What Memory Kernel Parameters to check as server using too much RAM

[amreek bansal]

Can someone kindly help.. We haev HP-UX server which has 32gb RAM and 13gb swap space. Its runs databases and has connections from OAS and Staffware, ist maxises out on RAM quite a lot. Swapinfo -tam always shows heavy ram usage, in the 90's.

Can any one advise which kernel parametes for memory I can check and see if the settings maybe incorrect, as we suspect the server is some how using the RAM inefficiently.
Thanks.

[Sajjad Sahir]

Please see this link

it will help u

http://docs.hp.com/en/939/KCParms/KCparam.PagingParamsOverview.html


sajjad

[Ganesan R]

Hi,

Check dbc_min_pct and dbc_max_pct values. dbc_max_pct might have set to default value of 50.

If you don't need that much buffer cache you can reduce to 10-15.

This value defines % of memory to be reserved for buffer cache.

[amreek bansal]

Thanks for the replies.

Ive checked settings for:
dbc_max_pct
dbc_min_pct

A while back we changed dbc_max_pct from 50 to 5. The dbc_min_pct was NOT changed, it remains on 5. Is there any problem with having both max and min on 5 ?

[likid0]

You can install kmeminfo(if you don't have it),
download from here:
ftp://eh:spear9@hprc.external.hp.com/memory.htm

and with:
#kmeminfo -- you can check your user/system memory usage
#kmeminfo -u -- memory per process usage

you could also use glance.

check out the process that are consuming more memory.

There are server parameters you can change to reduce the memory utilization, but you should first start by checking if the user process are using the ram they should.

[amreek bansal]

okay thanks, I'll look into kmeminfo, but you say there are several paramters that can be changed so memory, would be please be able to list these ?

[Ganesan R]

Hi Again,

>>Is there any problem with having both max and min on 5 ? <<

Obsolutely no. What you have done is static buffer.

If dbc_min_pct and dbc_max_pct is different value, it is called dynamic buffer. When the system is initially booted, the system allocates dbc_min_pct (the default is 5 percent) of memory for buffer pages. The size of the buffer cache will grow until dbc_max_pct value as new pages are brought in from disk.

For better understanding read this document.

http://docs.hp.com/en/7779/commonMisconfig.pdf

[likid0]

amreek bansal, in the pdf ganesan has linked on his reply you can check out some parameters to reduce memory use. But you have to think that when you lower kernel parameters, you may be loosing performance(inode buffer cache for example) so you have to check out you are not taking a big performance penalty.

But for example vx_ninode is a kernel parameter that the default(not allways, depends on you app) is to high, and by lowering it you can save some memory(not to much).

Take a look at the pdf it's well written.

[Bill Hassell]

You might lower dbc_min_pct to 2 (the lowest value). However, with 32 GB full, there is really nothing you can change in the kernel that make any visible difference. You might save a few MB here and there but nothing that will make a difference for 32 GB. You have three choices:

Add a lot more RAM or, tell the database admin to reduce RAM usage or, stop running so many processes on the system.

[Steven E. Protter]

Shalom,

Note that 90% memory use is not in itself a problem.

Lowering the buffer pool as recommended above can free some memory, but you should run some tests to see if that move benefits your performance.

http://www.hpux.ws/?p=6

SEP