topic Re: process performance in Operating System - HP-UX

process performance

jpcast_real — Thu, 29 Apr 2004 03:43:27 GMT

Hello ,

I must know why a process in a system takes so much time . The same process in another system is very fast . For this reason I must trace it. I must know how many kernel parameter resources is the process using at a certain period of time , I am thinking about using sysdef or kmtune. Are this commands reliable?

The problematic process makes an access to the share memory . How can I trace where is the bottleneck ??? Tusc , sar , ps .....

I need and advice from an expert...

Thanks in advance

Re: process performance

Steven E. Protter — Thu, 29 Apr 2004 03:47:42 GMT

http://hpux.connect.org.uk/hppd/hpux/Sysadmin/tusc-7.5/

That product will let you collect the data you need.

SEP

Re: process performance

jpcast_real — Thu, 29 Apr 2004 03:55:51 GMT

Steven,

do you know if it works for HP-UX 11i over a PA-RISK architecture.

Thanks

Re: process performance

Bharat Katkar — Thu, 29 Apr 2004 04:19:15 GMT

See the attached PDF, hope it helps u out!!

Re: process performance

jpcast_real — Thu, 29 Apr 2004 04:31:40 GMT

How can I know , if in a certain moment a kernel parameter size has been exceeded...????

Sysdef ????

Re: process performance

Nicolas Dumeige — Thu, 29 Apr 2004 04:47:38 GMT

Javier,

I tought kernel parameters couldn't be exeeded, if they're not enforced, there serve no purpose.
If a process exeeded fixed limits, the kernel will stop it. If that the case of the process you're tracing, you should get an error or a core dump.

If you think that the difference between the two execution time is due to kernel parameters, why not redirect sysdef to flat files on both server and use diff to check the differences.

Cheers

Nicoas

Re: process performance

jpcast_real — Thu, 29 Apr 2004 04:51:08 GMT

I think I have a memory bottleneck . I include here my researches:

Dartanan:/tmp/javi> sar -u 5 5

HP-UX Dartanan B.11.11 U 9000/800 04/29/04

11:38:37 %usr %sys %wio %idle
11:38:42 50 0 0 49
11:38:47 50 0 0 50
11:38:52 50 0 0 50
11:38:57 50 0 0 50
11:39:02 50 0 0 50

Average 50 0 0 50
Dartanan:/tmp/javi> vmstat 5 5
procs memory page faults cpu
r b w avm free re at pi po fr de sr in sy cs us sy id
2 0 0 173928 461115 24 13 15 0 0 0 59 471 1412 311 6 1 93
2 0 0 173928 461067 4 0 1 0 0 0 0 381 670 292 50 0 50
4 0 0 176304 461064 0 0 0 0 0 0 0 367 530 295 52 0 48
4 0 0 176304 461073 0 0 0 0 0 0 0 379 638 329 50 0 50
4 0 0 174772 461073 0 0 0 0 0 0 0 379 496 307 50 0 50
Dartanan:/tmp/javi>

Dartanan:/var/adm/syslog> echo total_lockable_mem/D | adb64 -k /stand/vmunix >
total_lockable_mem:
total_lockable_mem: 636895
Dartanan:/var/adm/syslog> echo phys_mem_pages/D | adb64 -k /stand/vmunix /dev>
phys_mem_pages:
phys_mem_pages: 786432

Re: process performance

Pete Randall — Thu, 29 Apr 2004 04:53:08 GMT

Javier,

The definitive tool is Glance. If you don't have Glance, there is a free 30 (or 60) day trial version on your application CD's. Install that, then fire up gpm and use the gui version to zero in on your process. It can tell you more than you ever wanted to know about what it's doing!

Pete

Re: process performance

Ted Buis — Fri, 30 Apr 2004 09:40:40 GMT

I can't resist putting in a plug for book,
HP-UX Tuning and Performance by Robert F. Sauers. For someone who is unfamiliar with the various tools and their use, this is very useful. Most importantly it provides a sound methodology for performance troubleshooting.