This has been a huge week in networking. Cisco finally launched the long-awaited Insieme effort to much fanfare. And in an attempt to preempt the Insieme launch, Arista scheduled their own announcements for a couple of days before. There are tons of articles about the technical details of what each launched. I have included just a couple here:
- SDNCentral: http://www.sdncentral.com/news/aristas-single-tier-spline-targets-enterprise/2013/11/
- Jim Duffy at Network World: http://www.networkworld.com/news/2013/110413-arista-275557.html
- Jim Duffy at Network World: http://www.networkworld.com/news/2013/110613-cisco-insieme-275666.html
- Enterprise Networking Planet: http://www.enterprisenetworkingplanet.com/datacenter/cisco-insieme-launches-new-application-centric-sdn-vision.html
The merits of the technical approaches will be debated by people far better versed in product nuances than I am. But the launches represent fairly divergent strategic views of the world. Wherever we see space between approaches, it means customers will have to make choices. And those choices create opportunity.
Admittedly confused by Arista
I would normally soft-peddle this one a bit, but frankly I am confused. Had this been a normally-planned launch, it would probably make more sense to me, but I don't understand the rush to get it out two days before the Insieme announcement. And I suspect that the decision will render what could be something pretty powerful architecturally-speaking into little more than a footnote in an otherwise news-heavy week.
The landscape this week is application-centric infrastructure. Knowing that, Arista basically announced a larger switch with still more price reductions. The mood in the industry (I would argue generally, but certainly at least this week) is innovation and, more specifically, on the software side. The Arista announcement, while always paying tribute to their remarkable flagship software EOS (and it IS remarkable), seems anachronistic to me.
Ahead of an impending IPO, Arista should be fighting to show that they have growth upside and healthy margins. Typically, innovation is a path to growth at higher margin. If you can create something that is fundamentally better than someone else's something, you can charge more. It is rare that you see an innovative company rotate so strongly to overall cost reductions. Their two major launches this year have both amounted to: bigger hardware, cheaper price per port. That seems completely counter to an innovation strategy. And as a long-term strategy, cost will asymptotically approach some margin threshold. Put differently, there is a short runway on this.
For customers who are particularly moved by price, Arista needs to be aware of the white box movement led ostensibly by Cumulus. The Credit Suisse report from a couple of weeks ago estimated that the TCO for a Cumulus ToR was 70% lower than a Cisco ToR. It would seem that Arista's price war ought to be aimed there. The worst possible outcome would be for Arista to be successful in convincing customers that networking is too expensive, only to watch them flock to the white box movement in response.
And all of this CapEx flies in the face of what is really driving movements like SDN, Network Virtualization, NFV, and DevOps: OpEx. This puts Arista in a bit of a strategic straddle. They are publicly talking price-per-port while also claiming TCO advantages tied to a right-sized, purpose built operating system in EOS. Striking this balance is going to be hard. The risk is that neither point lands, and customers go either to price (Cumulus) or OpEx (insert favorite SDN vendor here).
Where I get confused though is framing this around Cisco's launch. By pulling their activities in, they launched a bunch of hardware, which only serves to highlight that Arista is continuing to press hardware while Cisco is moving more into software. In doing so, the HW-SW divide gets accentuated, which reduces the impact of an architectural shift (the Spline architecture).
Insieme breaks the seal on Application-Centric Infrastructure
Aside from somewhat muddled messaging around a VERY complex topic, the Insieme launch had a couple of notable points beyond the stuff everyone will talk about. First, they subtly suggested that they will drop prices per 1GE/10GE ports. It was a subtle move, but it will essentially mute the Arista price advantage. This is going to force a redesign of the 7500 to improve COGS. I doubt this gets much play in the mainline press, but this means that Cisco is finally responding to the pricing pressure that Arista has exerted through rapid adoption of merchant silicon.
The reason Cisco can drop prices is that they are using merchant silicon as well. In an equally subtle comment, Soni referred to a "Merchant+" strategy (not sure how they spell it). Basically, Cisco is bringing merchant silicon alongside their own ASICs. This is interesting as it will impact merchant silicon pricing for everyone as volumes go up. It also means that the path for Intel will get even harder as more companies converge on a narrow set of merchant silicon. It will be interesting to see how this impacts the Broadcom/Intel landscape.
But aside from these subtle points, the Insieme launch was interesting because of the obvious pivot to applications. The storyline of the day was that the applications come first in an application-centric environment. That said, there was still a lot of discussion about edge policy and network provisioning, and a surprisingly little amount of focus on the DevOps frameworks required to orchestrate an application-centric infrastructure
It could be that DevOps frameworks are unnecessary because the system is so tightly integrated that everything is taken care of. If that is the case, users will be pleased that much of their pain has gone away. Of course, it will have been replaced with a deeper reliance on homogeneity, which might create tension in some environments.
Ultimately, workload orchestration appears to be a much more full-frontal assault on VMWare than any other networking player. While the temptation is to describe this as a Cisco/Nicira battle, I suspect we will find in time that this goes well beyond overlays and NSX. The real question is not whether Cisco and VMWare square off though; that outcome is predetermined. The real question is who else enters the fray. Once you start talking about applications first, application companies come into view. The networking world might want to pay more attention to Oracle than Arista in that regard. It could be where the real fireworks end up happening.