Re: Disk I/O

On Thu, 17 Mar 2005 06:39:30 +0100, rainer <Rainer.Klomps_at_t-online.de> wrote:

>Hello,
>
>we are using a 9.2.0.5 DB on a Windows Advanced Server machine having 8 *
>1.8 GHz CPUs, 8GB of MM.
>The database is used as well as DW as also as OLTP system. Data is stored on
>three Clariion RAID5 / fiber channel disk systems.
>When monitoring the database I can see lots of disk waits, which is expected
>when considering the relation CPU performance / number of disks.
>However, when looking at the Clariion performance monitor, I see between 40
>and 80 IO requests / sec and between 0 and 4 queued IO requests. The queue
>size does not raise during regular DB operation.
>When doing additional disk operations via OS, as e.g. copying a file, IO
>requests raise up to 600 / sec.
>
>The question is: Why is the DB waiting for IO, while the disks seem not to
>work at their limits?

So far the information you provided doesn't actually tell anything at all.
First of all, we don't know the number of I/O requests the disks should be able to cope with. In many situations also the existence of any I/O queue would be unacceptable.
Secondly, we don't know *what* kind of I/O this is. Read requests? Single block reads? Multi block reads? Write requests? Run statspack, run the statspack report and submit the results to www.oraperf.com and have it analysed.
Finally, it is a well known fact there is a penalty associated with running Oracle on Raid-5 disks. If you use the database both for OLTP and DW purposes, putting the database on RAID-5 devices isn't very appropiate.

--
Sybrand Bakker, Senior Oracle DBA