perfomence problem with hp-ux server(11i v3)

[periyasamy]

In one of my hp-ux server the cpu idle time in the output of top and sar always showing below 10%.I found three process taking more cpu time
from the top output.That process are belongs to some application.I want to know weather this broblem arrives due to that process or anything else i have to check to find,why processor idle time goes below 10%.if anybody knows pls help me.

I have added the top output as a attachment, pls find for your refrence.

[Tim Nelson]

From your top output ( limited point in time ) proceses wmqatc, wmqoti, and wmqfli are using most of the CPU resources.

Unfortunately I do not know what this application is.

In general, those processes are looking like CPU hogs and research needs to be done with the application support on why they are using so much CPU. From there if application support suggests changes to HPUX parameters then do so.

[Ganesan R]

Hi,

Only the top three process consuming more cpu cycles and it looks those process are belongs to specific application and running quit long time. We cannot comment anything about cpu usage by these processes. It is possible that these processes are cpu hunger and really need that much. In that case you need to consider to upgrade the cpu.

Sometime improperly written code may lead into this kind of cpu bottleneck.

In any case you need to work with that application vendor to find out the reason and fix.

[rick jones · Accepted solution]

If this is an Integrity system you might consider profiling those processes with Caliper:

http://www.hp.com/go/Caliper

otherwise if the system is an HP 9000 then you would use Prospect:

http://www.hp.com/go/Prospect

You might also consider some system call tracing with tusc or even just counts from Glance.

All that sort of data would be useful when you interact with the application author(s).

[Dennis Handly]

>I want to know whether this problem arrives due to that process

You can't tell without detailed knowledge of what the application is doing.
All you know from top(1) is that you should be talking to ceda. If you think that using 90% of a CPU is unfair and you don't have time to ask ceda, you can simply renice those 3 hogs:
renice -n 20 21127 21027 20953

You may want to get the whole process tree for ceda:
$ UNIX95=ENHANCED_PS ps -Hfu ceda

Perhaps knowing the parents may explain what's going on?

>Ganesan: running quite long time.

Yes. If a process is suppose to run that long, periyasamy should have seen it over and over. Perhaps this is just a bug and they need to be killed??

So some steps if you don't want to use Rick's suggestions:
1) Talk to owner.
2) renice so you can get work down.
3) kill -STOP each, so you can resume (kill -CONT) after you talk to owner.

[Bill Hassell]

I will take a different approach: you are wasting 10-18% of your available CPU power. There is nothing wrong with using 100% of all your CPU power. Nothing will start smoking and processes won't crash because all the CPU cycles are being used. Idle time = 0% is a good thing. You are getting your money's worth from the system in terms of CPU cycles.

Another metric that is more useful is the load average at 0.84, indicating that you are not using all your CPU cycles. Many sysadmins will report load averages from 2 to 50 are normal for their systems. It is also important to know whether you can rewrite the application. If not, ask your application supplier how to improve performance (if that is a problem).

[Dennis Handly]

>Bill: There is nothing wrong with using 100% of all your CPU power.

Yes but should these processes take 70 hours?

[periyasamy]

Hi
what is the defrence between cpu ussage(including idle time) and load average from the top output in HP-UX can anybody explain me.

[Wim Rombauts]

The idle time is the time that the CPU is doning nothing.
The LOAD is the average number of processes on the run queue.

As I understand it, this will give the following result :

If you have few processes that run continuously, the length of the run queue (=LOAD) remains low while the idle time is low.

If you have many processes that put themselves on the run queue continuously but run very short as soon as the get CPU, yoy van ave a high LOAD while you still have a high idle time.

[Dennis Handly]

>between CPU usage (including idle time) and load average

Basically CPU usage is a better measure of what the CPU is doing. A load average might start to be useful if the CPU was 100% busy.