paul@wayr.org wrote:
> Thanks,
>
> Unfornatly, i can't update pgsql to 8.3 since it's not in debian stable.
That's why backports.org was invented :-)
Or does can't mean "not allowed to"?
> So i'm going to play with work_mem & shared_buffers.
>
> With big shared_buffers pgsql tells me
> shmget(cle=5432001, taille=11183431680, 03600).
> so i do "echo 13183431680 > /proc/sys/kernel/shmmax" ( 10Go + 2Go just
> in case)
>
> but pgsql tells me again that it there's not enought shm..
> How can i compute the go shmmax for my server ?
I'm not seeing anything terribly wrong there. Are you hitting a limit
with shmall?
Oh - and I'm not sure there's much point in having more shared-buffers
than you have data.
Try much larger work_mem first, I think that's the biggest gain for you.
--
Richard Huxton
Archonet Ltd
