Mark Williamson wrote:
It would be especially neat if we could get at least a
little bit of
server space somewhere in Africa, Australia, and South America.
Having said that, it would also be easy/nice to get server space in
California or somewhere else in the Western US (Google would be
California, would it not?)
I think that it's always best to have everything distributed over
multiple locations in case of a disaster, natural or otherwise (this
way, we won't have to worry about Wikipedia in hurricanes, and if one
location gets broken into/burns down/explodes/disappears mysteriously
in a puff of steam, Wikipedia will not have problems because of it).
Domas has been advocating further expansion of the Florida colo, and I
can see his point. Even a 10ms RTT delay (say from Tampa to Miami) would
have a large negative impact on performance if we tried to have a single
memcached namespace stretching over both locations. Our idea for using
the European and Asian data centres is to move entire wikis. Each wiki
needs to be served by a master database server and a cluster of apaches
in close proximity to each other.
Because of this need for clustering, I don't think we can reasonably
expect to be able to cope with the Tampa colo disappearing without some
reduction in performance. It's not worth having a slow wiki 365 days a
year just so that we can recover quickly after a one in a million chance
event. However, it's reasonable to expect that there will be no loss of
data, and that we'll be able to get read-only service up fairly quickly.
Squid servers are a different story, they can be spread all over the
world. There's no preference for clumping them together except for
convenience of administration and maintenance. Performance-wise, it's
better if they're near the users. That's why we've set up a page for
organisations that wish to volunteer to host individual squids:
-- Tim Starling