hight load with idle 100% on k360

[Pouponnot]

hello,

I have a little probleme on a HP-UX 11.11 model
9000/800/K360.
My load is hight (for a old machine),
uptime
2:36pm up 427 days, 43 mins, 2 users, load average: 2.02, 2.05, 1.89

but:

root@fn001u02# sar 2 5
HP-UX fn001u02 B.11.11 A 9000/800 02/10/05
14:37:41 %usr %sys %wio %idle
14:37:43 3 2 2 92
14:37:45 0 0 0 100
14:37:47 2 1 1 95
14:37:49 1 2 1 96
14:37:51 1 0 0 99
Average 1 1 1 96

swapinfo -tm
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 512 228 284 45% 0 - 1 /dev/vg00/lvol2
dev 512 194 318 38% 0 - 1 /dev/vg01/lvol1
reserve - 443 -443
memory 477 381 96 80%
total 1501 1246 255 83% - 0 -

I don't understand why my load is hight.
Could you help me ?
Best Regards
Julien
Paris La Defense

[Pete Randall]

Interesting! It looks like you may be swapping. Try running "vmstat 1 5" and look at the "po" column. Anything over single digits in "po" can indicate a paging/swapping problem. I suspect your machine is not accomplishing much because it's too busy doing context switches to actually do any real work.


Pete

[Bill Hassell · Accepted solution]

The load average is not a measure of how hard the computer is working. Load Average is the size of the RUNQUEUE, the queue of programs that are running or ready to run. If you have a single program that uses 100% of the processor (assuming a 1-porcessor system) than the RUNQUEUE will be 1.00 ( probably slightly more as other programs are run). It is not unusual to have a RUNQUEUE of 5 or 10 (I've seen 100+ before) due to certain programs that run for a very short time (like a polling program) and then sleep for a short time. Not very much CPU activity, but start 100 copies of this program and the RUNQUEUE will be high, but system and user CPU load will be low (high %idle).

On the other hand, a program that consumes 100% of the CPU and another that just reads the entire disk as fast as it can will consume all the data processing resource in your system and the RUNQUEUE will be about 2. This short script will consume 100% CPU:

while :
do
:
done

And this program will read the disk as fast as possible:

dd if=/dev/rdsk/c0t6d0 of=/dev/null bs=128k

Run both of those at the same time on a single processor system and you will see a slowdown.

So uptime is only a staring point for learning what is happening on your system.

[Fred Ruffet]

What is the 3rd line in a "top" output ? It shows how many process you have, and how many are in the states sleeping, running.

Are you running any kind of DB ? It may be (assuming this is Oracle) that you have a too high memory allocated to that DB compared to your RAM.

Regards,

Fred

[Pouponnot]

root# vmstat 1 5
procs memory page faults cpu
r b w avm free re at pi po fr de sr in sy cs us sy id
3 0 0 169189 4559 8 3 0 0 0 0 0 0 48 -2 59 11 1190621216768
3 0 0 169189 4529 4 0 1 0 0 0 0 601 896 384 0 0 100
3 0 0 169189 4522 2 0 1 0 0 0 0 569 663 304 0 0 100
3 0 0 169189 4522 1 0 1 0 0 0 0 553 541 257 0 0 100
3 0 0 169189 4522 0 0 1 0 0 0 0 542 448 219 0 0 100

System: fn001u02 Fri Feb 11 10:51:20 2005
Load averages: 1.77, 1.50, 1.53
175 processes: 166 sleeping, 9 running
Cpu states:
CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS
0 1.57 64.6% 0.0% 7.4% 28.0% 0.0% 0.0% 0.0% 0.0%
1 1.98 64.5% 0.0% 10.0% 25.6% 0.0% 0.0% 0.0% 0.0%
--- ---- ----- ----- ----- ----- ----- ----- ----- -----
avg 1.77 64.3% 0.0% 8.8% 26.9% 0.0% 0.0% 0.0% 0.0%

Memory: 308544K (252716K) real, 1029912K (760388K) virtual, 18232K free Page# 1/5

Thanks a lot for your respons, it's realy very usefull to me.
Do you know if the load is a lineare calculate, 1 of load for 100 precessus for exemple ? If yes, what 's the ratio ?

regards,

[Pouponnot]

juste another small thing:
your explanation for load, is it the same for other unix system (solaris, linux, aix...) ?

[Florian Heigl (new acc)]

Julien,

yes the load is always calculated that way.
i.e. a friend of mine is working on huge data warehouse systems on aix.
they have 16 cpu's which might be all at 100%, but the load can be above 200 at times.

This could be translated as 'currently the system has enough processes to keep 200 cpus busy' but the emphasis is on 'currently'.

We also had an hp-ux based bootp and http server which would be accessed by all clients at times, the load was well above 140, but there was no performance issue in that. The requests were all fullfilled from memory and after 5 minutes it was back to load 0 again.

Also, in Your case You seem to have processes that simply sit in the queue but are not currently using a lot of cpu time, so maybe simply call that thing 'process load'

[Wasitthee]

I also have problem on HP unix all the time

when I displayed

Load Avg is : 0.13, 0,13, 0,14

233 processes, 221 sleeping, 7 running, 5 zombies

The IDLE% is 100% what does it mean but HP machine was hang.

free 1/14 pages

[Dennis Handly]

>The IDLE% is 100% what does it mean but HP machine was hung.

If the machine was hung, how did you run top?

Certain types of hangs, don't use any CPU, so the system will be idle.

What was hung?  NFS?