Re: Fundamental database design question

>Database block size is 8K, OS disk block size is 8K. Stripe size on all
>disks (note exception below) is 16K
 

>I still experience disk I/O bottlenecks occasionally.

Any multiblock read or write has to access multiple discs. Possibly a nominal 64 of them depending on platform and setup. This may have some bearing on your bottleneck.

I wouldn't stripe at less then 64K per disc, unless you found a very good reason.

--

Jonathan Lewis
Yet another Oracle-related web site:  http://www.jlcomp.demon.co.uk

Dave Bottom wrote in message <3JTl5.864$P2.16601_at_harpo>...


>Here is what we've done for our production (PeopleSoft) database.

>

>we have 60 disks running on two dual fiberchannel controllers (4

 controllers


>max)

>

>these 60 disks are hardware mirrored so the system only sees 30 devices

>

>Of these 30 devices we've done some striping and I have 14 logical volumes

>with 2 spares.

>

>Database block size is 8K, OS disk block size is 8K. Stripe size on all

>disks (note exception below) is 16K

>

>I've put oracle home on one disk (not stiped).

>I've got data, hopefully load balanced, on 7

>I've put my indexes on 2

>I've put my temp space on 2 of these with a large stripe size (64K)

>I've got my rollbacks and redologs on 2

>

>I still experience disk I/O bottlenecks occasionally.

>Some of this is poor SQL tuning, we're working on that.

>Some of this is poor archetecture, we're working on that. What I've

 recently


>done is take some tablespaces where multiple tables exist that get

>simultaniously read and written and broken them up into multiple

>tablespaces, migrated the tables, and spread them over multiple disks. It

>helps.

>

>I guess what I'm trying to say is theory is great but in practice you need

>to monitor your database load, and make changes based on what is happening.

>

>Dave

>

>