Smooth like btrfs: Inside Facebook’s Linux-powered infrastructure

Btrfs creator showcases Facebook’s open-source storage.

Facebook engineer Chris Mason is unequivocal about the primacy of Linux in Facebook’s storage infrastructure.

“If it runs on a computer, and it’s storing important data,” he said, “it’s running Linux.”

Mason, speaking at the Linux Enterprise End-User Summit on Monday in New York, joined Facebook just six months ago in order to spearhead the social network’s move to btrfs (usually pronounced “butter eff ess.”), the Linux-based file system that he created in 2008 while working at Oracle.

+ALSO ON NETWORK WORLD: Why Facebook's SDN switch won't affect Cisco's customers | OpenStack chair: Linux at the cutting edge of the cloud +

It’s a massive undertaking, to say the least. Mason cited some statistics from Facebook’s Lookback video project that help underline the scale of many Facebook plans – 11 petabytes of storage used, a peak outgoing bandwidth rate of 450Gbps.

What’s more, the pace of work at the company is often breakneck – the code behind the main website itself is updated at least twice a day, according to Mason. “Projects there move very fast,” he said. “Very soon after I joined, they had a selection of machines carved out to run btrfs.”

Accordingly, the ability to automate much of the management work involved in the storage infrastructure is crucial, Mason said.

“If it can’t be automated, we’re not interested,” he said. “We have so many machines and so much infrastructure that we have to be able to sit at a desk and hit a button and have it go.”

Another reason Mason and his team have been able to keep Facebook’s massive storage infrastructure from sputtering during the transition is the use of a stable Linux kernel, without too many in-house patches applied to the official stable branch.

“We’ve never had real problems caused by stable release,” he said. “Being so close to the mainline kernel allows us to add new features.” Most Linux machines used at Facebook run on version 3.10 of the Kernel or above, and the internal distribution is based on CentOS.

That’s not to say that the project has been hiccup-free, of course.

“It turns out that there are little things that happen when you move from one kernel to another,” Mason said.

Btrfs is what’s known as a copy-on-write file system, which means that it operates in a substantially different way than traditional filesystems. Mason told in 2008 that the idea is to let Linux storage scale more efficiently across larger volumes. For many, the Facebook adoption will look like vindication.

Facebook has a history of headline-grabbing moves into open-source technology – the company’s Open Compute Project, created to help spur open-source data center hardware design, launched in 2011, and Facebook contributes code to a host of open-source software projects.

ADDENDUM: Mason contacted Network World after this story was published to clarify that the entirety of Facebook’s infrastructure isn’t slated for a move to btrfs.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2014 IDG Communications, Inc.

SD-WAN buyers guide: Key questions to ask vendors (and yourself)