Oracle 9i required Kernal changes and HP-UX 11.0 on v-2250

[Philip Ladouceur]

Has anyone had problems with the kernal parameter changes in HP-UX 11.0 on a V2250 box for the ORacle dictated kernal changes...
I am specifically changing
maxdsiz from 335544320 to 1073741824
maxdsiz_64bit from 1073741824 to 2147483648
maxssiz from 8388608 to 134217728
maxssiz_64bit from 8388608 to 1073741824
maxswapchunks from 5148 to 16384
msgseg from 13844 to 32767
nflocks from 3461 to 4096
vps_ceiling from 16 to 64

These changes I haven't changed in awhile, for it is imperative that I feel secure about these changes. Please advise if you find anything wrong here or if you have had any issues with these Oracle dictated changes..
Thanks in advance
Phil

[Steven E. Protter]

I have been running Oracle on 11.00 for a number of years now.

I see nothing wrong with these changes. They make your kernel more similar to the ones I've used in production than before.

If you are concerned about breaking Oracle or your system, I don't think the changes will hurt.

I have never had a problem with Oracle recommended kernel changes except sometimes they don't tell you to raise values enough.

With shared memory, changes are somewhat subjective, because you need to take into account other shared memory apps running on the server.

SEP

[Jeff Schussele]

Hi Phil,

Not much danger there, but the maxswapchunks doesn't need to increase UNLESS you are actually increasing swap size.
Now IF you see a recommended change for timeslice=1 DO NOT DO IT.
Leave timeslice=10

My 2 cents,
Jeff

[Ravi_8]

Hi Phil

your kernel values suits for Oracle 8i.

No harm in changing the values.

[Stephen Andreassend]

Why is there a recommendation against the popular recommendation to change timeslice?
What experiences have you encountered on this matter?

[Tony Drake_2]

The timeslice should never be set to 1.

In previous threads it was said that this change was an error in a common template.

Having timeslice set to 1 causes thrashing, sometimes severely on a CPU loaded system.

What happens is this:

Process a gets on cpu for its timeslice of 1, but can't get done, so it gets swapped out, and another process comes on, and this happens over and over.

Timeslice is also used by HP to manage priority waits by changing the priority based on the number of timeslices a process has been running or waiting. What ends up happening is that no process can ever run to completion because the priorities change so quickly.

This is an over-simplification, but I have seen it on some systems here and it is a nightmare.

Tony