Here's the current rundown:
In California: * larousse is running a squid proxy to the web servers in Florida * gunther is making a backup dump * ursula is not doing much of anything * geoffrin, pliny and susan are very sad, wishing someone would take away their pain
In Florida: * suda is running a master database * zwinger is serving mail and is the main fileserver for the other machines * bart, bayle, moreri, and vincent are running identically-configured apaches. They share their work directories by NFS, so uploads etc should remain quite in sync. The squid picks between then round-robin. Right now only bart is running a memcached, but it should work to split the cache over them all, and they'lll ask each other for the bits they don't have. * browne and coronelli have test squid installations that Gabriel is setting up, testing failover systems etc. * isidore isn't doing anything just yet but should soon start syncing over a backup of zwinger's stuff.
So far things seem mostly in order. Zwinger's occasionally freezing up for a couple seconds, may be something that does too much IO. Trying to track this down.
Eventually we'll want to move the squids over to florida, which means changing DNS. This'll cut out the 80ms round-trip across the continent for every hit, as well as give us bigger beefier caches.
-- brion vibber (brion @ pobox.com)
On Thu, 12 Feb 2004 06:08:21 -0800, Brion Vibber wrote:
- browne and coronelli have test squid installations that Gabriel is
setting up, testing failover systems etc.
These two systems are configured as siblings, they share their storage and ram as one big cache without duplicating information.
We've done some load balancing tests to find the optimal configuration, particularly in case of an imbalance between the Apaches. Dynamic load balancing that takes the individual load of each Apache into account isn't set up yet, given the high numbers of requests this might not be necessary anyway as the differences between the requests tend to even out. In case one Apache is overloaded or down the requests are passed on to other machines, so fail-over is automatic and transparent to the users.
Heartbeat is installed on both machines, but not yet configured. This could be done together with an update to Linux 2.6 which should have better performance on the Squids than 2.4.
wikitech-l@lists.wikimedia.org