On 05/03/07, Tim Starling <tstarling(a)wikimedia.org> wrote:
> I said "In Tampa there's lots of big
disks, these go to three big
> database servers - one for English Wikipedia, two for the other 200+
> projects - and these go to hundreds of webservers and caches and
> proxies in Tampa, Paris, Amsterdam and Seoul."
> Is that about right as a one-sentence description of our setup?
It's close. There are three master database
servers, but they're not any
bigger than the 12 or so slave database servers, so it might be more
accurate to just say that we have 15 database servers. Caches aren't on
dedicated servers, so I would say "webservers and caching proxies" rather
than "webservers and caches and proxies", leaving the role of the various
backend caches unsaid. We no longer have any active servers in Paris.
"A bunch of disks serving three master database servers - one for
English Wikipedia, two for the other 200+ projects - going to hundreds
of webservers and proxy servers around the world, in Tampa, Amsterdam
and Seoul."
The DB split is useful info because en:wp is such a huge bugger.
A tier you didn't mention, which the journalist
may or may not find
interesting, is that our load balancing frontend is LVS-DR on commodity
servers, with geographic DNS to balance between clusters.
He was going "wow" about a non-profit running a top-10 site on no
money, and it was the best not-quite-technical answer I could give him
that I thought would be reasonably robust against a journalistic game
of telephone :-)
- d.