Kernel Paramaters for HPUX-PA-RISC server with 8Gb Memory

[Gousset]

Could someone provide some kernel tuning for the following server:

Model 9000/800/N4000-75
uname -a HP-UX rmsph001 B.11.11 U 9000/800 633309331 unlimited-user license
# CPUs 4 (including iCOD CPUs)
8Gb Memory

The server is running Oracle 9.2.0.6 and soon to be upgraded to patch 9.2.0.7. Currently we are having many performance issues which are causing frequent core dumps on the platform related to OCI drivers. We hope that upgrading to the latest OCI drivers may fix the solution, but would like to make sure the kernel paramaters are fine tuned. Please find attached the current paramaters:

[A. Clay Stephenson]

It's a little difficult to make suggestions in a vacuum (e.g. are you using raw/io for Oracle?) but a few things jump out at me. You are running a fixed buffer cache (which is fine) but your size is 300Mib (bufpages = 76800) is a bit low. I would bump it up to somewhere in the 800MiB range --- and possibly higher depending upon free memory and use of cooked vs. raw Oracle i/o.

Your maxssiz values are much too large and this will steal space for data -- especially in 32-bit land. Reduce maxssiz to no more than 32MiB and maxssiz_64bit to no more than 64MiB -- only very badly written software would ever need stacks larger than this. I would also bump up maxtsiz to 128Mib and maxtsiz_64bit to 512MiB or so. You should also bump scsi_max_qdepth up and ninode can be reduced to 800 or so to save a little memory. Ninode only applies to hfs filesystems and I'll bet the only hfs filesystem on your box is /stand.

What you should really be doing is gathering metrics; tune, gather more metrics; rinse and repeat.

Now having said all of this, you are going to find that tuning your database and tightening your SQL is going to make 10X more difference than OS tuning.

[Jaime Bolanos Rojas.]

Gousset,

When it comes to hp-ux tuning- oracle tuning, there is a lot of things to take in accound for all the different needs and servers out there.
This forum is filled with question about tuning oracle and unix, as you go search on them, some examples:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1043965

http://forums1.itrc.hp.com/service/forums/bizsupport/questionanswer.do?threadId=1042183

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=708837

Search or post a question on specific performance issues, and I am sure somebody will come with an answer for you.

Regards,

Jaime.

[Steven E. Protter]

Shalom,

How many instances of Oracle and SGA structure are critical to adnwering this question.

If you are already runing 9.2.0.6 you have the required parameters in place.

I'd check:
dbc_max_pct
dbc_min_pct

Make sure they are close together and low. Oracle is better than the OS at caching its data.

I'd check parameters vx_inode (similar) is not set to zero because that lets the OS pick and change things.

You might wish to collect some performance data and look for specific issues.

http://www.hpux.ws/system.perf.sh

SEP

[A. Clay Stephenson]

The dbs_max_pct and dbc_min_pct values are meaningless in this context as bufpages is set to a non-zero value. If you want to run a dynamic buffer cache then set bufpages to 0 and then tune the dbc_xxx_pct values but I prefer to run static buffer caches. By making one of the chief variables (buffer cache size) a constant, it is much easier to measure the effect of other tunables.

[nanan]

It is very hard to suggest a proper parameters without your system performance info

You may need to check where the bottleneck was; mem, cpu or IO etc..

It looks like fine when it comes to your parameter itself.
but you have to find the real effective things out to tune your system properly

you may need to tune SQL if the system show very high CPU usage while the other idle
may to need to adjust mem memangement like cache size accroding to hit rate

....



Use vmstat, sar or glance you can

[Gousset]

Please find attached some performance stats i gathered from running "system.perf.sh". Also find attached the crontab jobs.