bottle neck

[Norman_21]

Hello experts,

I've been monitoring the glance closly and I've realized that sometimes a mesg pops up states there is a Bottle Neck!! Also, the mesg disappears real fast. The system sometimes gets very slow and there is only few users logged in!!
However, at first I thought it's a memory leak. But the Glance Plus output doesn't seems to be a Memory Leak.
Also, the dbc_max_pct is 50 and the dbc_min_pct is 5.
The uptime is:
4:29am up 7 days, 42 mins, 14 users, load average: 1.04, 1.17, 1.23
The swapinfo -ta is:
Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 2023424 0 2023424 0% 0 - 1 /dev/vg00/lvol2
dev 2043904 0 2043904 0% 0 - 1 /dev/vg00/lvol4
reserve - 194288 -194288
memory 1559004 1221720 337284 78%
total 5626332 1416008 4210324 25% - 0 -

Attached is the Memory and the pro. from glance plus.
Your advice will be highly appreciated.
Points will be awared.

[Norman_21]

here is the proc output from glance plus.

[James Murtagh]

Hi,

If you are experience a spike in performance it can be quite hard to catch. If you have the Motif version of glance, gpm, you can use the symptom history graph (main - advisor - symptom history) to keep an active track on this. If you find the default interval is too much (15 secs) you can change this from measurements menu (main - configure - measurements).

You can also set up some sar jobs at low interval settings to try and catch it. I suspect CPU spikes at this point.

Regards,

James.

[Norman_21]

Hello,

Any documents, good cook books or something!!

Thanks.

[Ramkumar Devanathan · Accepted solution]

Maybe this doc would help.

http://h21007.www2.hp.com/dspp/files/unprotected/devresource/Docs/TechPapers/UXPerfCookBook.pdf

- ramd.

[James Murtagh]

Hi,

I'm sure there are a few white papers etc on the ITRC, although as I don't have a contract I can't get access to them!! (only work here after all :-) )

However, found this guide, its pretty good:

http://h21007.www2.hp.com/dspp/files/unprotected/devresource/Docs/TechPapers/UXPerfCookBook.pdf

Do a search on the forums for this, I know doc ids have been given before. I've also attached a doument.

Regards,

James.

[Ramkumar Devanathan]

James,

that's the same link we posted at almost the same time! ain't that cool?

- ramd.

[James Murtagh]

I know!

Great minds and all that! :-)

[T G Manikandan]

The buffer cache really makes a impact here.

The default value for the buffer cache is 50% of the memory(dbc_max_pct) which takes half of the memory.

You can bring that down to 300-400MB.
If you are using 2GB memory then the dynamic buffer cache maximum 15-20%.

Bringing this change can result in improving the bottleneck.

You can also check what processes are really occupying/using the memory resources using

UNIX95= ps -e -o ruser,pid,vsz=Kbytes|more
There are lot of documents for the performance and Tuning in Tech base.

Do a search there.

revert


Thanks

[Norman_21]

Thanks TG. I just want to make sure before I do any changes. This a production server so I have to scheduale for that too.
James,
Since you think it is CPU spikes, do you think this can be solved by adding more CPUs.

All, please share.
Thanks.

[James Murtagh]

Hi,

I was only guessing really, looking at the min/avg/high figures from your glance output and the process table. It could as easily be a disk IO related issue. You will have to go through the performance docs and try and narrow it down. If you are just getting spikes in performance I'd imagine tuning the system would be the way to go. Even if it is a CPU issue I can't imagine buying more would help much in this circumstance.

If you have any specific questions, or need help to analyse any results you get, please let me know and I will do my best to help.

Regards,

James.

[Norman_21]

File system defragmentation resolved the issue!
Thanks to everyone...