Re: NAS, SAN or any alternate solution ? - Mailing list pgsql-performance

From Grega Bremec
Subject Re: NAS, SAN or any alternate solution ?
Date
Msg-id 20040720122335.GD16410@elbereth.noviforum.si
Whole thread Raw
In response to NAS, SAN or any alternate solution ?  (bsimon@loxane.com)
Responses Re: NAS, SAN or any alternate solution ?
List pgsql-performance
...and on Tue, Jul 20, 2004 at 09:52:56AM +0200, bsimon@loxane.com used the keyboard:
> Hi all,
>
> I've been searching the list for a while but couldn't find any up-to-date
> information relating to my problem.
> We have a production server with postgresql on cygwin that currently deels
> with about 200 Gigs of data (1 big IDE drive). We plan to move to linux
> for some reasons I don't have to explain.
> Our aim is also to be able to increase our storage capacity up to
> approximately 1 or 2 terabytes and to speed up our production process. As
> we are a small "microsoft addicted" company , we have some difficulties to
> choose the best configuration that would best meet our needs.
> Our production process is based on transaction (mostly huge inserts) and
> disk access is the main bottlle-neck.
>
> Our main concern is hardware related :
>
> Would NAS or SAN be good solutions ? (I've read that NAS uses NFS which
> could slow down the transfer rate ??)
> Has anyone ever tried one of these with postgresql ?
>
> I would appreciate any comments.
> Thanks in advance.

Hello Simon,

We're testing 3ware Escalade 9000, which is a hardware-raid SATA
controller with VERY good support for Linux (including direct access
for S.M.A.R.T. applications, which is a serious problem with other
RAID controllers), featuring RAID levels 0, 1, 10, 5, JBOD, up to
12 SATA channels (that's 3ware Escalade 9500S-12, they also come in
4- and 8-channel versions, up to four cards can be fitted into a
system), up to 1GB battery-backed ECC RAM (128MB out-of-the-box)
and most of all, excellent tuning guides that actually manage to
exceed the scope of merely making you come up with good benchmark
results for that controller in a specific test environment.

Our preliminary tests show that a setup of four 250GB SATA Maxtors
that aren't really qualified as fast drives, in RAID5 can deliver
block writes of 50MB/s, rewrites at about 35MB/s and reads of
approximately 180MB/s, which is rougly 2.5-times the performance
of previous Escalades.

You can find more info on Escalade 9000 series, benchmarks and
other stuff here:

    http://www.3ware.com/products/serial_ata9000.asp
    http://www.3ware.com/products/benchmarks_sata.asp
    http://www.3ware.dk/fileadmin/3ware/documents/Benchmarks/Linux_kernel_2.6_Benchmarking.pdf

Oh, and not to forget - the price for a 3ware 9500S-12, the version
we're testing ranges between EUR1000 and EUR1500, depending on the
contract you have with the reseller and the intended use of the
device. SATA disks are dirt-cheap nowadays, as has been mentioned
before.

I do agree on the reliability of cache-usage setting those drives
report though, it may or may not be true. But one never knows that
for sure with SCSI drives either. At least you can assert that
proper controller cache sizing with drives that usually feature
8MB (!!!) cache, will mostly ensure that even the largest amount
of data that could fit into a hard disk cache of the entire array
(96MB) will still be available in the controller cache after a
power failure, for it to be re-checked and ensured it is properly
written.

Hope this helps,
--
    Grega Bremec
    Senior Administrator
    Noviforum Ltd., Software & Media
    http://www.noviforum.si/

Attachment

pgsql-performance by date:

Previous
From: "Steinar H. Gunderson"
Date:
Subject: Re: Odd sorting behaviour
Next
From: Grega Bremec
Date:
Subject: Re: NAS, SAN or any alternate solution ?