Performance problem?

[Mott Given]

An N4000/04/440Mhz server with 6 GB memory running HP-UX v11.00 has a paging rate around 300 pages/sec while showing 64% free memory and only running 40% CPU busy. Why is there so much paging when so much memory is free? Kernel files are shown in the attached file. The system has 25% swap used, 4 GB memory free, memory utilization 37%, 50% of the response time spent in I/O wait, and there are 2000 defunct processes/hour. Sysadmin considers the defunct processes to be no problem as they are cleaned up periodically by an automatic job.

Mott Given

[Steven E. Protter]

You have an application creating a performance problem.

Whats running on the box and why do so many processes being spawned and orphaned?

SEP

[Mott Given]

There are a number of Oracle databases and MQSeries running on the box.

Mott

[Sanjay_6 · Accepted solution]

Hi Mott,

Maybe this could explain why there is so much paging,

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000063237969

The itrc doc id is S3100003449.

The pagingis probably being done some application code.

Also the number of defunct processes you hve mentioned should be a matter of concern if it is not known how they are being created. Cleaning them periodically is a good thing, but you need to find out, why they are getting created in the first place.

Hope this helps.

Regds

[Steven E. Protter]

Oracle is a known quantity. It does not do this on our systems. Do make sure the system if fully patched, June 2004 or later Quarterly release and anything on metalink.oracle.com that Oracle has added to the requirements list.

The other application may be the issue. You should check the ownership of the defunct processes. If it ties back to the user id that owns application then follow up with that vendor for your next step.

If it ties back to the oracle binary owner then file a tar at http://metalink.oracle.com

SEP

[doug mielke]

One question to ask the DBA would be " have indexes been toughed lately?"

I don't see it contributing to the defuct process, but if full table scans are being used because of the lack of indexes (or cost vs rule based optimization changes cause indexes to be ignored)waiting for i/o can go way up.

sar -b can show how much is being read from buffers, and if reads are low (ours run in the high 90%) system can spend much time trying to find a block, while it reads in tons of table info.

[Mott Given]

Thanks to everybody. The defunct processes are being created by an ORacle application.