PRM and CPU capping

[Matthew White_3]

I think I've run into a wall with PRM and need to upgrade to WLM, but I'd like to run my problem up the flagpole and see if anyone has any wit or wisdom to help me with.

We have a four processor server with three different applications. We want to limit one of the applications to use a maximum of two CPUs for licensing purposes. We do not want to limit the other two applications - we want them to be able to use as much of the box as they want. As far as I can tell, there is no way to use PRM to absolutely cap application one at two CPUs (or 50% of available CPU power) without locking the other applications out of 50% of the box.

Question one - am I correct, or am I missing something?

Question two - if I am right, does WLM give me the kind of granularity I need to accomplish my stated goal.

Thank you for your help.

[RAC_1]

It has been a while since I worked on PRM. If my memory serves me right, you can configure it in two modes-share mode and pset mode.

In share mode, you can give % of CPU to application and in pset mode, you can dedicate cpus to the application. So if you configure it in pset mode, you can give 2 cpus to your application pending to other stuff. I think your problem will be resolved if you configure it in pset mode. I am not sure capping function is applicable to pset mode or not.

Anil

[Devesh Pant_1]

Matthew,
PRM can do the capping for you.
WLM is a layer above PRM and provides automatic resource allocation and application performance management through the use of prioritized service-level objectives (SLOs).

With PRM you can have PRM groups depending on what you want to do. I think Psets will best suit your needs

Please refer to the following for CPU capping
http://docs.hp.com/en/B8733-90011/ch05s02.html#daibbdhi

thanks
DP

[Matthew White_3]

The problem with using PSETs is that it puts one or more CPUs off-limits to processes in other groups. What I want is to be able to restrict group A to a maximum of two of the four CPUs while not restricting groups B and C in their CPU usage at all.

[melvyn burnard]

Question one - am I correct, or am I missing something?

You are correct. CPU Capping is across ALL PRM groups.

Question two - if I am right, does WLM give me the kind of granularity I need to accomplish my stated goal.

You should be able to configure WLM such that you assign whole cpu's to a group but others can use percentages of cpu.

WLM has two Non-Metric goal configurations:

Fixed Allocation:
Has explicit entitlement request
No goal is specified
CPU usage:
Usage goals specify a CPU utilization range
Has maximum and minimum CPU request bounds
WLM internally tracks the workload groupâ s actual CPU usage versus its CPU allocation
Useful when you want to automatically size the allocation based on what a workload needs


It would be worth a read of the manual available at http://docs.hp.com/en/netsys.html

[Jonathan Fears]

Hello Matthew. So first you can configure PRM to work for both PSET and FSS groups together. You could make the group for one application a PSET with 2 CPUs and have the other 2 be FSS groups that share the remaining 2 CPUs. Though this does not seen to be what you want as, like you stated, the 2 CPUs in the PSET are off-limits to the two FSS groups.

With WLM you can create workload groups and give them a absolute hard limit if desired. This gives you more flexibility than using PSETs in PRM because it does not bound 2 CPUs to that workload group and make them unavailable to others. Through WLM's use of SLOs (Service Level Objectives) and other keywords, specifically gmaxcpu, you will be able to create the configuration you are wanting. Take a look at the following site to learn more about WLM: http://www.hp.com/go/wlm. A link from this website also gives you the ability to directly contact us in the development lab with general product. questions.