"Hernan J. Gonzalez" wrote:
>
> -- Brian Baquiran <[EMAIL PROTECTED]> wrote:
> > Hi,
> >
> > I'm using Postgres 6.5.2 in a high-volume
> > application that has on the order of a
> > hundred inserts and selects (but mostly inserts) a
> > minute.
> Some suggestions:
>
> 1) Increase the -B and -S options
> (I use -B5800 -S4000 )
Is there a way to determine the correct parameters based on the current load?
Oracle at least gives you stats on how often it has to go to disk for data.
> 2) Increase the default linux shared memory
> (this may be necesary for the previous
> item to work)
> (for example, in the postgresql.init script
> add the line:
> echo "100000000" > /proc/sys/kernel/shmmax
> )
>
> 3) Turn off flush to disk
> (option -F )
> (very little risk)
I'd rather not risk this...
> 4) Use a 4kb disk-block size
> instead of the default 1kb
> (i'm not sure wheter this matters...)
It's already 4kb. I think that's the default.
> 5) Make a "vacuum analyze" daily
Ok, I was just doing plain "vacuum"
> 6) And the standard one: be careful with
> the indexes. Adding (and also deleting)
> an index may dramatically change the
> performance.
Since I do mostly inserts on some tables, I may delete the indexes (indices?) on
those.
Brian
--
[EMAIL PROTECTED]
http://www.baquiran.com
US Fax: (603) 908-0727
AIM: bbaquiran
