This week’s big networking news is that Facebook announced that they have built their own switch (codenamed “Wedge”) and the corresponding networking software on top (“FBoss”). While the specific technical details are still coming out, I’m more interested in why they did it. Facebook is an advertising company on steroids, not a networking vendor. They are not going to make money selling this switch to end customers, so why invest the engineering effort and time to build their own networking stack?
The answer is for the same reason Google builds their own switches: it is a tacit assertion that the networking infrastructure provided by traditional networking vendors simply does not work for hyperscale networks. Yes, one result is lower-cost switches, but that’s almost the least important aspect. The primary reason is agility: if you want to deploy new applications and services at the rate that Facebook, Google, et al. do, you need programmatic control of the entire network stack to automatically deploy and test new services. This is critically important because fresh applications, faster time to market, and rapid innovation are the lifeblood of modern companies -- especially hyperscale. The second reason is operational efficiencies. Working box-by-box at a traditional CLI, there are simply not enough CCIE’s in the world to keep up with the rate of change in hyperscale networks -- not to mention the cost of employing them. It is a huge resource investment to design, test, and deploy custom hardware and an even bigger one to build the corresponding software stack. But Facebook, Google, and--by rumor--others are all doing this because their traditional network alternatives simply do not work for them.
While pundits might dismiss this trend as for hyperscale only and thus not applicable to enterprise, service provider, or other companies, they miss the most important point: the Wedge causes trickle down innovation for everyone. The merchant silicon and white box networking movements are just now making headway into traditional enterprise and service providers because the hyperscale folks deployed them five years ago. Even further back, traditional companies were debating Solaris versus Windows NT (pause for nostalgia… aah; done) until the hyperscale folks showed the way with Linux. It is fair to say that the hyperscale providers have consistently lead the entire data center movement and that their hard work has eventually trickled down to benefit us all. Facebook’s Wedge and FBoss are just the newest examples of that innovation. When you combine these innovations with other projects from Open Compute including ONIE and our own Open Network Linux, the path to tomorrow’s data center network becomes clear. In other words, just like bare metal switches and Linux, this technology is coming to a data center near you -- it’s not a matter of “if” but “when”.
Original Facebook Press Release: Introducing “Wedge” and “FBOSS,” the next steps toward a disaggregated network
--Rob Sherwood, Big Switch CTO