Operating System - Tru64 Unix
1748140 Members
3432 Online
108758 Solutions
New Discussion юеВ

kernel idle explanation

 
Danilo Martinez
Occasional Advisor

kernel idle explanation

Alpha gs60 with 14 cpus.
Tru 64 5.1A, the top tool shows the following output

oad averages: 11.31, 11.68, 11.94 13:17:57
1705 processes:11 running, 50 waiting, 486 sleeping, 1122 idle, 36 zombie
CPU states: 36.4% user, 0.3% nice, 31.7% system, 31.5% idle
Memory: Real: 8939M/14G act/tot Virtual: 1749M/77656M use/tot Free: 230M

PID USERNAME PRI NICE SIZE RES STATE TIME CPU COMMAND
1048576 root 0 0 17G 395M run 17.1H 145.00% kernel idle
1076743 oracle 51 0 9166M 38M run 263:29 99.70% oracle
1076745 oracle 51 0 9143M 5586K run 288:36 98.60% oracle
1555646 oracle 42 0 9135M 4694K run 5:02 34.90% oracle
1417109 oracle 42 0 9135M 4300K run 40:34 24.00% oracle

The question is why the kernel idle is 145%??
Thanks
Dan
7 REPLIES 7
Ralf Puchner
Honored Contributor

Re: kernel idle explanation

maybe it a calculation issue if using more than one cpu..
Help() { FirstReadManual(urgently); Go_to_it;; }
Dave Bechtold
Respected Contributor

Re: kernel idle explanation

Hi Danilo,

I'd suspect the "top" utility is improperly calculating the CPU %. Make sure you have a currect version of "top" and use "ps alxmp 0" to review the "kernel idle" process and all it's associated threads. Also, if you have a GUI see if dxproctuner provides the detail you need.

Hope this helps,
Dave Bechtold
Danilo Martinez
Occasional Advisor

Re: kernel idle explanation

Thanks for your answer,

the ps command doesn├В┬┤t show anything for process id 0

root#com035>ps alxmp 0
UID PID PPID CP PRI NI VSZ RSS WCHAN S TTY TIME COMMAND
root#com035>

Bye D
Dave Bechtold
Respected Contributor

Re: kernel idle explanation

Hi Danilo,

The "ps alxmp 0" command showing nothing would bother me. I tested it on a V5.1A+PK3, V5.1B+PK2, and V4.0G system and all displayed expected results for pid 0 "kernel idle".

What happens if you do "ps -ef" and grep for "kernel", does it find "kernel idle" process?

Dave Bechtold
Danilo Martinez
Occasional Advisor

Re: kernel idle explanation

root#com035>ps alxmp 0
UID PID PPID CP PRI NI VSZ RSS WCHAN S TTY TIME COMMAND
root#com035>ps -ef |grep kernel
root 524288 0 153.7 Oct 05 ?? 1-23:46:43 [kernel idle]
root 739294 731965 0.0 00:13:43 pts/4 0:00.01 grep kernel
root#com035>ps alxmp 0
UID PID PPID CP PRI NI VSZ RSS WCHAN S TTY TIME COMMAND
root#com035>
Hein van den Heuvel
Honored Contributor

Re: kernel idle explanation

> Alpha gs60 with 14 cpus.

Is it a GS160 with 14 cpus, or a GS60 with 4
cpus? Or a GS140 with 14 cpus and just 1 memory module? But there is 8GB memory, and the largest Turbolaser memory was 4GB per MS7CC-GA module no?
So it's a Wildfire with 14 cpus and just 8GB?

> The question is why the kernel idle is 145%??

Because you do not have enough memory and probably not enough memory interleaving.

> CPU states: 36.4% user, 0.3% nice, 31.7% system, 31.5% idle
>Memory: Real: 8939M/14G act/tot Virtual: 1749M/77656M use/tot Free: 230M

That's way to much system time, and too little free memory. Your system is likely to be paging is brains outs, causing all sorts of serilization backups. You are likely to be in 'non-linear' overload space. Reduce the load and see CPU usage go up!.

Please clarify some more about this setup, maybe a vmstat or somesuch, but from what you show here you 'simply' need much more memory. Mere money will fix this problem :-). Be sure to add memory, not replace by larger density. By adding correctly you will increase interleaving which will increase effective CPU speed.
Please contact an (HP) support person to verify my suggestion and help you verify the sizing of your system.
If indeed more memory is needed and added, be sure to generously add some to the Oracle SGA(s) to hopefully increase cache hits and further reduce IO wait time / idleness.

Good luck,
Hein.
Johan Brusche
Honored Contributor

Re: kernel idle explanation

"ps alxmp 0", showed nothing in this case because the system is a cluster member.

"ps alxmp 524288", would have given the
desired result, for this node with memberid=1

Johan.

_JB_