nbuf

[Jeffrey F. Goldsmith]

Last week my manager was tring to figure out why our server was running slow. He found that the nbuf was at 100% so he changed it from 780 to 8000. Now our paging has increased 10 fold. We are going to reduce the nbuf back to 780 but does anyone know how to increase performance on my server.?

L2000 with HPUX 11.0

[Pete Randall]

Do you have Glance installed? Look at memory, processor utilization, and I/O contention. Those are the three big ticket items. If you're waiting on any of those, that's what you need to address.


Pete

[S.K. Chan]

Get the performance cookbook from this thread, it'll a good read.
http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xb8225bd3782dd711abdc0090277a778c,00.html
Hope it helps ..

[Mark Greene_1]

nbuf works in concert with bufpages. They can both be set to zero to do dynamic page allocation, but I'd call your software vendor first to make sure they play well with that.

Glance is your best bet to look at performance. You can do other things like run ipcs -ca |wc -l once every 30 minutes during the day for a week or to in order to gauge semaphore useage. Run an uptime and vmstat 2 10 as well, then you can do some correlation between the three to see how memory is being used during the day.

HTH
mark

[Zafar A. Mohammed_1]

You also have to check dbc parameters and check other related parameters.

Check this links also:
http://docs.hp.com/hpux/onlinedocs/939/KCParms/KCparam.FilesysParmsOverview.html

http://docs.hp.com/hpux/onlinedocs/939/KCParms/KCparams.OverviewAll.html

Thanks
Zafar

[Wodisch]

Hi,

"nbuf" tells your kernel how big the UNIX buffer cache is to be. And the UNIX buffer cache is NEVER paged, so by increasing it, you give away RAM, which then is not available to your running processes. Hence they start to page :-(

FWIW,
Wodisch

[James R. Ferguson]

Hi Jeff:

You clearly created memory pressure when you increased the amount of memory that filesystem buffers were allowed to use.

Stephen Cuillo' performance paper as memtioned by S.K.Chan is certainly a good read.

Often, tuning is a balancing act between memory. I/O and processor as you have seen here. While appropriate kernel parameters help, they really only act as fences or limits. Good coding makes the best use of the resources. Bad coding (or bad SQL) can consume everything.

Regards!

...JRF...

[Jeffrey F. Goldsmith]

I do have glancepluse have been using it to see what my performance is. I just rebooted my server because I changed nbuf back to 780. When I looked at nbuf via glancepluse i see that there are 1050 available and 100% is being used. What else should I look at?

[A. Clay Stephenson]

I strongly urge you to set nbuf to zero. With a non-zero bufpages that produces the optimal setting in almost all cases.