High CPU utilization by NICE

[Phani_4]

Hi,

Server: K380 ( 6 CPU)
OS : HP-UX 11.11

From TOP o/p NICE field consuming high CPU utilization, almost 60% it is consuming, where as USER - 15 % & SYS - 25 % ...

But when i look at vmstat o/p the CPU idle time is 100 % & the CPU run queue length is varying from 8 - 11

Can somebody help me why CPU is busy with NICE...because of the reason the CPU is IDLE for 0 %.

Looking forward for your prompt reply.

Regards,
Phani

[Nicolas Dumeige]

I don't get how nice can consume CPU as it's a priority envelop for another command...

[KapilRaj]

This is from my readings thru the forum . I think u need to apply some patches for top !. There were many issues posted for the same problem and ultimate solution is some patches !

Kaps

[Jeff Schussele]

Hi Phani,

All NICE means is that procs are not running at the default priority of 20. They've probably been started in the background which shifts priority to 24. IF the run queue is always occupied, then I seriously doubt the idle IS actually 100%. Can't be or you'd have nothing in the queues. They key to CPU perf is actually the response times & the priority queue. IF the pri queue builds it means that procs are getting bounced out of there CPU time & they're not getting their work done while there. Run queues can be normally quite high with no impact, but how does glance/gpm show the pri queues?

Rgds,
Jeff

[Phani_4]

Hi Nicolas,

Please find the attached which is having CPU statistics, which were taken by TOP, VMSTAT & SAR -U

Regards,
Phani

[Phani_4]

Hi Jeff,

Please find the attached two screen shots of Glance on CPU.

Regards,
Phani

[Nicolas Dumeige]

You're right, there's something wrong with vmstat output. How 11 process would be waiting on CPU resource with at the same time CPU idle 100% ?

I guess the only possible reason would be CPU affection to a specific user or project.

But the other tools seems to give quite a different perspective on the activity.

You must also have noted that the runQ and the load average show different result : 10 versus 1.58. Whith 6 CPU, the load average is quite low and doesn't seem coherent with 11 process waiting on CPU time.

Adding all CPU consumption pourcents give 592.51% with is close to full CPU utilisation.

sar also show activity with one figure idle time for all CPU. The activity is spread across all CPU.
--> vmstat output is very doubtfull.

Nicolas

[Phani_4]

Do we have any fix for this abnormal behaviour.

Phani

[Jeff Schussele · Accepted solution]

Hi (again) Phani,

Yep - you need patch PHCO_25979 - URL here:

http://www2.itrc.hp.com/service/patch/patchDetail.do?BC=patch.breadcrumb.main|patch.breadcrumb.search|&patchid=PHCO_25979&context=hpux:800:11:11

Should fix you right up.

HTH,
Jeff

[Phani_4]

Hi Jeff,

After installing the above said patch, now vmstat o/p looking fine on CPU...

Now what should in need to fix the NICE CPU utilization behaviour..why NICE is consuming more CPU..

Please help me.

Regards,
Phani

[Jeff Schussele]

Hi Phani,

As I stated earlier, it's not NICE consuming the CPU. It's that processes running at priorities *other* than the default 20

Widen your screen as far as possible & run the following

ps -efl | more

Then check the PRI column. I think you'll find that there are a LOT of processes that are *not* at 20 - but at higher (>20) or lower (<20) priority. And *all* these will be reflected in the NICE column. These are the processes that are putting the load on your system.
ONLY processes at 20 will be in the "normal" columns.

HTH,
Jeff

[Phani_4]

Hi Jeff,

Thanx a lot for your clear explaination..

Regards,
Phani

[Jeff Schussele]

Ooops - I had that backwards.
PRI values >20 are *lower* priority & <20 are *higher* priority.
Rule is lower the number - higher the priority.

Rgds,
Jeff