Performance Issue

[Dario_1]

Hello Everyone:

I am having an interesting situation with an rp7410 (HP-UX 11.11), which is running Oracle 8.1.7. The problem is that the server occasionally stops for about a second in simple commands like ll (directories with more than 40 files) or dmesg. (I also have complains from users). The box never has more than 7 users on and I have never seen the load over 1.02. I thought that the problem was related to a bad configuration of the LAN Card, which was set to HALF-DUPLEX and AUTONEGOTIATION. I changed the LAN Card and the port (Swith) to 100 Full-Duplex MANUAL but we are still having the problem.

Any ideas will be appreciate.

Thank you.

[Chris Vail · Accepted solution]

I've found that the problems like the one you mention are usually due to a temporary CPU utilization spike. However, there are a lot of possibilities. If you have Glance Plus, you'd do well to have gpm running as you work on the system. Then, the next time you notice the pause, check the GPM output, see if its a CPU spike.
However, massive paging can cause the same symptoms, so too can a disk I/O problem. With the data you've supplied, t'will be difficult to identify. We'll need to know what else you have in the 7410: number and type of disks, amount of memory, SGA size, how the drives are mounted, and just a lot of detail. However, we'll be glad to help you if you post it.


Chris

[Ken Hubnik_2]

Any external disk??? Scsi or Fiber?? Perhaps you are having resets. Anything in syslog.

[Dario_1]

Thank you for your replies.

Syslog is not reportin any errors. We have 8GB of memory and 4 processors. The external Disks are running in 2 HP Disk System 2100 enclosures. Each one of them with 2-18GB and 2-36GBs. We do not have Glance installed in that box but I will go ahead and istall at least the trial version to be able to get more information.

[A. Clay Stephenson]

Because this is an Oracle server, it's possible that a bad timeslice setting was used (based upon an archaic and bad kernel template). Checxk to see that your timeslice has not been set to 1. It should be 10.

[Dario_1]

Ken:

Couple of weeks ago we had some resets problems. HP replaced two of the disks. I was thinking that that was going to be the end of the problem but it was not.

Regards

[Dario_1]

Clay:

Thank you for your reply. It is set to 10.

timeslice 10 - (100/10)

Regards.

[James R. Ferguson]

Hi Dario:

In the absence of empirical data as noted, this is difficult to second-guess. Another possibility that comes to mind is a very large buffer cache that is periodically (~ ever 30 seconds) being flushed.

Since you're running Oracle which handles its own buffers, using a large Unix buffer cache is quite wasteful. Cehck the settings of your 'dbc_min_pct', 'dbc_max_pct', 'nbuf' and 'bufpages'. No more than perhaps 400MB of buffer cache should probably be configured.

Regards!

...JRF...

[Dario_1]

Hello James:

Thank you for your help. They are set to 50 - 5 which is the default. I will make the changes as soon as I can reboot the machine.

Regards

[Dario_1]

James:

I forgot to add bufpages and nbuf. Here they are:

bufpages 0 - (NBUF*2)
nbuf 0 - 0

Thank you again.

[James R. Ferguson]

Hi (again):

To enable the dynamic buffer cache, use the 'dbc_min_pct' and 'dbc_max_pct' and set 'nbuf' and 'bufpages' to zero.

I'd choose something like <2> for the 'dbc_min_pct' and <5> to <8> for 'dbc_max_pct' depending upon your total memory given the guideline of ~400MB buffer cache.

At the very leazt, you will free up memory and eliminate the overhead of both Unix and Oracle handling buffers.

After changing the settings, evaluate your system and let us know the results, please!

Regards!

...JRF...

[Dario_1]

All:

Thank you for all your help. I will change the Kernel Parameters over the weekend and will advise of any changes.

Regards