CPU IO wait always more than 60% and swap is zero

[Former Member]

Hi All,

Our system is AIX and oracle. We got a problem is that CPU IO wait is always large than 60% and swap is zero. Is it reasonable? System is slow now, is it the reason?

Thanks

Wilson

[Former Member]

Hi All,

Thanks for your reply and recommendations. I will try and let your know what I find.

Thanks

Wilson

[Former Member]

Some more recommendations for you.

Assuming you are running 10g database

from sqlplus

1. @$ORACLE_HOME/rdbms/admin/addmrpt.sql running this script for a specified snapshot interval ie start snap id and end snap id should reflect the problem time, the output could be text html file then verify what are the wait events causing performance bottlenecks also the disk layout and their throughput will be reflected.This output will also reflect the highly expensive sql statements if any using huge resources like cpu .

2. IO contention can also be verified using iostat (os tool) or catio.sql(oracle supplied script, you wud need to search it out from metalink I don't remember the metalink number right away).

3. Topas is another good tool for AIX to detect which processes is holding up more cpu ,runqueue etc.

Regards

Kausik

[markus_doehr2]

> Our system is AIX and oracle. We got a problem is that CPU IO wait is always large than 60% and swap is zero. Is it reasonable? System is slow now, is it the reason?

I'd interpreted that more as "your I/O subsystem is too slow/overloaded".

Markus

[Former Member]

What should we do? anything i can prove it? like to capture some figure

Thanks

Wilson

Edited by: wilsonwu on Jun 2, 2010 3:28 PM

[markus_doehr2]

> What should we do? anything i can prove it? like to capture some figure

You can check in the SAP system

- SM50/SM66

- ST06

- DBACOCKPIT

on OS level you can use 'sar' and 'iostat' to get statistics about the I/O load.

Markus

[stefan_koehler]

Hello Wilson,

> What should we do? anything i can prove it? like to capture some figure

You can use nmon to analyze your environment: http://www.ibm.com/developerworks/aix/library/au-analyze_aix/

You can drill down the I/O bottlneck to the disk and monitor its throughput, etc. It is also possible to run nmon in "background capture" mode and you can import the generated file into "nmon analyzer" (http://www.ibm.com/developerworks/aix/library/au-nmon_analyser/) and generate nice charts, etc.

If you want to go in detail on disk level - you can use filemon to get detailed information about the I/O on that disk, etc.

Regards

Stefan