Separate database and web servers

This is an archived history discussion from March and May 2003, when there was a single server hosting the site. Also see Wikimedia cache strategy evolution during 2003 for coverage of that topic during March through December 2003. For current information see Wikimedia servers and cache strategy. Jamesday 01:34, 26 Aug 2004 (UTC)


Everybody in the universe recommends running database and web services from separate machines.

Advantages:

  • They're not fighting each other for dominance of one machine -- cpu and memory are precious resources, and they both want it at the same time. Not to mention the disk...
  • From there, a "simple" step to set up some minimal failover / round-robin / whatever.

Disadvantages:

  • We don't have a second machine yet. :)

We might also want a separate mail server to ensure the mailing lists don't go down when the main server(s) is/are loaded/crashed/eaten by llamas.

Status as of 2003-05-18

edit

The English wikipedia web front-end has been moved to a new separate server. The new machine is actually slower (old parts) but is only doing apache, occasionally tex, and some outgoing sendmail. With some limited server-side caching of pages enabled, load average is down in the 1-3 range on both servers and response time is pretty good all day long.

Success so far!

what's 1-3 in terms of what it was before?
Nearly a factor of ten lower for peak hours. --Brion VIBBER