On 13 Nov 2002 at 9:21, Josh Berkus wrote:
> I also just read Tom's response regarding reserving more RAM for kernel
> buffering. This hasn't been my experience, but then I work mostly
> with transactional databases (many read-write requests) rather than
> read-only databases.
>
> As such, I'd be interested in a test: Calculate out your PostgreSQL
> RAM to total, say, 256mb and run a speed test on the database. Then
> calculate it out to the previous 600mb, and do the same. I'd like to
> know the results.
I would like to add here. Let's say you do large amount of reads/writes. Make
sure that size of data exceeds RAM allocated to postgresql. Testing 100MB of
data with 256MB or 600MB of buffers isn't going to make any difference in
performance. If this is the practical scenario, then Tom's suggestion is a
better solution.
IMO postgresql buffers should be enough to hold data requierd for
transactions+some good chunk of read only data. Read only data can be left to
OS buffers for most part of it..
HTH
Bye
Shridhar
--
Sometimes a feeling is all we humans have to go on. -- Kirk, "A Taste of
Armageddon", stardate 3193.9
