On Wed, Apr 30, 2003 at 03:09:52PM -0600, scott.marlowe wrote:
> What about commit_siblings and commit_delay? I haven't really played a
> lot with those. I would think increasing the commit delay and the number
> of siblings should give better, but slightly bursty performance.
This depends very heavily on your actual traffic. If you have
primarily write activity, it is a definite advantage, because you get
a boost from ganging writes to disk. If your traffic leans more
toward reads, I find that increasing these settings is a net loss:
you just end up waiting for nothing.
BTW, it is tricky to construct a correct model load to test this,
because you have to model both the load and its distribution.
A
--
----
Andrew Sullivan 204-4141 Yonge Street
Liberty RMS Toronto, Ontario Canada
<andrew@libertyrms.info> M2P 2A8
+1 416 646 3304 x110