Kernel parameter tuning comments please

[Gavin Clarke]

I've got a couple of L Class servers in a serviceguard cluster, runnning an oracle 8.1.7 database, 4G memory, what else do you need to know?
Attached are the tunable parameters pretty much as delivered, I've changed timeslice to 10 on recommendation (and having looked at the forums quite a bit) and semume to 4096 which seems rather high to me (what happens if it is too high?) The rest I'm hoping you'll give me some hints with.

Well over to you, I hope you're all up to your usual levels of generosity with advice.

[Kent Ostby]

Gavin -- I'll leave the kernel parameter recommendations to others, but I would recommend that you up the NODE_TIMEOUT value for the ServiceGuard cluster to 8 seconds (8000000 milliseconds in the cluster ascii file).

This will prevent some types of cluster reformations if your system experiences heavy network traffic or "mini-hangs" of the Operating System.

Best regards,

Kent M. Ostby

[Sridhar Bhaskarla]

Hi Gavin,

Your bufpages is looking interesting. It is set to NPROC*3. It is coming to 664*3= 1992 pages.

So, the buffer cache you have is 7.9 MB. Have a glance session open, type "m" and see how much is being listed under buffer cache. If it is only 8MB, then your application may slow down unless you are using raw partitions for your database. Also do not put a formula for bufpages.

I did a quick calculation. Verify it with glance.

-Sri

[Hai Nguyen_1]

Your maxswapchunks is set as follows:
maxswapchunks 4096

Is it true that you have a 8GB swap space?

Hai

[James R. Ferguson · Accepted solution]

Hi Gavin:

To enable the dynamic buffer cache (I would) you need to set *both* 'nbuf' and 'bufpages' to zero (0). Your 'nbuf' is non-zero.

Also, your 'maxvgs' of 80 is very high. If you really don't need 80 volume groups (!), I'd reduce that to something reasonable. Ten (10) is the default which allows volume groups 0-9. About 4-8KB of kernel locable memory is consumed for each increment in 'maxvgs'.

Regards!

...JRF...

[monasingh_1]

I will change maxswapchunks to max so that in future you do not have to reboot the machine.

Also check your glance output and see how yeur system parameters are being used (t: System table report) and see if anything is close to 75-80, then increase them as well.

If you are using 11i then few parameters are dynamic but at 11.0 you should change the potential parameters as mentioned above to a good enough value to avoide a requirement of reboot.

hope this helps..

[Gavin Clarke]

This is all excellent stuff, I forgot to mention that the L Classes run HPUX 11.0.

It's true there's 8G swap space and 1992 buffer pages. I think I'll be changing nbuf to zero and maxvgs down. NODE_TIMEOUT I'm a bit worried about, this is because serviceguard is new to me, I suppose it can't hurt.
I will also be changing maxswapchunks to max although perhaps after I've seen what the other changes do.

As for the system table report, this is also new to me, would I be right in saying the system needs to be under heavy load for these to show problems?

I awarded points by looking at which I could do most immediately, although all of this is really good information.

Sorry about the slow response, I've been out of the office for a few days.