Re: Oracle IO tuning tips/practices on Windows

NetComrade wrote:
> >
> >Nial Litchfield was puting together a compilation of resources
> >for tuning Oracle on Windows. Perhaps you can swing around
> >his website and ask?
> >http://www.orawin.info/services/index.php
>
> Nial needs to tune her blog software a bit :)
> Thanks for the link

"his". Not "her".

> >> I've never dealt with Oracle on Windows before, but someone asked to
> >> help them out with IO performance problems (they can get the dell
> >> powervault SCSI disk array to push data to about 80Megs/sec at the
> >> (same) time when Oracle is only pushing about 20M/sec)

How do they get the scsi disk pushing 80Mb/sec?

Another thing, do some checks with Windows performance monitor: you want to see what size queues there are in this disk device, there is a specific monitor for that.
You might also consider a defrag of those disks (or checking it) : there is just way too many waits on "sequential reads". And my guess is the allocation unit/cluster size might be off mark. Having two block sizes though means ideally you need two partitions, each with its own optimal block size. I don't know how dbfmbr will react with two block sizes but I know that "sequential reads" are not affected by its setting. Unless there is something very atypical going on.

> The db is on the array, OS on local disks

Still, check that no one has created a paging file in the array.

> That's why i needed some Ora/Win resource, I have no idea how to check
> those things on Windows.

Yong mentioned chkdsk, go with that.

> It's SAME.. RAID 10 across 12 disks (6disk stripe)

Given that you're getting mostly sequential read waits - which are direct reads with indexing- it is probably a good idea to set the stripe size pretty small. It's only an advantage with large values if you're doing lots of FTS. Otherwise, you are basically doing IO you might not need to do.

> We run Oracle 9.2.0.6 on RH4 AMD

and so you should!... Received on Thu Feb 02 2006 - 23:57:00 CST