Comments on: Speeding up Koha with memcached

By: chris

chris — Thu, 26 Mar 2009 21:16:20 +0000

The language query might be small, but the code wrapped around it is heavy. By caching the resulting data structure as a serialised object. We can cut the execution time of every page on the opac by a factor of 10.
As you point out, caching the results of a database query, when that query is small and fast isn’t much of a win. In some cases its a loss. But if you then do processing of the resulting data, say parsing xml, or building a complicated hash by doing lots of other small queries, then caching the resulting data structure is a big when.

By: Clay Fouts

Clay Fouts — Tue, 24 Mar 2009 20:10:15 +0000

I think the real caching wins will come from storing MARC binary data blobs for searches. Anything that requires a lot of serialized processor time, like parsing XML data. The systempreference and language queries are annoying and numerous, but as you point out they’re also pretty lightweight.

Threading and parallelizing these sorts of queries will also enable Koha to take better advantage of multi-core systems. That would be much more difficult to pull off, I imagine.

By: Chris

Chris — Mon, 23 Mar 2009 12:59:51 +0000

I did some work on this last year, and a bit more this year.

http://blog.bigballofwax.co.nz/2008/06/22/testing-koha-with-memcached/

http://blog.bigballofwax.co.nz/2008/06/28/too-much-weights-not-enough-speed-work/

http://blog.bigballofwax.co.nz/2008/07/18/a-geeky-way-to-spend-your-lunch-break/

My caching branch in git is here
http://git.workbuffer.org/cgi-bin/gitweb.cgi?p=koha.git;a=shortlog;h=refs/heads/cache

http://git.workbuffer.org/cgi-bin/gitweb.cgi?p=koha.git;a=blob;f=lib/Koha/Cache.pm;hb=refs/heads/cache
Might be of interest to you in the work you are doing.