Facebook, in its quest to serve its more than 1 billion users, has redesigned its network architecture using a fabric aimed at keeping the site running even the social network continues to scale. The architecture change will also encompass new networking technology, such as the commodity switches that Facebook announced at the Open Compute Project.
Jay Parikh, the VP Infrastructure Engineering, at Facebook and Najam Ahmad, director, technical operations, infrastructure at Facebook explained the new networking fabric onstage Wednesday at the GigaOM Structure conference in San Francisco.
“The needs of our application are such that everything is virtually real-time and connected,” Parikh said. “So our traffic going across from machine to machine far exceeds the ratio of machine to user.”
The company has moved to a fabric-based network architecture using 10 gigabit per second connections, not only because its inbound and outbound traffic is huge and growing, but because the traffic inside its data center is 1,000 times larger that the traffic sent in and out of its data center.
Facebook is moving from what it calls a four-post network that is comprised of several layers of switches. It developed that design in response to previous network failures, where one switch going down would call a service outages.
The fabric architecture is also an opportunity for Facebook to disaggregate and open up the network — which is one of the last closed, black box parts of web infrastructure. This is where the Open Compute Switch design that debuted in May will come into play. Ahmad noted that moving to this kind of architecture will allow Facebook to focus much more on managing network via software, instead of via hardware.
Check out the rest of our Structure 2013 live coverage here, and a video embed of the session follows below:
A transcription of the video follows on the next page