UNIX buffer cache limit

[Dan Moloisane]

I have a 40 GB RAM, 16 X 1000GB CPUs rp7420 server.
I need to tune the UNIX buffer to a most recommended size.

PLS advice on how much can you set the size.

[RAC_1]

Depends on what you run on the system.
With oracle, I tune it max to 800mb. ORacle does it's own buffering, so setting it at 800mb is enough for OS cache buffering.

Also I set dynamic cache. Use of dbc_max_pct and dbc_min_pct

[Arunvijai_4]

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=721990
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=954245

Some good threads to learn ..

-Arun

[Stuart Abramson]

We always set dbc_max_pct to 10%. (min defaults to 5% I think.) But you have a BIG RAM, and may not need that much. Start with 10% and see how it goes.

[Geoff Wild]

You can hard net nbuff to 800 MB...

And/or just set the dbc_max_pct and dbc_min_pct to a setting that works out to somewhere between 500 - 800 MB of ram....

2% for max and 1 % for min would work.

Rgds...Geoff

[Ted Buis]

If you have OnLineJFS, there are mount options so that you can avoid double buffering from Oracle and HP-UX which wastes RAM and time doing memory to memory transfers.

[Dan Moloisane]

Thanks Guys for your effort.

[Andy Torres]

This attachment has great info on tuning buffer cache, as well as other commonly misconfigured resources:

http://docs.hp.com/en/5580/Misconfigured_Resources.pdf

[Dan Moloisane]

Thanks for the responses.
One more question,
I have OlineJFS currently installed on HP11.11. I need Oracle 9.2.0.4 to bypass UNIX buffer cache.What commands and options do I need to use and can it be done online while the database filesystems are mounted and Oracle is running.

[Orhan Biyiklioglu]

Check this thread:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=248308

hth

[RAC_1]

The options that you require are convosync=direct and mincache=direct. This can not be done online. (in fact there -o remount option, but does not make sense in your case)

Mount with above options.

[Jean-Luc Oudart]

Hi

just a note. You should patch your Oracle database server to latest.
9.2.0.7

cf Metalink Note:189908.1 "ALERT: Oracle9i Release 2 (9.2) Support Status and Alerts"

Regards
Jean-Luc

[Tim Sanko]

Dan,

What it really depends on is your mass storage. Is it an EMC DMX, if so minimize your buffer cache as it is basically irrelevant. If it is on a jbod, max it as far as you can and still have stuff work...

The real key is to look at the wait states and check sar while the DB is running.

If there is an unacceptable number of disk io waits, it is time to bump up the buffer cache.

If it has a low amount of hits it may not matter about the way things are configured...


run sar -a and check the memory stats, disk stats and cpu stats...

Tim

[Dan Moloisane]

Ok, my dbase runs as a package on a cluster environment. How do I add the meancache=direct and convsync=direct options within the package control file.
The line looks like this within the control file:
LV[3]=/dev/vg01/lvol4;
FS[3]=/mp2hc500/pgt1/dbase01; FS_MOUNT_OPT[3]="-o rw"; FS_UMOUNT_OPT[3]=""; FS_FSCK_OPT[3]=""; FS_TYPE[
3]="vxfs"

Thanks