After getting very, very frustrated with MySQL replication (and seeing the hoops people have to hop through to get replication to scale), I've decided I'm not that interested in pursuing designs that put a database at the center of the universe.
Reading through
Tim O'Reilly's database war stories helped cement this idea for me. Especially interesting was the
bloglines and memorandum interview and the
facinating topix.net architecture (found from a comment).
Another datapoint is
LiveJournal. LiveJournal uses MySQL for storing posts, but they've
a lot of
work to minimize the use of MySQL as they've grown. We hear a lot about Google's
BigTable and
GoogleFS all the time. But we can't use them because the source isn't available. LiveJournal has written similar systems in
MogileFS and
memcached and, unlike Google's work, these are available to everyone.
I'm going to start looking at how I can build systems around these two utilities and to minimize or eliminate my dependence on a SQL server.