High LoadAverages are certain CPUS

[Frank W. Smith]

Environment: N4000/8 way HPUX 11.0 June 2000

Certain CPUS show very high loadaverages others are normal.
CPU 0, 4, 5, 6, and 7 low Average of 1.5 or less
CPU 1, 2, and 3 high Average of 6 or higher

Reboot corrects but problem comes back.

How can I find which processes are causing the que?

[Vincenzo Restuccia]

top shows CPU utilization processes.

[Frank W. Smith]

I'm very familiar with both the TOP utility and the GLANCE utility. The GLANCE utility actually is the only one that shows the problem on the CPU by Processor screen.

Neither glance or top indicate any busy processes. I'm trying to determine what makes up the runque. I assume a process is active but stuck for some reason or an HPUX bug?

[Steven Sim Kok Leong]

Hi,

glance will be able to show you the processes that are running on each cpu.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[David Child_1]

Frank,

Top should show you which processes are running on which CPU (see example -- 1st column in process list). Make sure that top isn't aliased to 'top -h' as this will hide individual CPU stats.

Once you discover a process that is suspect (like PatrolAg which is running on CPU 6 in my example), take a look at that using glance for more detail.

[Frank W. Smith]

Dave:

I'm not getting my point over. Looking at your top you have a slight inbalance in the loadaverage. CPU 0 is 2+ and most other CPUS are <1.

In my situation CPU 3 has a 6+ loadaverage, and the other CPUS have < 1.5. I don't see any processes in CPU 6 that are heavy consumers, and I don't understand why ONE CPU would be so heavily loaded with other almost idle.

Thanks,

Frank

[Steven Sim Kok Leong]

Hi,

Are you running predictive? Predictive had a problem with its daemons (eg. memlogd) occupying 99-100% of CPU time on a single CPU. A patch resolves the CPU hog.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Thierry Poels_1]

Hi,
just want to mention: if you have one super mega heavy process you will have one overloaded cpu, whilst the others might be idle. Your server cannot divide the load of one single process over several cpu's. The process might however be switched to other cpu's regularly.
But you measures seem to be extreme, and need a closer look. Like mentioned, predictive might be the cause.
good luck,
Thierry.

[Paula J Frazer-Campbell]

Hi

Fire up STM, highlight the cpus, select tools, exercise and run.

This will check out the cpus.

Also as previously mentioned cpus do not share what might be a load hungry process.

Paula

[Tim D Fulford]

Try glance with s option to select a process. This shows which CPU the selected process is on. This is not quickest way but it works.

I assume you want is a list of the processes running on each CPU at any given point. From this you can look at the individual processes & track down the culprits.

I would imagine that something like processor affinity is switched on for some of your processors to bind them to 3 of your 8 CPU's!

Good luck

Tim

[Asad Malik]

HI
check whether these multiprocessor patches are installed. PHNE_21433 and PHNE_21767. We have the same issue and it was resolved by installing these patches.

[Todd Larchuk]

I think what you are seeing is a difference in the averaging time.

What measurement interval are you using for Glance? If you increase the measurement interval of Glance to something very long, say 5 minutes, then you will see a very nicely balanced AVERAGE load. If you could have an instantaneous measurement interval, you would see some cpus at 100%, busy, and others at 0%, idle. That is what is really happening, but what performance tool report is an average value rather than an instantaneous value. The longer the averaging time, the smoother the values are, usually.

The values that 'top' reports are very heavily averaged. In fact, the cpu utilization is the exponentially weighted averages that the kernel uses to assign the process priorities. If you watch top while a process starts, executes, and stops, you will see cpu values that slowly ramp up, stay high, and then slowly ramp down. This is not what the process is really doing! This slow ramp up and down is due to the exponential averaging of the cpu utilization.

[Sandip Ghosh]

You must have already checked but to make it sure, have you checked if there is any runaway process or not which is hogging that CPU?

Sandip

[Bill McNAMARA_1]

what application are your running.. does it do mpctl's (ie push process execution onto certain CPUs) or is it the kernel doing it all.

I can post up a program (or you can search for it here - processor affinity) that will allow you to push a PID onto a certain CPU.

Later,
Bill