Google use MySQL— see: http://www.mysql.com/customers/view/?id=555
and in particular: http://www.bytebot.net/blog/archives...ysql-at-google

It would not be practical to spread individual tables of a relational database across different servers. The typical approach is to use replication of the whole database. One web application might have one master database server and three slave servers (for example): read operations can be made against any server, while write operations are made against the master which then propagates the changes to the slaves. The master is responsible for keeping all of the slaves in sync. This approach works for most high traffic applications because read operations are typically much more frequent than writes.