Silva response times
By Nick Dawe, on 6 April 2009
As Silva users will be only too aware, at the turn of the new year the CMS suffered a number of severe performance issues. Silva has now grown to accommodate over 2000 registered users and over 500 different websites (including sandbox sites), and the system began to suffer extremely slow response times as a consequence.
However, after our visit to ETH, Web Services and IS have implemented a new ‘ZEO server’ setup , which splits the Silva server into 5 separate instances, each taking care of particular tasks. The result has surprised us all a little:
[Edit] The diagram above should, of course, be titled ‘Server response time’, not ‘page load time’ – apologies for the confusion.
The big drop shown on this graph is when the ZEO server setup was implemented, and apart from a ‘blip’ on 13th March (caused by the day’s file system problems), it’s been remarkably consistent.
As our Silva service continues to grow, we’ll obviously be keeping an eye on performance, but we’re still quite amazed by the difference that this setup has made.