Re: Parallel Query and disk striping

Lisa Lewis wrote in message <6o04np$d3e$1_at_usenet01.srv.cis.pitt.edu>...
>Can anyone tell me the considerations for physical layout with the
intention
>of using the parallel query option. I understand that in any good db
>design, tablespaces should be spread across multiple disks.

Yes, and that doesn't change when you stripe. Just consider the stripe set(s) each as one big disk, and follow the normal Oracle installation principles.

>When using
>parallel query, does it matter if you use OS disk striping, RAID
>technologies or manual striping of tablespaces ( via creating TS from
>multiple datafiles on separate disks).

Yes. Speed. Cost. Effort. RAID levels and striping methods offered. All of these things vary from platform to platform, implementation to implementation. Look *hard* at your own kit before you leap.

Likewise, look long and hard at Oracle's PQO - it's not as simple as it seems. Test whatever you implement to death, and test different layouts with real-world SQL on your real-world data.

You've also got to make a descision on *why* you are striping - fault-tolerance? Performance? Mix of both?

>Any suggestions/experiences will be
>greatly appreciated. If you use OS disk striping or raid technologies, I
>assume you would want to always have more than one stripe set ( one for
>indexes, one for tablespaces ) in addition to some disks that are not
>striped for redo's, rollbacks, etc. ???

Yep. And anything that is heavily random I/O you might not want to stripe at all, except for fault-tolerance. You might also want to look at raw devices.

>
>Thanks
>Lisa
>
>
Received on Thu Jul 09 1998 - 02:23:40 CDT