- 100+ servers
- 5+ million accounts, about half active
- 50M+ dynamic page views/day
The document explains all the steps over five years from one shared server to the current infrastructure, with users spread around database clusters, two proxy/load-balancing layers (at the IP and HTTP levels), and caching handled by memcached (also used by Slashdot, Wikipedia, Meetup – handles 100,000 queries/second at peaks with a 90-93% hit rate). It all started as a "college hobby project"!
Update (somewhat related): Adam Bosworth: Where have all the good databases gone.
05/30/07 update: A bunch of presentations on scaling websites: twitter, Flickr, Bloglines, Vox and more., Inside Myspace.com, What powers Curse Gaming.
08/24/07 update: High scalability blog.