HPE Community read-only access December 15, 2018
This is a maintenance upgrade. You will be able to read articles and posts, but not post or reply.
Hours:
Dec 15, 4:00 am to 10:00 am UTC
Dec 14, 10:00 pm CST to Dec 15, 4:00 am CST
Dec 14, 8:00 pm PST to Dec 15, 2:00 am PST
System Administration
cancel
Showing results for 
Search instead for 
Did you mean: 

Unbalanced CPU Utilization in rp7420

 
Mirror breaking
Contributor

Unbalanced CPU Utilization in rp7420

Hi Friends,
We have a rp7420 with single cell board and 8 CPUs.
I run sar for 12 hour and got the data.
While i checked the data I found that load on CPU 4,5,6,7 is much higher then 0,1,2,3.
Please any one can suggest why this is happening and what is the solution of the case.
For oyur referance I am attaching sar report with it.
Regards
Ankit Agrawal
8 REPLIES
Mark McDonald_2
Trusted Contributor

Re: Unbalanced CPU Utilization in rp7420

Hi I have not looked at the attachment yet.

What applications/DB's are you running? This may be quite normal and nothing to worry about.
Mirror breaking
Contributor

Re: Unbalanced CPU Utilization in rp7420

Hi,
Thanks for your reply.
We are running SAP APO on the server with oracle.
Regards
Ankit Agrawal

Re: Unbalanced CPU Utilization in rp7420

why are you concerned?

None of this is showing any CPU bottlenecks. Maybe you only have around 4 processes that ever need much CPU time.

Have a look in "sar -qM 2 20" are you seeing any high numbers in the cpu runq-sz?

HTH

Duncan

HTH

Duncan
Mirror breaking
Contributor

Re: Unbalanced CPU Utilization in rp7420

I know there is no CPU bottleneck.
But is it logical that 4 r getting highly used and 4 are not.
And my application team wants that every cpu will work evenly.
Not like present scenerio.

Re: Unbalanced CPU Utilization in rp7420

This reminds me of a very highly paid consultant from one of the big SIs telling me "that L3000 server is unbalanced cos it has 3 CPUs - it should have 2 or 4"! I asked him if he was concerned that the server might topple over cos of the uneven distribution of weight!

Seriously though.. if there's only a small number of processes doing work, then only a small number of CPUs are going to get used - when you look in glane or top do you see the same small set of processes running all the time?

Anyway you can tell your apps team that there's nothing you can do to "balance" this out (not that it matters if there's no performance hit). If they want to see more "balance" they need to look at their application parameters and see if there are ways to increase the degree of parallelism in the app (maybe starting more server instances?) I'm afraid I'm not familiar with LiveCache so can't really comment on how you'd do this.

HTH

Duncan

HTH

Duncan
Fernando_4
Occasional Visitor

Re: Unbalanced CPU Utilization in rp7420

What is your Oracle's version?
fcmts
Avinash20
Honored Contributor

Re: Unbalanced CPU Utilization in rp7420

Hi Ankit,

You need to check from the Application end
They need to design the Application to evenly distribute on the CPU available

OS will give the resource as the application demands.

I believe you should consult your application vendor, if they like to balance it on all CPU.
"Light travels faster than sound. That's why some people appear bright until you hear them speak."
Hein van den Heuvel
Honored Contributor

Re: Unbalanced CPU Utilization in rp7420

>> And my application team wants that every cpu will work evenly.
>> Not like present scenerio.

And WHY would they want that?
To keep you busy? because it looks cute?

If there is a need to engage the other processors, then the OS will. Trust it for that!

Typically you want a process and its children to stay on the same CPU to increase CPU cache characteristics.

'Moving' a process has a cost, and is only done when there is a clear need.

For SAP applications, work tasks are accepted by the dispatchers and assigned to a first free work process... NOT round-robinned. If a small number of work processes can handle the work, then those can stay on the CPUs originally assigned to them and the others will not get active much. If they happened to be assigned in high-number to low-number order, then this would explain what you observe and would indicate that this is optimal.

To know exactly what is happening, one would need to know the exact OS version, teh scheduler policy selected, any pset, or mpctl setting opted for and so on.

You may want to review the Robert Sauers "HPUX tuning and performance" book before digging into this further.

hth,
Hein.