To answer his question, yes virginia, we (people that use MySQL) use replication a lot. Especially at mid-grade levels of scaling where custom sharding and replication middleware hasn't necessitated itself yet but vertical scaling is out of the question.
I'd say the bigger issue with Postgres now is that middleware like http://code.google.com/p/vitess/ for it hasn't been deployed in the large yet. Most Postgres scaling anecdotes I've heard were:
"Well we put 64 gb of ram in the server and installed a RAID array of SSDs and stopped writing dumb unindexed queries."
Well that's just dandy, but what if my indices don't even fit in the ram of a single machine?
> I'd say the bigger issue with Postgres now is that middleware like http://code.google.com/p/vitess/ for it hasn't been deployed in the large yet.
That simply isn't true. Take a look at PgPool (http://www.pgpool.net/mediawiki/index.php/Main_Page) - it provides many of the same core features as Vitess, goes beyond to provide replication and flexible sharding/partitioning and is a good number of years more mature than Vitess.
PL/Proxy is similar middleware (though implemented as stored procedures) developed by and used at Skype for massive Pg partitioning.
There are dozens more packages like this at Pgfoundry (http://pgfoundry.org/). What specifically do you feel is missing?
You get the feeling that Postgres needs some marketing. It's a great piece of software but if people can't easily find out if it's been used at massive scale they aren't going to trust it.
Like it or not most deployments of MySQL are probably because people can easily hear about it being used for massive databases. The decision isn't based on technical merits (or lack of).
I'd say the bigger issue with Postgres now is that middleware like http://code.google.com/p/vitess/ for it hasn't been deployed in the large yet. Most Postgres scaling anecdotes I've heard were:
"Well we put 64 gb of ram in the server and installed a RAID array of SSDs and stopped writing dumb unindexed queries."
Well that's just dandy, but what if my indices don't even fit in the ram of a single machine?