Re: Server load planning

Dan,

You can get a sense of how much memory you will need by the shorthand
presented in table 16-2 for calculating the value of SHMMAX:

http://www.postgresql.org/docs/8.0/static/kernel-
resources.html#SYSVIPC-PARAMETERS

Otherwise, you'll need to include some estimate of work_mem and
maintenance_work_mem based on your knowledge of your queries:

http://www.postgresql.org/docs/8.0/static/runtime-config.html#RUNTIME-
CONFIG-RESOURCE

As far as disk I/O and contention at that level, I'm not sure how that
will be affected by sheer number of connections. There's a simple
utility in contrib called pgbench that you could use to do some
testing.

-tfo

--
Thomas F. O'Connell
Co-Founder, Information Architect
Sitening, LLC

Strategic Open Source — Open Your i™

http://www.sitening.com/
110 30th Avenue North, Suite 6
Nashville, TN 37203-6320
615-260-0005

On Mar 27, 2005, at 11:38 AM, Dan Sugalski wrote:

> I'm trying to get a handle on how an app I'm looking to roll out's
> going to impact the server I'm connecting to, and what sort of
> capacity planning's going to be needed to make it all work relatively
> well.
>
> I'm looking at around 250-300 simultaneous users, nearly all of them
> doing interactive work. (Curses-based screen stuff for the most part)
> I'd not too worried about the server we've got for them being able to
> handle that except... for reasons that are fairly annoying, I'm
> looking at somewhere in excess of 9K simultaneous connections to the
> database server, and I'm not in a position to cut that down any. (The
> app suite's written in an old 4GL that assumes an ISAM database. We're
> porting to a modern database and runtime, but we have to preserve the
> DB semantics of the original database. Nasty, but there you go)
>
> I know each of the back-end processes is going to suck down some
> resources on the server, but am I going to hit coordination or
> inter-process sync delays with that many different back ends going at
> once? (And is there a good way, short of just running some load tests,
> to estimate the costs involved?)