which kernel parameter for user memory?

[Tan Shirley]

Hi,

Would appreciate if you guys could advise me which kernel parameter I can tune in my HP-UX v11.00 so as to reduce the amount of memory each user process is taking up. I have two similar HP-UX machines, but one is N-class, the other is L-class. They have the same applications running, but the N-class is serving much more users as compared to the L-class. From the Glance in my L-class, I noticed that the RSS for each user process is in mb, whereas it's in kb in my N-class machine. I can't tune the kernel in L-class to be exactly the same as the N-class because the N-class is used by much more users and has got more databases.

In a nutshell, I'll need to tune the kernel in my L-class so that each user process takes only a few hundreds kb reflected under the RSS column in the Glance

Thanks so much !

Regards,
Shirley

[U.SivaKumar_2]

Hi,
Setting dbc_max_pct to 20 % will increase the user memory space.

regards,
U.SivaKumar

[Dietmar Konermann]

Hi, Shirley!

You are talking about the RSS, the resident set size of your processes. Usually the kernel determines the working set size of the processes which is the key factor for their RSS. You cannot easily restrict this directly from the OS side (unless you use things like 11.11 memory resource groups).

What you can change directly are things like maxdsiz/maxdsiz_64. But they do not restrict the RSS... here were talk about the vritual set size VSS.

Maybe you should have a look at the memory management white paper for more details. Not an easy stuff! :)

http://docs.hp.com/hpux/onlinedocs/os/11i/mem_mgt.html

Regards...
Dietmar.

[Mike Stroyan]

vps_pagesize is the most likely parameter to cause processes to have more RSS. The default value is "4", meaning that pages default to 4KB each, (4096 bytes). If the vps_pagesize value were very high then all pages would default to large sizes. A value of 1024 would make every page bring in 1MB of RSS. That would make "small" processes consume at least 3MB for one page each in its text, data, and stack regions.
While some programs get a big performance boost from large pages, it can be much more efficient to control that with "chatr +pd +pi" settings for particular programs instead of forcing all processes to use large pages.