Re: Solaris 9 tuning - Mailing list pgsql-performance

From Paul Johnson
Subject Re: Solaris 9 tuning
Date
Msg-id 28097.80.5.160.4.1107975762.squirrel@www.gradwell.com
Whole thread Raw
In response to Re: Solaris 9 tuning  (Josh Berkus <josh@agliodbs.com>)
Responses Re: Solaris 9 tuning
Re: Solaris 9 tuning
List pgsql-performance
Hi Josh, there are 8 internal disks - all are 18GB@10,000 RPM, fibre
connected.

The O/S is on 2 mirrored disks, the Postgres cluster is on the /data1
filesystem that is striped across the other 6 disks.

The shared_buffers value is a semi-educated guess based on having made 4GB
shared memory available via /etc/system, and having read all we could find
on various web sites.

Should I knock it down to 400MB as you suggest?

I'll check out that URL.

Cheers,

Paul.

> Paul,
>> I would like to know what /etc/system and postgresql_conf values are
recommended to deliver as much system resource as possible to Postgres.
We
>> use this Sun box solely for single user Postgres data warehousing
workloads.
> What's your disk system?
>> shared_buffers = 500000
> This is highly unlikely to be optimal.   That's 3GB.   On test linux
systems
> up to 8GB, we've not seen useful values of shared buffers anywhere above
400mb.    How did you arrive at that figure?
>> sort_mem = 2097152
>> vacuum_mem = 1000000
> These could be fine on a single-user system.   sort_mem is per *sort*
though,
> not per query, so you'd need to watch out for complex queries spillling
into
> swap; perhaps set it a 0.5GB or 1GB?
> Otherwise, start with the config guide at
www.powerpostgresql.com/PerfList
> --
> Josh Berkus
> Aglio Database Solutions
> San Francisco





pgsql-performance by date:

Previous
From: John Arbash Meinel
Date:
Subject: Re: Tell postgres which index to use?
Next
From: "Marinos J. Yannikos"
Date:
Subject: Re: GiST indexes and concurrency (tsearch2)