Problems with performance!

[Lorenzo Facello]

Hi All,
at the moment we run an application using API requests, the processes take memory without release it!!!!... until the all memory is occupied.
we have a UNIX 11.0 - oracle -
in attachement you find some output info:
sysdef
swapinfo -a
top
Can I change some kernel param to solve the problem??
any suggestions??
thanks for help
L.F.

[Sridhar Bhaskarla · Accepted solution]

Hi,

If the process is taking memory without releasing it later, then it's called memory leak unless the process really needs it. You will need to work with the developers/vendors to get it fixed. To confirm the issue run

UNIX95= ps -e -o 'vsz pid args' |sort -n

periodically and observe the processes whose vsz values are gradually increasing.

Your total swap utilization (swapinfo -t) is around 52% which is enough. You also have 1.3GB of free memory available. Your CPU usage is very low when you ran your top. Since you have configured static buffer cache of around 600MB, you should be alright in that area also.

Only issue is with your timeslice value which is 1. I suggest you set it back to 10 which is the default value. Other than that I don't see any issues with your configuration.

-Sri

[Steve Steel]

Hi


Look at

ftp://eh:spear9@hprc.external.hp.com/memory.htm


Good info and tools


Steve Steel

[Lorenzo Facello]

Thanks guys...
interesting stuff!
Now I need to launch the command
UNIX95=1 ps....
when we do some test... we should notice something strange when we get the problem...
just a question...
but the sz has to 1/4 of the vsz?
thanks again
L.F.

[Fred Ruffet]

Have a look in man ps.

sz is size in pages and vsz is the same size in KB. As a page is 4KB...

Regards,

Fred

[Bill Hassell]

The man page for ps is very helpful here. VSZ memory is displayed in Kbytes while SZ is displayed in memory pages (4Kbytes each). Since memory pages are pretty strange for humans to figure out, use VSZ. But the important note here is that your application(s) need rewriting. There is nothing you can do in HP-UX to prevent the programs from (mis)using memory.

Note also that using all of memory is not necessarily a bad thing...unless the memory is not being used efficiently. Your swapinfo shows that not all memory is used and no swap space is used. When your memory is full, quiet processes will be rolled out to the swap area. Use the command:

swapinfo -tm

to see your total virtual memory. You probably have dbc_max_pct set to 50 in the kernel. Change that number to approximately 400-600 megs--note that dbc_max_pct is percentage of total RAM so you'll have to calculate the percentage value.