Re: Postgres performance comments from a MySQL user

At 15:20 12.06.2003, Justin Clift said:
--------------------[snip]--------------------
>Probably it's a good idea to have some mention of this, as even though we
>should alter the source to higher defaults for our next release, there are
>potentially
>people that would read a message like this and go "wow, didn't know that",
>then tune their existing installations as well.
--------------------[snip]--------------------

Be careful with increasing memory defaults - I just took over a RH7.2
server that still had the SHMALL and SHMMAX settings left at their default
(2MB), for a 2x1000/1GB machine! Turning up the shared_buffers and sort_mem
parameters immediately caused postmaster to fail, of course. Could turn out
messy with newbies, IMHO.

Of course I immediately gained a recognizable performance boost by stuffing
up the 2 OS parameters to 128MB, and setting both shared_buffers and
sort_mem to 4000, even before I dropped in two indexes on some heavily
filled tables where queries were executing sequential searches for 2 rows
out of a million... *sigh*

I believe the idea of an intelligent install script would be near perfect.
It should check the current system hardware and OS memory settings, make
intelligent decisions for manipulation of SHMALL and SHMMAX as well as
shared_buffers and sort_mem, and ask the user/installer to let it modify
these values. Should be a valuable tool, even for finetuning performance
later on.


--
   >O     Ernest E. Vogelsinger
   (\)    ICQ #13394035
    ^     http://www.vogelsinger.at/