From: | Greg Smith <gsmith(at)gregsmith(dot)com> |
---|---|
To: | Benjamin Minshall <minshall(at)intellicon(dot)biz> |
Cc: | pgsql-performance(at)postgresql(dot)org |
Subject: | Re: AMD, Intel and RAID controllers |
Date: | 2009-10-01 18:43:41 |
Message-ID: | alpine.GSO.2.01.0910011424130.27583@westnet.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
On Thu, 1 Oct 2009, Benjamin Minshall wrote:
> I'm basically looking at something in the ProLiant DL380 series which
> boils down to Intel Xeon 5500 or AMD Opteron 2600. Are there any
> notable performance concerns regarding Postgres on either of these cpus?
The Xeon 5500 series are very impressive performers. The only reason I
could think of for why someone might want the older AMD design is if it
saved enouch money to buy more disks for an app limited by those.
> RAM will likely be in the 16GB range. Any comments on bus speeds or other
> issues related to the RAM?
The Xeon 5500 models I've tested didn't seem to vary all that much based
on the memory speed itself. It is important to pay attention to where the
major breaks in bus speed on the processor are though, because those bumps
really mean something.
> What is the opinion on HP disk controllers? The standard controller on this
> server line is the Smart Array P400 (512MB BB cache) although the option is
> available to go up to P600 or P800. I plan to need about 500GB (8 146GB
> disks, raid 10). Are the HP controllers worth my time or should I be looking
> elsewhere?
Those are reasonable controllers, and the list archives here are filled
with a bias toward the P800.
http://www.nabble.com/Experience-with-HP-Smart-Array-P400-and-SATA-drives--td20788664.html
is a good sample, there are more. The important thing to realize is that
RAID5 performance on the card is going to be awful no matter what you do,
since you're using RAID10 you should be fine.
> Finally, I'm thinking 10k RPM SAS drives are appropriate. Does the
> substantial price increase for 15k RPM drives really show in the overall
> performance of the storage array?
If your app is limited by disk seeking and general latency, those can make
sense. Ideally, you'd get enough RAM for caching that you're not hitting
the disks hard enough for the difference between them to matter so much.
--
* Greg Smith gsmith(at)gregsmith(dot)com http://www.gregsmith.com Baltimore, MD
From | Date | Subject | |
---|---|---|---|
Next Message | david | 2009-10-01 19:10:35 | Re: Best suiting OS |
Previous Message | Greg Smith | 2009-10-01 18:16:27 | Re: Best suiting OS |