topic Re: Kernel parameters in Operating System - HP-UX

Kernel parameters

Lorenzo Facello — Wed, 24 Nov 2004 09:09:58 GMT

Hi everybody,
I have some troubles with 4 kernel param
I have an hpux 11.00 with oralcle and some application
4 CPU
6GB RAM
4GB SWAP

Timeslice - on my server is set to 1, but I read it should be set to 10. isn't it?

bufpages and nbuf -
I have 4692 for bufpages and 8092 for nbuf
I read if you put zero values you get the dynamic buffer cache... but I would like to know how these 2 param work and how they can increase the system performances... if it's possible!

swapmem_on - is set to 0... it should be to 1??

in attach some info

Thanks in advance for help!
L.F.

Re: Kernel parameters

Steven E. Protter — Wed, 24 Nov 2004 09:12:15 GMT

My timeslice is set to 10. System runs nicely

My swapmen is set to 1. This enables swap to be reserved, device and filesystem.

Again, production system running well.

SEP

Re: Kernel parameters

RAC_1 — Wed, 24 Nov 2004 09:17:25 GMT

timeslice - leave it as 10. (This is default and do not change it.) If oracle has recommanded you to set to 1, do not do that.

bufpages and nbuf - If you want to use, dynamic buffer cache, set those two values to 0 and set parameters dbc_max_pct and dbc_min_pct. In most cases (with oracle) setting of 10 % and 5 % suffices. If values of bufpages and nbuf are non zero static buffer cache is set.

swapmem_on - With 6GB RAM, I would not set it to 1. Setting it to one, inflates the swap space value. (swap space + 75 % of RAM) This is just for kernel calculation. It is helpful, if your system is not oging to swap at all. Proves very helpful on systems with large RAM. (for me large RAM starts at atleast 12 GB)

Hope this helps.

Anil

Re: Kernel parameters

John Poff — Wed, 24 Nov 2004 09:17:56 GMT

Hi,

I think the Monolithic Database kernel parameter set used to set the timeslice to 1. As mentioned you should put it back at 10. We had a box that had it set to 1 for a while. It didn't run noticeably faster with the timeslice at 10, but long running jobs that contended for CPU time did seem to do better.

JP

Re: Kernel parameters

Victor BERRIDGE — Wed, 24 Nov 2004 09:19:14 GMT

Hi,
Since you have less device swap than RAM swapmem_on should be set at 1 in order to use for swap what the extra...
Now reduce your buffers:
set
dbc_max_pct 10
dbc_min_pct 3

Once that done you can try with a bigger SGA and tune the way you mount your file sytems if you use vxfs and have onlineJFS

All the best
Victor

Re: Kernel parameters

Pete Randall — Wed, 24 Nov 2004 09:19:30 GMT

Lorenzo,

First, timeslice should definitely be set to 10. This should never be changed.

From SAM's "Help on Configurable Parameters":

" * bufpages = 0, nbuf = 0: Enables dynamic buffer cache.

* bufpages not zero, nbuf = zero: Creates bufpages/2 buffer headers and allocates (bufpages x4 Kbytes) of buffer pool space at system boot time.

* bufpages = 0, nbuf not zero: Allocates nbuf*2 pages of buffer pool space and creates nbuf headers at boot time..

* bufpages not 0, nbuf not zero: Allocates bufpages pages of buffer pool space and creates nbuf buffer headers at boot time. If the two values conflict such that it is impossible to configure a system using both of them, bufpages takes precedence."

The setting of swapmem depends. On systems with lots of available memory it is typically enabled. On smaller systems and workstations, it might be better to disable it.

Pete

Re: Kernel parameters

Geoff Wild — Wed, 24 Nov 2004 09:19:35 GMT

Yes - set timeslice to 10.

Swapmem_on set to 1

http://docs.hp.com/cgi-bin/otsearch/getfile?id=/hpux/onlinedocs/939/KCParms/KCparam.SwapMemOn.html&searchterms=swapmem_on&queryid=20041124-071617

My bufpages/nbuf are set to 0

Rgds...Geoff

Re: Kernel parameters

Victor BERRIDGE — Wed, 24 Nov 2004 09:22:10 GMT

But if you want swpmem_on set to 0, then add more swap so that it equals the RAM...

All the best
Victor

Re: Kernel parameters

Sridhar Bhaskarla — Wed, 24 Nov 2004 09:30:41 GMT

Hi Lorenzo,

As others indicated, set your timeslice to 10 otherwise your system will spend quite sometime in doing context switching.

Interestingly you have set only 4692 pages for bufpages which means you have around 18MB for buffer cache!!!... That's too low. Either set your nbuf and bufpages to 0 and make your max_dbc_pct and min_dbc_pct to 8 and 2 respectively or set your bufpages to 128000 and nbuf to 0 so that buffer headers will be automatically determined.

Set your swapmem_on to 1 otherwise you will not be able to use all of your memory as you have only 4GB of swap.

-Sri

Re: Kernel parameters

Lorenzo Facello — Wed, 24 Nov 2004 09:58:54 GMT

ok.... almost clear...
just 2 question
I'm using the static cache (so nonzero values to bufpages and nbuf)
bufpages 4692
nbuf 8092

It's enough?? or I should change something?
how does it work?
thanks
L.F

Re: Kernel parameters

Sridhar Bhaskarla — Wed, 24 Nov 2004 10:23:09 GMT

Hi,

You obviously didn't read my message. You are only setting 4692 pages of buffer cache which is around 18MB and this is too low. I would set it to around 400-500 MB means, set it to 128000 pages. Also set nbuf to 0 and let the system calculate the buffer headers.

Either you should have the filesystem set to use no buffer cache (using OnlineJFS options) or with optimum size for oracle. Otherwise it can impact the performance.

-Sri

Re: Kernel parameters

A. Clay Stephenson — Wed, 24 Nov 2004 10:33:11 GMT

Your bufpages value is much too low. Your current setting is just a little over 18MB. I would increase it to about 300-400MB if you are using raw/io for Oracle (or the vxfs mount options convosync=direct,mincache=direct) which bypass the buffer cache. If you are using cooked i/o then 600-800MB of cache is about by suggestion.

300-400MB => 76800-102400 bufpages
600-800MB => 153600-204800 bufpages

In almost every case leaving nbuf set to 0 for the default behavior is a good idea.

Currently, your buffer cache is much too small to handle all the routine UNIX tasks well even if all your Oracle i/o is raw. At a barebones minimum, you should increase it to 100MB or so.

Re: Kernel parameters

Lorenzo Facello — Wed, 24 Nov 2004 10:36:15 GMT

ok!!!!
sorry SRI but I didn't refresh the page...
THANKS ALL!!