Run queue query

[Srinikalyan]

Output of vmstat :
procs memory page faults cpu
r b w avm free re at pi po fr de sr in sy cs us sy id
9 0 0 3770998 382269 321 45 6 0 0 0 0 2415 1966951 1251 44 7 49
9 0 0 3770998 381883 2 0 0 0 0 0 0 2334 2785076 1633 60 7 32
9 0 0 3770998 381883 0 0 0 0 0 0 0 2315 2767665 1473 62 7 31
9 0 0 3770998 381883 0 0 0 0 0 0 0 2310 2783469 1385 60 7 32
9 0 0 3770998 381883 0 0 0 0 0 0 0 2314 2769268 1326 61 8 32

No of processors : 4
The run queue has more processes than the number of CPU's which indicates that the CPU is overloaded. But CPU idle% is above 30%. Does this indicate all the 9 processes in the run queue are able to run with only 70% CPU and none of them are waiting for CPU ..??
Is this correct?
Thanks,
Srini

[Dennis Handly]

>which indicates that the CPU is overloaded.

Not really. It could indicate there is a lot of waiting for I/O or swap. Or lots of process starting and ending.
What does "swapinfo -tam" show?

>... with only 70% CPU and none of them are waiting for CPU?

That seems reasonable, the problem is elsewhere.

[Srinikalyan]

Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 32704 2139 30565 7% 0 - 1 /dev/vg00/lvol2
reserve - 29358 -29358
memory 32701 8177 24524 25%
total 65405 39674 25731 61% - 0 -
I have been digging on the significance of run-queue in vmstat and understood that having
runq -size > no of processors , is not good..

But this server of ours has been running like this , without significant performance issues from users. So am trying to understand what is happening. The sar -u doesnot show any %wio .
I notice that the sar -q output shows the runq size as <2 , when vmstat shows it as 11 at the same time.

10:59:57 runq-sz %runocc swpq-sz %swpocc
10:59:59 1.8 50 0.0 0
11:00:01 1.8 100 0.0 0
11:00:03 2.0 50 0.0 0
11:00:05 1.6 82 0.0 0
11:00:07 1.5 80 0.0 0

Average 1.7 72 0.0 0

What is the difference between both?

Thanks,
Srini

[Dennis Handly]

dev 32704 2139 30565 7%
reserve - 29358 -29358
memory 32701 8177 24524 25%
total 65405 39674 25731 61% - 0 -

It doesn't look like you are swapping.

>understood that having runq -size > no of processors, is not good.

Not always.

>without significant performance issues from users. So am trying to understand what is happening.

You should not just be looking at the run queue. You should be looking at CPU percentage and other performance criteria.

>The sar -u does not show any %wio.

So you don't have an I/O bottleneck.