you will be behind, not just a little, but woefully.

    The port asic on the latest Nexus 7000 F1 linecard supports a whopping 2 ports of 10GbE, and doesn’t even do L3… brilliant.

    The port ASIC on the Arista 7050 supports L2 and L3 with 64 ports of 10GbE – same table sizes or larger than the Cisco Nexus 7000 F1 board from what I can see as well. Modern silicon architecture.

    But more important is the OS architecture. I don’t think closed systems with a CLI as the only interface are going to be the winners – especially ones that have an antiquated message-passing system that relies on tightly coupling state to application. A simple test of any OS: kill Spanning Tree or a Routing Protocol. Does your network flap? Do you rely on graceful restart delayed timer hacks? Do you have all sorts of protocol churn? If it does – get a new OS.

    dg

  • Pingback: Show 45 – Arista – EOS Network Software Architecture – Webinar

  • Mike B

    I don’t personally understand the merchant vs. custom debate these days. The issue used to be that generalized processors and ASIC’s weren’t fast enough or have the flexibility that was required of purpose built gear. However these days, Fulcrum, Broadcom, and the Mellanox’s of the world have merchant silicon that rivals anything that was custom spun. Even Cisco purchases a ton of Cavium “mechant” silicon for thing all across the product families.

    I am not saying that there is never a purpose to spin your own, but these days it’s hard to be pressed to make an argument that the performance,reliability and flexibility of the merchant silicon is worse than the custom spun stuff.

    For me the benefit of Arista is that they are small and agile. The Nexus came as a result of Cisco spinning the Nexus folks out and allowing them to innnovate and build a cool product and then absorbed them back into the fold.

    Every single commication I have had with Arista has left me with a positive impression from when we demo’d gear and had issues to honest communication about product plans and feature support.

    Most of the networks I am specifying now are Arista Core and TOR with Cisco Edge and WAN. (Arista doesn’t do POE or have 3560 equiv switches)

    Mike B