Advice: HA Storage Appliance for Postgres

[Rich Holland]

Danny Howard wrote:

> I'd like to maybe set up a dedicated disk solution where I can store the
> database on a RAID array on its own hardware unit, maybe with a
> hot-failover RAID controller -- the idea being that the backend storage
> will be high availability, and if one machine fails, I can bring the
> data up on another server (though I may have to run fsck or the postgres
> clean sequence first)

Sounds like you want a SAN.

> I have a little experience with NetApps, as well as with some dedicated
> hardware that connected to the servers via SCSI cables.  I think the
> latter is a better solution, especially if I can connect multiple
> servers to the same RAID hardware, and bring up the disk on one or the
> other.  (Having a Datacenter Tech walk over and swap a cable could also
> do ...)

If you really want SCSI, I know the Dell PowerVault arrays will handle
multiple hosts with the PERC/4 RAID controllers; that's what we use for
NT-based clustering solutions for ERP systems.

> - I think I want to avoid anything called "SAN" right?

Why?  You can set up a simple SAN with a single storage array using
protected disks.  Set up a couple of small switches and put at least one HBA
in each host.  If you want REAL high availability, use dual HBA's with
something like PowerPath from EMC.  You can mirror the database within the
same array or to a second array at a remote location on the same SAN
(another building, etc).

> - These days, is SATA versus SCSI versus fiber versus copper connects
>   important?

SATA is a lot cheaper and nearly as fast as Ultra320 SCSI from what I've
seen.  Most of the high-end arrays still use SCSI, but there are SATA arrays
out there that will do what you want.

> - Anyone have a particular vendor they like?  

Anyone that won't be out of business next year.  IBM (Shark), EMC, Hitachi,
Dell (EMC reseller and PowerVault 22xS SCSI arrays)

- Is an NFS/NetApp solution viable?

- Do smaller disks provide better performance in RAID contexts?

Depends on what kind of RAID you're using.  Normally smaller disks means
more spindles, which equates to better performance.  If you're using it for
a database, RAID 0+1 (striped & mirrored) is your best bet for performance
and protection.  Then use the vendor's array technology to mirror that to
another set of disks elsewhere (same array or a different one, doesn't
matter).
--
Rich Holland        (913) 645-1950        SAP Technical Consultant
print unpack("u","92G5S\=\"!A;F]T:&5R(\'!E<FP\@:&%C:V5R\"\@\`\`");