Thread: Storage Options

Storage Options

From
Jeff Davis
Date:
I am trying to decide what kind of storage options to use for a pair of
good database servers, a primary and a read-only that can be a failover.
Here is what I'm thinking so far:

(1) We have a nice NetApp that can do iSCSI. It has a large (multi-GB)
battery-backed cache so it could potentially perform the transactions at
a very high rate. However there are many other applications accessing
the NetApp over NFS, so I am not sure what performance to expect. Any
suggestions about using network storage like this for the database? Will
the database make huge demands on the NetApp, and force my department
spend huge amounts on new NetApp hardware?

(2) I read with interest this thread:
http://archives.postgresql.org/pgsql-performance/2006-08/msg00164.php

Is there any consensus on whether to do WAL on a RAID-1 and PGDATA on a
RAID-10 versus everything on a RAID-10? How does the number of disks I
have affect this decision (I will probably have 4-8 disks per server).

Some of the applications I initially need to support will be a high
volume of simple transactions without many tablescans, if that helps.
However, I expect that these servers will need to serve many needs.

Any other suggestions are appreciated. Is there a common place to look
for hardware suggestions (like a postgresql hardware FAQ)?

Regards,
    Jeff Davis


Re: Storage Options

From
Michael Stone
Date:
On Mon, Aug 21, 2006 at 02:50:51PM -0700, Jeff Davis wrote:
>the NetApp over NFS, so I am not sure what performance to expect. Any
>suggestions about using network storage like this for the database?

Don't. Unless you're using a very small (toy-scale) database, the netapp
storage is way too expensive for the kind of usage you see with a
database application. You're much better off buying much cheaper storage
twice and using a database replication solution than either choking a
really expensive netapp or getting lousy performance from the same. The
netapps have their niche, but database storage isn't it. (Peformance in
general really isn't it--the advantages are managability, snapshotting,
and cross-platform data exchange. It may be that those factors are
important enough to make that a good solution for your particular
situation, but they're generally not particularly relevant in the
postgres space.)

>Is there any consensus on whether to do WAL on a RAID-1 and PGDATA on a
>RAID-10 versus everything on a RAID-10? How does the number of disks I
>have affect this decision (I will probably have 4-8 disks per server).

You can't get a good answer without testing with your actual data. I'd
suspect that with such a low number of disks you're better off with a
single array, assuming that you have a good bbu raid controller and
assuming that you're not doing write-mostly transaction work. But
testing with your actual workload is the only way to really know.

Mike Stone