Joshua Jackson wrote:
Here's the question, gnome's bugzilla has over twice as many bugs as
we have, is quite speedy and doesn't seem to suffer from the OOM
killers that our bugzilla has. So what's the difference? Did gnome
just toss hardware at the problem to make it go away or have they done
something to make bugzilla work for them?
I think throwing hardware at the problem is the wrong approach in this
case, as its just delaying the problem that has made the new hardware
seem like the solution...which will no doubt creep up again.
Don't get me wrong, the donation of hardware from gni is greatly
appreciated. I'd just to see that we try and see why we have the
problem in the first place as well. As I'm sure that this problem will
creep up again
Another technique is to change high transaction tables to Innodb table
format. Innodb is going to be roughly 30% slower than MyISAM for selects
and take up much more space on disk approx 3-5x larger. However it has
row locking which solves the contention issue. A good example of mixed
table types is actually Mediawiki which uses Memory for hitcounters,
Innodb for pages and revisions, and MyISAM for everything else.
Bugzilla has 50 tables in its schema, but converting bugs and
bugs_activity to Innodb might cause more problems than it solves.
Normally a decision to change tables format is accompanied by some
normalization of your tables and changing queries to get the best
performance out of each table type. It's possible that changing a few
tables would work and have no downside, but if that were the case I'd
expect upstream to have suggested it.
Here's a paper on general Mysql scaling that's pretty interesting and
easy to read if you don't have much db background.
http://www.danga.com/words/2005_mysqlcon/mysql-slides-2005.pdf
kashani
--
gentoo-dev@gentoo.org mailing list
