A Look Inside Wikipedia’s Infrastructure (2008)

Despite being one of the world’s busiest sites, Wikipedia runs on fewer than 300 servers from a single data center in Tampa.

As a non-profit running one of the world’s busiest web destinations, Wikipedia provides an unusual case study of a high-performance site. In an era when Google and Microsoft can spend a half-billion dollars on one of their global data center projects, Wikipedia runs on fewer than 300 servers from a single data center in Tampa, Fla. It also has servers in Amsterdam at the AMS-IX peering exchange.

The engineers on the Wikipedia team may not take themselves too seriously, but they are serious about performance. That’s in keeping with Wikipedia’s guiding principles, which emphasize community over commerce (the site runs no ads) and getting excellent mileage out of its donations. Wikipedia maintains high 99-percent availability, and the usage data for Wikipedia includes some mind-boggling numbers.

  •  50,000 http requests per second
  •  80,000 SQL queries per second
  •  7 million registered users
  •  18 million page objects in the English version
  •  250 million page links
  •  220 million revisions
  •  1.5 terabytes of compressed data

Wikipedia is powered by the MediaWiki software, which was originally written to run Wikipedia and is now an open source project. MediaWiki uses PHP running on a MySQL database. Mituzas said MySQL instances range from 200 to 300 gigabytes. In addition to Squid, Wikipedia uses Memcached and the Linux Virtual Server load balancer. Wikipedia also uses database sharding to set up master-slave relationships between databases.

Additional technical details on Wikipedia’s infrastructure is available in 2007 presentations by Mituzas and WikiMedia’s Mark Bergsma.

Mituzas summed up his view of Wikipedia’s operations in a blog post about his Velocity presentation: “As I see it, in such context Wikipedia is more interesting as a case of operations underdog – non-profit lean budgets, brave approaches in infrastructure, conservative feature development, and lots of cheating and cheap tricks (caching! caching! caching!).”