On Thu, 2007-09-20 at 16:38 -0400, Bill Moran wrote:
> In response to Robert Fitzpatrick <lists@webtent.net>:
> Why does everyone leave of the IO subsystem? It's almost as if many
> people don't realize that disks exist ...
>
> With 2G of RAM, and a DB that's about 3G, then there's at least a G of
> database data _not_ in memory at any time. As a result, disk speed is
> important, and _could_ be part of your problem. You're not using RAID
> 5 are you?
Yes, using RAID 5, not good? RAID 5 with hot fix total of 4 drives. All
SATA 80GB drives giving me little under 300GB to work with.
Also, my nightly backup does a pg_dump of the one database and vacuums
only that database as there are no other except template#'s. Then it
does a pg_dumpall. Now, I noticed that we have the -dD flags on
pg_dumpall, not sure why, I took them off. But the strange thing I am
finding is while my one database using a 'pg_dump -F c' only comes out
at 930MB while the pg_dumpall results in 3GB, is that due to the use of
INSERTS by using -dD?
> > max_connections = 250
> > max_fsm_pages = 204800
> > shared_buffers = 128MB
>
> Unless this machine runs programs other than PostgreSQL, raise this to
> about 650MB. You might get better performance from even higher values.
> The rule of thumb is allocate 1/4 - 1/3 of the available RAM to
> shared_buffers ... subtract the RAM that other programs are using first.
Yes, it runs a few other things like Postfix+amavisd-maia+SA+clamAV, but
low priority MX so it gets little unless the primary is not responding.
Other than that, I use it to run the web GUI (php) for this amavisd-maia
mail server where users can view spam/ham caches. Can I determine the
amount of memory everything else is running by stopping postgres and
look in top to see what is being used?
Thanks for the other pointers...!
--
Robert