ps question on rs6000

zepplin611 · 01-18-2004, 06:07 PM

Greetings AIX'ers

When doing some machine audits on an IBM rs6000 (aix 4.3.3.0 <--- lowest ML), i typed:

ps -aux and the following returned (it has been truncated for shortness here):

# ps aux|more
USER PID %CPU %MEM SZ RSS TTY STAT STIME TIME COMMAND
root 516 24.8 1.0 8 6352 - A Aug 04 239054:44 kproc
root 774 23.7 1.0 8 6352 - A Aug 04 227883:55 kproc
root 1032 23.1 1.0 8 6352 - A Aug 04 222423:01 kproc
root 1290 23.0 1.0 8 6352 - A Aug 04 221316:33 kproc
----- MORE HERE BUT DELETED -----

What are commands: kproc ? And why are the chewing up the cpu percentage?

This is a 4 cpu machine. While typing ' w ' the following returns:
# w
07:09PM up 167 days, 4:11, 8 users, load average: 1.68, 1.39, 1.47

however there are no jobs called kproc running on the machine (that is, jobs called kproc executed by regular users).

Any help is greatly appreciated.

zepplin611

crabboy · 01-18-2004, 10:33 PM

kproc are the kernels idle processes. There is one for each CPU. The PIDs will always be 516, 774, 1032, 1290 for a 4 processor machine.

Here is the output from one of our 4 processor boxes.

Code:

USER       PID %CPU %MEM   SZ  RSS    TTY STAT    STIME  TIME COMMAND
root       516 24.8  0.0   12 16140      - A      Dec 17 46544:50 wait
root       774 24.7  0.0   12 16140      - A      Dec 17 46287:13 wait
root      1032 24.5  0.0   12 16140      - A      Dec 17 45860:19 wait
root      1290 24.3  0.0   12 16140      - A      Dec 17 45555:39 wait

iainr · 01-19-2004, 05:53 AM

Worth mentioning that the CPU is always doing something. The idle process is the process it runs when nothing is happening so it does not show up most stats.

When the CPU becomes busy, idle goes down (of course).

zepplin611 · 01-19-2004, 09:17 AM

thanks crabboy and iainr....i know this was an "AIX" question, but does linux have equally observable 'idle' processes that one can monitor?

thanks again,

zepplin611

jc2it · 02-03-2004, 11:35 AM

Is the size of your idle process percentage a good indicator of how your hardware is handling the software? Also, is this cumulative or the current status? What I mean is does the idle process percentage calculate all of the idle time from the last reboot, or is this just the most recent five minutes?

iainr · 02-03-2004, 02:35 PM

Quote:

Originally posted by jc2it
Is the size of your idle process percentage a good indicator of how your hardware is handling the software? Also, is this cumulative or the current status? What I mean is does the idle process percentage calculate all of the idle time from the last reboot, or is this just the most recent five minutes?

Run vmstat (e.g. vmstat 2). Ignore the first line (which shows stats since the last reboot) and look at the others which show current system performance.

The last four columns are usr, sys, idle and wait. Usr + sys is the total time the CPU is spending doing things. As a rule of thumb, on a normal system with interactive users (i.e. not doing batch processing) you want this to be below 70%. If the CPU usr + sys is consistently high, that's an indication that the system has too little processing power.

The "wait" is the system waiting for I/O. Again a very rough rule of thumb, but on a normal interactive workload you want this to be below 30%. If wait i/o is high, you can use iostat (or topas, or a few other tools) to identify hot spots. Idle in itself isn't very useful - you need to know whether the non-idle time is doing CPU work or waiting for I/O.