Video: Arista Networks and technology talk

By | November 2, 2009

While at SNW Europe I had the opportunity to visit the Arista Networks booth with my video camera in hand, and the guys on booth duty were good enough to allow me to shoot some footage while we talked about their products.

Enjoy the video and then feel free to comment on my thoughts below –

First up, Arista is start-up networking company, and like all networking companies…….. they have Cisco in their crosshairs.

Although low cost seems to be their trump card, to be fair to Arista they are also touting performance as another key advantage.  Better performance, literally at a fraction of the Cisco price – if the guys at the booth are to be believed.  However, as I mention in a previous post, there is more to value than price.  It may turn out to be very short sighted to jump all over short term cap-ex savings at the true expense of the long term good.

However, the current down economy is probably creating ‘too good to pass-up’ opportunities for companies like Arista, who come in at a value price-point- most companies being forced to think more and more about cost, thus creating windows of opportunity for those coming in at significantly lower price-points.

However, when at the booth, the Arista guys did a great job ensuring that their products didn’t have that ‘value’ feel.  Despite the fact that they are apparently way cheaper than Cisco, they weren’t over-stressing that point and wanted to talk a lot about performance.

Technology and performance overview

So a quick overview of what Arista have on the table from a product and performance point of view….

Basically they are offering high-speed low latency 10Gbps Ethernet products.  Oh and apparently they are cheap…cheap enough for my garage??.

However, if you listened closely to the video, they only have DCB/CEE “on the roadmap”.  This, in my opinion is a huge shortfall in the current shipping products.  As each day goes by I am becoming more and more of the opinion that to be a true data center switch, you need CEE and FCoE.  Yes, 10Gbps without the goodness of CEE and FCoE might suffice for now, however, it will be found wanting in the very near future.  More on this later though….

I am impressed at the published latencies and port count as well as the dual hot-swap power supplies and reversible airflow (to change the airflow from front-to-rear, to rear-to-front requires swapping out the fans).  However, my gut reaction is that this is almost a gimmick in comparison to features like CEE and FCoE. Sorry that I keep coming back to this but I can’t stress it enough.
 

NOTE:  In the interest of fairness, should point out that when at HP Colorado Springs recently I was knocked back when I was told that one of the hardest things to engineer for new servers and blade chassis is the power supply units (PSU).  So I am careful not to totally dismiss the reversible airflow and redundant hot plug PSUs as totally insignificant.

7100 Series Data Center Switches

The 10Gbps 7100 series SFP+ based switches come in either 24 or 48 port-count models and you can choose between a non-oversubscribed and oversubscribed models.  You can also buy similar 10GBASE-T models which take standard RJ45 connectors and work with existing Cat 5/6/… cables.

When first at the booth (without my camera) I raised my eyebrows when introduced to the 10GBASE-T model.  As I raised my eyebrows the guy giving me the tour (different guy to the one on the video) pre-empted me and said “your are going to say ‘Wow’ aren’t you. But I had to admit that I was actually going to say “Why”…..

My initial reaction was why would anybody want to do that from a power cooling and latency standpoint.  Essentially SFP+ up to 10 metres has people talking about ~0.1W nominal power draw whereas 10GBASE-T RJ45 cables are closer to 4W at approx 10metres, but obviously offers greater distances at the expense of more power.  And that’s not to mention the inferior transceiver based latency that 10GBASE-T has compared to SFP+. 

Obviously a 10GBASE-T model may make the jump to 10Gbps Ethernet simpler for some people, I personally believe this approach is a false economy of sorts and hedges up the route to the full benefits of 10Gbps, and greater, based on CEE. 

Pass-Through module

They also had a 10Gbps Pass-through module on display.  This pass-through module has 14 internal and 14 external linespeed connection, with no oversubscription and is designed for the IBM BladeCenter H.  The module only has 4 physical ports on the external faceplate and achieves the 14 external port count via 10GBASE-CR QuadSFP to SFP+ splitter cables.  This is the copper twinax 4-way splitter cable seen in the video.  The module and is apparently very competitively priced compared to competitive offerings from the likes of BLADE Network Technologies.  Although features and functionality might be less than that of the BLADE Networks offering, apparently performance is not.

Nice, but no game changers

So, a lot of talk about feds and speeds.  Interesting for some, but talking feeds and speeds, so Im told, is sooooo last year.  EMC and HDS used to do it all the time until analysts and customers told them they didn’t care.  Feeds and speeds are fine, but not everything (at least not to the vast majority).

So for feeds and speeds the Arista kit seems to be very good and very competitively priced. If latency and performance are critical for you then Arista are definitely worth a look.

However, away from feeds and speeds was where there were signs of weakness. 

In my opinion, for a networking product to be a true Data Center product it has to play in the converged networking space (IP, FC, RDMA..).  At the moment, Arista products do not.

First, there is no Converged Enhanced Ethernet (CEE/DCB).  For me this is vital to the future of Data Center networking.  I know they guy on the video said that Arista doesn’t do anything that is not standards based, alluding to DCB/CEE standards not yet being fully ratified.  However, Im willing to bet they do jumbo frames, and I’d like to see them produce the IEEE 802 standards documents for jumbo frames πŸ˜‰

Sorry, that was a cheap shot, but I think comments like these are poor and will not help you out in the future when you are playing catch up.

Second, these are pure Ethernet products.  No Fibre Channel ports.  This is understandable as they have no FC legacy or expertise. 

In the very short term, while people use the edge in approach for implementing a unified fabric, this means that Arista products will not be an option.  But more importantly, beyond this, once end-to-end FCoE becomes mainstream, where will this leave Arista?  Having no FC expertise that I’m aware of will make implementing FCoE (FCF and all the other FCoE related functions) in their products very time consuming and difficult.  They will likely have to acquire a company that has FCoE or hire some engineers who know it, and hire them fast!

Although Data Center networking is converging on Ethernet, it is not converging on the Ethernet that is widely deployed today.  In fact, the Ethernet that it is converging on (CEE) bares less resemblance to todays 1Gbps lossy Ethernet than most realise.  So just because you are an Ethernet company today, and Data Center networking is convergi ng on Ethernet, does NOT automatically mean you are well placed for the future.

Looking to the future

So…….. normally, it is the large companies, those with the large market share and legacy architectures that struggle to change with the times (think Cisco).  The newer and more agile companies (think Arista) often beat them out of the blocks and are tearing away around the first bend before the likes of Cisco are even out of the blocks.  Not so this time! 

Cisco are actually heading around the first corner with early deployments of CEE and FCoE on the new and expanding Nexus platform and NX-OS.

Far from being caught flat-footed or even found trying to protect their existing markets, Cisco have made bold moves into the Blade Server market with UCS as well coming to market with the new and ever expanding Nexus platform and NX-OS.  If anything Cisco are already heading around the first bend with early deployments of CEE and FCoE.

Its one thing to make a better or faster “switch” than Cisco.  Buts altogether another thing to build better solutions.  Right now Hypervisors are driving change and shaping the future.  Addressing the challenges that Hyperviors bring are vital.  From my perspective Im not seeing a lot from Arista to help here πŸ™

Management Big Hitters

Before finishing, it is worth quickly mentioning the very impressive list of networking legends that is heading up the Arista management team.  A quick scan of the management team listed on the Arista website reads like a who’s who of Data Center networking and is nicely capped off with Andy Bechtolsheim who has pedigree when it comes to picking winners.  These guys know their networking, oh ……. and they know the competition πŸ˜‰  So don’t write them off just because of what I say :-S

Finally

With the lack of a networking group at IBM, it would be remiss of me not to suggest that Arista may be a future acquisition target for Big Blue πŸ˜€  I had to say it.

I'd love to have been more impressed with Arista.  Who knows, may be they have a game changer up their sleeve.

Nigel

You can follow me on Twitter where I talk about storage technologies (@nigelpoulton) and I am also available for hire as a consultant.

3 thoughts on “Video: Arista Networks and technology talk

  1. Douglas Gourlay

    Nigel,

    Thank you for the post and write-up on us. 

    As you can imagine, I have a bit of an opinion of CEE/DCE/802.xyyy/etc.  The Arista products have hardware support for the lossless transmission of Ethernet frames via the Priority Flow Control mechanisms.  We do have to enable the software portion of this and as you accurately indicate we, "have this on our roadmap."  The funny thing though is while I concur with you that there is absolutely significant value to be delivered with FCoE that is not our current target in the market.  Where we have been extremely successful so far is in the low-latency, high performance parts of the network.  This usually equates to financial risk analysis, gene modeling, high frequency trading, market data feeds, fluid modeling, and quantum chromodynamics applications.  Net-net:  the areas where IT is delivering the maximum business value as measured by the lines of business.

    FCoE is more on the cost optimization side of general IT.  FCoE saves cabling, and thus, hopefully, saves money.  For the portion of the servers that require block-based connectivity where iSCSI does not meet the performance requirements, within the scope/scale of a large layer-2 network FCoE will be a solid solution.  Right now there are very few FCoE capable switches on the market, most of those deliver FCoE trading off performance, cost, latency, and Layer-3 routing.  We felt that in our target markets FCoE was not a top priority and have had this echoed frequently by our many customers.

    Will we deliver it?  We will start with the lossless extensions to the Ethernet standards so we can safely carry the payloads and then if and hopefully when the FCoE market develops into a reality we will be right-timed with the control plane features necessary for the deployments we are targeting.

    I don’t think FCoE is a ‘must have’ for where Arista switches are being deployed successfully today.  I don’t think that 80% of the servers in the world require or need block based storage access via FC, and I think for those that do there are a variety of technologies that are competing for this sub-20% of server attach. 

    On another note- I do appreciate you identifying with the physical hardware design we have as being very optimized for the data center environment – I think we are quite unique in that and the open Linux implementation of our software that allows 3rd party applications to be installed.

    dg

  2. Nigel

    Thanks for the comment Douglas,
    Clearly yourself and the management team at Arista have enough networking peigree to know a winner when you see one and have greater insight into roadmaps than I do.
    But for me…. CEE hits the sweetspot of many of todays demands and emerging growth areas – cloud, convergence, green….  No CEE = No cloud, no convergence, no green – IMO.  I also see CEE happening faster than most people expected.
    While concentrating on bringing high performance non-CEE 10Gbps solutions might be addressing a gap, it seems a little niche and short-term – although I suppose it gets your foot in the door.  But if you can couple that price and performance with CEE and possibly FCoE then that is something that could cause some ripples.  Thats altogether harder though.
    Im interested in EOS though. I blogged a while ago that I felt Enginuity is the crown jewel of EMC Symmetrix and tweeted the other day that IOS/NX-OS was/is the crown jewel of Cisco.  So I agree that that the real value can be in the SW rather than the HW. The booth guys at SNW Europe made a big play about EOS but mainly around its reliability and self-healing capabilities rather than any value-add it brings to the table.  This was also in line with their comments re the hardware – "faster, cheaper, faster, cheaper…" but not so much about value-add or inteligence.   
    Some background as to where Im coming from.  With data center networking I (personally) think that the value going forward is not just in faster faster faster, but in inteligence like CEE – creating a new platform to build next gen services on….  Same goes for software like EOS.  Yes its important that its self healing etc but so is NX-OS. What does EOS bring that is more than that? We all know that everybody writes apps for Windows despite it having a terrible track record for reliability.  Sadly quite often the best/most reliable tech doesn’t always win πŸ™ 
    So what is so good about EOS and What does EOS do that puts it out there ahead of NX-OS? I didn’t get that from the booth guys – although in their defence I was pushed for time.
    Appreciate any thoughts you might have on this.
    Nigel

  3. Etherealmind

    Nigel

    DCB isn’t going to change how a Server person sees the network. While it is a fundamental change to how networking "IS", it doesn’t really change the network insofar as a server perceives connections. Given that FCoE still isn’t ready for commercial use, and most servers aren’t capable, then declaiming the lack of unfinished standards, and unused vapourware isn’t really viable. EOS, in fact any other network vendor, will support DCB when the time comes. Don’t believe the hype from Cisco, DCB won’t be here until the middle of next year at the earliest

    greg

Leave a Reply

Your email address will not be published. Required fields are marked *


*

You can add images to your comment by clicking here.