Can Facebook’s different take on SDN scale down from hyper?
Mon 23 Jun 2014
In his latest exclusive blog for The Stack, Chris Swan reviews Facebook’s announcement of its new switch, Wedge, yet finds the software the more interesting part. Is this hyperscale approach to software defined networking (SDN) likely to have a broader appeal?
![Chris_Swan21[1]](https://thestack.com/wp-content/uploads/2015/08/Chris_Swan211.jpg)
Facebook have announced their own switch design, codenamed ‘Wedge’, saying that it’s already being tested in their production network.
In many ways the switch is unremarkable; it uses the same Broadcom Trident II merchant silicon ASIC that most other high end ‘white box’ top of rack (TOR) switches use, and it uses Linux on a commodity compute platform as its operating system. The physical packaging and power systems fit in with the server designs that Facebook has previously donated to the Open Compute Project (OCP), making it almost the purest expression of networking equipment as a commodity.
Where things get interesting is with the ‘FBOSS’ software, which Facebook describes as offering ‘a hybrid of distributed and centralized control’. This moves away from the Open Network Foundation (ONF) definition of Software-Defined Networking (SDN) being the ‘the physical separation of the network control plane from the forwarding plane, and where a control plane controls several devices’, and there is no mention of OpenFlow in relation to Wedge.
What Facebook have done at a physical level is put the control plane (based on an OCP ‘Group Hug’ microserver) right next to the forwarding plane. Since that control plane is just a Linux server it can do a bunch of stuff autonomously, but it can also be treated as a node in a large scale distributed architecture. By creating a Thrift-based abstraction layer on top of the forwarding plane APIs, Facebook is now able to treat network configuration as just another set of services, claiming that ‘with “FBOSS,” all our infrastructure software engineers instantly become network engineers’.
‘FBOSS’ and Thrift are fine within the confines of the Unicorn ranches of Facebook’s own infrastructure, but it remains to be seen how far that approach will penetrate outside of Facebook. Whilst there’s a commitment to donating the hardware designs to OCP, there is, perhaps, some work to be done on having a software stack that’s deployable for those that don’t live and breathe Thrift.
What Facebook is doing here has some echoes of Google’s Andromeda SDN/NFV architecture, but the discipline of hyperscale operators and how they manage their estate doesn’t necessarily scale down to other users. The bottom line here could be that whilst Google and Facebook want to carefully shape how network traffic moves across their data centres to wring out every last cent of efficiency in edge utilisation, most others just want cheap and reliable networking equipment.
Chris Swan is CTO of CohesiveFT
