I have just added a new web server to the load balancing setup. It is currently configured to answer only when the main server goes down (as opposed to sharing the load with it). This should get rid of the "503 Service unavailable" messages that we were regularly seeing on the web server these days.

There may be issues with some sites related to that change. We have tested a few sites (a Drupal and a Tikiwiki) and things seem to be running fine, but if you see weird behaviour, please tell us the exact time at which it was encountered so we can diagnose the problem.

Note that this does not yet improve performance in the cluster, but merely improves reliability. We will shortly deploy a dedicated server that should improve performance as well.