MirrorBrain is an open source framework to run a content delivery network  using mirror servers. It solves a challenge that many popular open source  projects face -a flood of download requests, often magnitudes more than any  single site could practically handle. Some very popular Open Source projects  rely on MirrorBrain. 

The central (and probably the most obvious) part is a "download redirector"  which automatically redirects requests from web browsers or download programs to  a mirror server near them. 

For mirror choice and load-balancing for the mirrors, MirrorBrain uses  geolocation and global routing data. The selection algorithm can be finely  tuned, but is designed to have sensible defaults and work mostly automatically.  In addition to network topography and country, the approximated geographical  distance of mirrors is also taken into account. Mirror servers are monitored and  scanned for files. 

Furthermore, MirrorBrain can automatically keep cryptohashes of all files and  can serve these on request. It can even generate Metalinks (RFC5854) and  Torrents in realtime. (The Torrents are uniquely useful in that they contain  webseeds - webseeds that work, specific to the client who requested  them!) 

How much traffic can MirrorBrain handle? Lots of, and you don't need a room  full of servers. Let's look at an example: handling all downloads of  OpenOffice.org, a busy site, could be called boring. A simple 512MB box  can do this. Handling all openSUSE.org traffic (20-40 millions hits per day) is  still relaxed (commodity hardware, server load: zero point something). That is  comparable to all downloads from Apple's App Store worldwide (200 per  second, as reported in autumn 2010). Do you need more?