On Tue, 1 Jul 2003, Michael Mattox wrote:

> My understanding is to use as much effect cache as possible, so figure out
> how much ram you need for your other applications & OS and then give the
> rest to postgres as effective cache.
> What I learned to day is the shared_buffers 25% of RAM guideline.

Note that the best guideline is the one that your testing shows you makes 
the most sense.  If you never access more than a few megs at a time, then 
there's no need to have 25% of a machine with 1 gig given over to the 
database's shared buffers, it's better to let the machine cache that for 
you.  If you access hundreds of megs at a time, then 25% of RAM is a good 
idea.  Usually 25% of RAM is about the max that gives good results, but in 
some corner cases, using more still makes sense.  Usually at that point, 
you've also increased sort_mem up a bit too, but be careful, sort_mem is 
PER SORT, not per backend or per database cluster, so it can add up very 
quickly and make the machine run out of RAM.

Setting these settings is a lot like playing Jenga (the game with the 
wooden blocks stacked up where you pull one out and put them on top one at 
a time.)  Everything seems just fine, the machine's getting faster and 
faster, everybody's loving life, then you crank one up a little too high, 
cause a swap storm, and the whole thing slows to a crawl.

---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
      subscribe-nomail command to [EMAIL PROTECTED] so that your
      message can get through to the mailing list cleanly

Reply via email to