High load while idling

[Timo J]

Included is the output of one system(rp2470, 1GB RAM. Why is the load averages so high (4.15, 3.79, 3.71), while the system is really doing almost nothing.

[KapilRaj]

there are some user processes isn't it ?. what is user1 doing with foo and java commands ?.

Regds,

Kaps

[Michael Tully]

commands foo and blah are not doing too much. The problem will most likely be memory. This could be associated with the java processes.
What is the output of:

# swapinfo -tam

and
What is the size of your buffer cache usage?
You can find what it is set to from your kernel parameters.

[Shaikh Imran]

Hi,
This load not to be considered high.
As per the logs attached.
The user user1 is using the java programs and thus you have 4.15%.
And the system is idle for 92.1% which is not bad as per the attached log.
Also at any given point of time there will be some daemons running (cmcld & vxfsd as seen in the log ) so i don't think you will be able to get the ideal results.


Regards,

[Fred Ruffet]

I see two things :

This machine has only one CPU. It can be scheduling issue. I had similar problem running big programs on a one CPU machine, where adding a second CPU had make the server go better. I don't really know how to have proofs of thios problem as long as it doesn't show many process in a running state, and a low nicing percentage. All I can say is that my server looked exactly the same.

Second thing would be to look at IOs. Could you issue a "sar 5 5" and look at wio ?

regards,

Fred

[Trond Haugen]

A high load is a INDICATION of a bottleneck. In your cast it is not CPU.
It could be memory. Run 'vmstat 5 30' and check for page outs (po).
It could be disk. Run 'sar -d 5 30' and check for high disk utilization or large disk queue lengths.
Or if the system is not percived as slow, it could be normal.

Regards,
Trond

[malay boy]

Well,that's what happen to me.The system is slow even after startup and turn up the memory is not enough.You can used vmstat to see the problem.But do get "glance" trial pak to see more.

regards
mB

[Timo J]

Thank you all for fast replies. Included is the requested additional information.

[Trond Haugen]

At the time you ran swapinfo there was literary no swap used. The vmstat output shows one period with 8 po's. So there where just a peak of memory preasure.
that 'sar -d' ouput shows no high disk utilization nor is there a big run queue. So there seems to be no disk bottleneck ether.

There are other posible bottlenecks but if the system is not percived as slow, I would say the load is normal.

[Hein van den Heuvel]

Speculation...
It could be a matter of 'bad timing'.

If some thread wake up periodically, pretty much at the same time, then an outside observer looking in at that same time will always perceived there to be a significant queue for the CPU. That observer might not see that this queue will soon be resolved and the system wil be idle for the remaining 90% of the indicated interval.

At a very high level you see this with folks submitting cron/at job at midnite or on the hour every hour. All you need tro do to fix this is to submit in an hour or in 24 hours allowing the absolute time to drift.
In this case is may be java schedulign a cleanup thread every exact second, whether it needs it or not.

I always wonder about the tele-conferences. Those more often than not start on the whole or half hour. That must create a terrible demand peak!

Cheers,
Hein.

[Ted Buis]

If you want to know why the run queue is high then see which processes are block and why they are blocked, but I don't think you have a problem necessarily. If application 1 locks a file that other applications want, those applications will be blocked until, it is released by application 1. So, it is very easy to have a senario where many processes are blocked, but this is not a system problem. If the users are complaining then maybe the application needs to be rewritten to improve performance, but the operating system can do nothing.