Re: Performance Tuning Document? - Mailing list pgsql-general

From Steve Wolfe
Subject Re: Performance Tuning Document?
Date
Msg-id 003d01c1d749$f7f1fbe0$d281f6cc@iboats.com
Whole thread Raw
In response to Performance Tuning Document?  (Matthew Kirkwood <matthew@hairy.beasts.org>)
List pgsql-general
> if u imply that you'd better leave huge amounts of memory to the OS
buffer
> cache rather than give them to the DB buffer cache then I strongly
disagree.

  Good, because that wasn't exactly what I was implying.  I was implying
that increasing either disk cache or shared buffers to ridiculous limits
*at the expense of the other* can potentiall be wasteful and even
counter-productive.

> A good on-topic reading is "Avoid Buffered I/O"  by Steve Adams
available at
> http://www.ixora.com.au/tips/avoid_buffered_io.htm

  Thanks, if I ever switch from PG to Oracle, I'll keep that in mind.
Now, for a little bit of REAL WORLD experience.

  Once you've got your sort memory and shared buffers to certain levels,
increasing them isn't going to help you.  In my case, I increased them
until I stopped seeing performance increases, then quadrupled them anyway.
Increasing them further is *not* going to help me.

  However, keeping the database in memory cache *does* help me.  Even
under very significant load (4 processors going full-tilt!), the disk
lights only blink *occasionally*, and that's a good thing.  Disk
bottlenecks really suck.

> also, I don't want to "always keep the *entire* database in disk
cache" - I
> want to keep cached only the frequently accessed parts of the data

  Why?  Do I/O bottlenecks excite you?  If you have the RAM, not using it
is wasteful.  Disks are a place to store data for when the power goes out,
not where you want to do your database work from.

steve



pgsql-general by date:

Previous
From: Oxeye
Date:
Subject: Plpgsql Question
Next
From: "Arguile"
Date:
Subject: Re: plperlu