Strange High Load on idle system

[Koldenhof]

Hi,

I have a HP-UX B.11.11 U 9000/800 system here in the office.

We use it for testing our application, but it has a problem. When I do a TOP command on it, the system has a load of 0.60-1.50 when i expect it to be more around 0.00 - 0.09 becuase there are no processes putting any load on the system. On another system i do see the expected load average of 0.00 - 0.09. I have attached a TOP output

[Luk Vandenbussche]

Hi,

The load is OK.
Your CPU is 100 % idle

LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS
0.46 0.0% 0.0% 0.0% 100.0% 0.0% 0.0% 0.0% 0.0%

No reason to worry

[Ranjith_5]

Hi,

IDLE 100% means there is no load on the system. This is absolutely right. No need to worry. You need to be concered if there is high values for CPU, SYS, USER, LOAD.

Your top output is absolutely fine since there is no load on your system at present.

See more about the top command here in the man page.

http://docs.hp.com/en/B2355-60105/top.1.html

Regards,
Syam

[Koldenhof]

Hi All,

I am still not conviced that these numbers are of no concern. Why is the load average displaying values around 0.60 to 1.2 when the system is idle ? I have another HP-UX system which shows load averages 0.00 to 0.02 when the system is idle. Also all my Linux boxes always show load averages of around 0.00 when idle.

Please advice on why the load average is displaying these values.. I know the system is completely idle because when i actualy look at the system it's quiet ( as far as a server can be quiet ;)) and there is no harddisk activity or network activity.

[Ranjith_5]

Hi,


For a better accurate data collection on performance you need to use glance or OVPM by HP. This will give you an exact idea.

Regards,
Syam

[Torsten.]

If a system is really *completely* idle, it is switched off or frozen.

There are *always* some running processes. Your attachment shows 5. Even if you want to measure the activity, you will need resources for that and you will enforce activity.
BTW, top is showing load averages. See man top for more information.
There is no reason to worry about.

[Koldenhof]

Thank you all for your responses.

However, I still have not received an answer to my question.

To illustrate a bit more, I have captured a top from a system which has the expected load average.

Load averages: 0.00, 0.00, 0.00
133 processes: 112 sleeping, 20 running, 1 zombie
Cpu states:
LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS
0.00 0.0% 0.0% 0.0% 100.0% 0.0% 0.0% 0.0% 0.0%

As you can see, there are also 20 processes running. But, the load average is as expected, 0.00

Why could this be so different as on the other system ?

Could it be that the system get a very large amount of inboud network traffic it's handeling in the background ? Or might it be that it's constantly swapping memory pages in and out ?

[Bill Hassell]

The "load" is not what you think it is. The load value is the average runqueue size, that is, how many processes were running or waiting to run over the measurement period. The percentage CPU time is measured in a different manner and thus the discrepancy. The most common cause of a non-zero load with all zeros in USER and SYSTEM is automounter or a similar daemon that is short-lived and may spawn several subprocesses.

Consider what happens when 50 processes all start to run at the same time and each process makes a short system call that consumes virtually no time, then goesa to sleep. The runqueue would be 50 for a short time (milliseconds) and none of the processes consume more than few CPU cycles, so %SYS and %USER is 0.0 but load is non-zero.

So, look at all your daemon processes (SAMBA or CIFS, automounter or autoFS, or similar. During a non-critical time, try stopping these processes and see if the load goes away.

There's nothing wrong with your system, it is just one of the many artifacts in trying to represent things that happen far too fast for humans to follow.

[Steven E. Protter]

look for a daemon process that should normally consume no cpu resources that does.

We found a aystem where lvmdevd has hours and hours of cpu time when it should not. This daemon makes sure the volume group can recover. There should be one per volume group.

If its consuming lots of cpu resources and there is no lvm activity or warnings in /var/adm/syslog/syslog.log then there may be a problem.

I don't at this time know what the problem is, but you may see something on your process list.

export UNIX95=1
ps -efH | more

This will show process hierachry, intenting the child processes.

It might help you find the problem process if not the solution.

Your load factors and expectations, as BH notes are not realistic. Before you can state an expected load factor, you need to know its calculated.

SEP