Re: Scaling further up - Mailing list pgsql-performance

"By lots I mean dozen(s) in a raid 10 array with a good controller."

I believe, for RAID-10, I will need even number of drives. Currently,
the size of the database is about 13GB, and is not expected to grow
exponentially with thousands of concurrent users, so total space is not
of paramount importance compared to performance.

Does this sound reasonable setup?
10x36GB FC drives on RAID-10
4x36GB FC drives for the logs on RAID-10 (not sure if this is the
correct ratio)?
1 hotspare
Total=15 Drives per enclosure.

Tentatively, I am looking at an entry-level EMC CX300 product with 2GB
RAID cache, etc.

Question - Are 73GB drives supposed to give better performance because
of higher number of platters?

Thanks,
Anjan


-----Original Message-----
From: Fred Moyer [mailto:fred@redhotpenguin.com]
Sent: Tuesday, March 02, 2004 5:57 AM
To: William Yu; Anjan Dave
Cc: pgsql-performance@postgresql.org
Subject: Re: [PERFORM] Scaling further up


On Tue, 2004-03-02 at 17:42, William Yu wrote:
> Anjan Dave wrote:
> > We have a Quad-Intel XEON 2.0GHz (1MB cache), 12GB memory, running
> > RH9,
> > PG 7.4.0. There's an internal U320, 10K RPM RAID-10 setup on 4
drives.
> >
> > We are expecting a pretty high load, a few thousands of 'concurrent'
> > users executing either select, insert, update, statments.
>
> The quick and dirty method would be to upgrade to the recently
> announced
> 3GHz Xeon MPs with 4MB of L3. My semi-educated guess is that you'd get

> another +60% there due to the huge L3 hiding the Xeon's shared bus
penalty.

If you are going to have thousands of 'concurrent' users you should
seriously consider the 2.6 kernel if you are running Linux or as an
alternative going with FreeBSD.  You will need to load test your system
and become an expert on tuning Postgres to get the absolute maximum
performance from each and every query you have.

And you will need lots of hard drives.  By lots I mean dozen(s) in a
raid 10 array with a good controller.  Thousands of concurrent users
means hundreds or thousands of transactions per second.  I've personally
seen it scale that far but in my opinion you will need a lot more hard
drives and ram than cpu.