Message-ID: <450999EC.8070601@modgraph-usa.com>
Date: Thu, 14 Sep 2006 11:05:32 -0700
From: "Craig A. James" <cjames@modgraph-usa.com>
User-Agent: Thunderbird 1.5.0.5 (X11/20060719)
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: RAID 0 not as fast as expected
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

I'm experiment with RAID, looking for an inexpensive way to boost performance.  I bought 4 Seagate 7200.9 120 GB SATA drives and two SIIG dual-port SATA cards.  (NB: I don't plan to run RAID 0 in production, probably RAID 10, so no need to comment on the failure rate of RAID 0.)

I used this raw serial-speed test:

   time sh -c "dd if=/dev/zero of=./bigfile bs=8k count=1000000 && sync"
   (unmount/remount)
   time sh -c "dd if=./bigfile of=/dev/null bs=8k count=1000000 && sync"

Which showed that the RAID 0 4-disk array was almost exactly twice as fast as each disk individually.  I expected 4X performance for a 4-disk RAID 0.  My suspicion is that each of these budget SATA cards is bandwidth limited; they can't actually handle two disks simultaneously, and I'd need to get four separate SATA cards to get 4X performance (or a more expensive card such as the Areca someone mentioned the other day).

On the other hand, it "feels like" (using our application) the seek performance is quite a bit better, which I'd expect given my hypothesis about the SIIG cards.  I don't have concrete benchmarks on seek speed.

Thanks,
Craig