Performance issue with Tru64 ES40

[Ronnie Doggart]

Hi,

I have been asked to investigate a performance problen on an ES40 running Tru64 V5.1 PK3 and Oracle. The symtoms were that every so often the users would complain about slow response. I installed collect on the server and can see that at particular times the cpu sys time can hit 100% utilitization for between 10-30 seconds. But I cannot identify what is causing this.

The system is not swapping, there is free memory, and I/O drops of to zero when the cpu sys activity spikes.

Any suggestions on how I should proceed.

Ronnie

[Mohamed K Ahmed]

Ronnie,
What version of Oracle you are running?
Please post the output of the command ps aux.
Notice the size of the VSZ and RSS for the oracle instances, they might be high compared with the amount of memory that you have.
Also there are certain paramters to tune the usage of Oracle instances to the physical memory, one important one is shared memory pool size.

Mohamed

[Ronnie Doggart]

Hi,

The memory config on the system is 5GB with only 3GB being used at present, the SGA is 480MB with 190 user processes.

I do not believe that this is an Oracle performance problem, as ordinary unix users also get frozen out when the sys time peaks.

Ronnie

[Michael Schulte zur Sur]

Hi,

what Oracle version do you use? Do you have the Enterprise Manager? Check the effeciency of the indexes. We had a lot of trouble with 4 cpu 4100. Idle time zero and all people complaining about bad performance. Too many full table scans, too many sorts.

greetings,

Michael

[Hein van den Heuvel]

(hey, is this BELFST::DOGGART ? :-)

The system time thing is interesting.
Possible simplelock going wild for a while?
Or the VM subsystem looking for bits and pieces? Or the file system looking for pages to flush?

I would recommend getting a little visibility preferably through DCPI , but perhaps using kprof? I'd just have dcpi run
very coursely around the problem time window,
then take a regular snap, (preferably just before 'every so often' :-), and then a snap again as soon as possible after a reported problem window.

I recall a problem once with 'smoothsync' and an application having 10s of thousands of files open, causing the system to take out locks 'to check what coudl be flushed' too long.

An other problem I recall was with largish SGAs (bigger then you have), regular shared memory and process faults. The system woudl walk too many pages to find out what to toss or not. The solution then was to enable GH.
You may want to verify your ssmthreshold, gh_chunks, new_wire and big_pages settings.

Cheers,
Hein.

[Ralf Puchner]

a) Be sure the Oracle mandatory software requirements are satisfied (patches, versions). In case of the "old" Tru64 5.1 version there should be no unknown issue after installing all required patches named by Oracle.

b) Another starter is to check if the Oracle related tunings have been done.

Btw. Version 5.1 is not supported anymore, version 5.1A or 5.1B are the last available versions.

[Joris Denayer]

Ronnie,

There is more info about standard support and Previous version support at
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=347200

Joris

[Ronnie Doggart]

Hi Hein,

Yes I am BELFST::DOGGART that was a long time ago in my Digital days.

Ronnie