Blog Posts tagged ‘architecture’

With a proud tear in the eye but a heart full of excitement, Anchor announces a parting of ways with Github as their hosting management provider.

Since we first got our hands into their systems about two years ago, Github has grown. A lot. For our sysadmins it’s been a process of gradual but continuous change, and it’s an eye opener to step back and take in the sheer scale of it all. We’d like to share that with you, and also talk about what the changes mean at a higher level.

Taking stock

Late in 2009, our project lead Matt Palmer penned a few technical posts about the size of the new architecture. Comprising some 17 physical servers and a dozen or so VMs, this was a huge upgrade for the whole architecture at the time.

Today Github has 48 pieces of hardware, and about twice as many VMs, with more coming online all the time. There’s about five times as much repo data being hosted, too. And that’s after they figured out how to dedupe all those copies of XBMC that people keep forking.

We’d like to think that this is a testament to the solidarity of our design. It’s not flawless by any means, but it’s proven to be very robust, manageable, and scalable enough to keep pace with Github’s growth.

Self sufficiency

So where to now? A recent count says Github has over a million users, about an eight-fold increase, and their team has grown similarly from five to 47 54 55 56! (for the sake of comparison, Anchor has gone from 18 to 30 staff – our sysadmins are denser).

Github now employs three full-time sysadmins in addition to their customer-facing technical support team, which means around-the-clock coverage. Anchor’s management services let customers leverage our size to get a high level of service at an affordable cost, but as a massively technical company working on a large scale, it makes even more sense for Github to bring it all in-house.

This is something that’s been in the pipeline for a while, and it’s been a pleasantly painless process of handing over the reins. Github’s sysadmins now handle the regular stream of diskspace alerts and other blips on the radar, and new VM services that we know nothing about appear on a regular basis (seriously guys, what’s a “cheddar”?), all thanks to the fully-automated build systems that we developed when the project started.

We wish the Github crew good luck and smooth sailing for the future. They’ve been a great bunch to work with and we look forward to more Octocat adventures.