Unix system performence very slow

[skravi17]

Hi

We have HPUx - 11.0 - N class Server.
Total physical memory - 8GB
Total Disk space used for oracle 11i database
- 80 GB.
Raid configuration 1+5
Total users - average - 300 daily.
Now we are facing peformance problem.
Every day we get disk usage 100% and cpu usage
100%.(Disk bottleneck & cpu bottle neck) Here i am attaching our kernel parameter list. Any one can advise me how to rectify the problem.

regards

ravi

[skravi17]

Hi


Here i am attaching mem_activity,disk,cpu
activity for one day.

regards
ravi

[skravi17]

Hi

cpu activity.

[skravi17]

Hi

Here i am attaching all the files.

[Michael Tully]

First you need to do this:

Your buffer cache is too high.

Suggest you change for starters:
dbc_max_pct - 5%
dbc_min_pct - 2%
eqmemsize - 30

What sort of disk array are you using?

Do you have any sort of timeout values other than default for your LUN's?

# pvdisplay -v /dev/dsk/c9t0d0 |more
and look for the IO timeout in seconds. If your using an external disk array these should be set to about 180 seconds.

[Stefan Farrelly]

Well, apart from dbc_max_pct being too high (but this will reduce anyway once memory pressure builds) I cant see any problems. What you are seeing is a feature of Oracle and having 300 users on your server.

Oracle by nature will hit all your CPU's as hard a possible (to the max) in order to meet the requests of all the users you have on your server. This is fine - thats what you would want - use all the cpu available in order to meet requests as fast as possible.

Disk looks fine also. Its busy at times but wio% doesnt get too high for too long. Again - this is a feature of having that many users on your server - Oracle hits the disks hard as possible in order to meet the requests as fast as possible.

If your users are complaining or poor performance then this is simply because you have too many users on your server! Either reduce the number of users or increase the specification of your server (more&faster cpu's, more memory, more and daster disk).

[Michael Tully]

Look at this posting from some more suggestions on Oracle related kernel parameters. The High %usr on CPU may indicate that there is a limited amount of space in your kernel for any sort of number crunching.

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x7acacf38d6bdd5118ff10090279cd0f9,00.html

The CPU bottleneck could be because it is waiting on disk IO. Looking at your disk IO, this suggests that one controller is getting flooded. One way would be to move some information around on your disks. Do you utilise any sort of disk striping?