Using WLM to enable TiCAP CPUs within a npar

[kavanagh_1]

I am preparing to setup WLM to automatically enable TiCAP CPU cores within a npar when the active CPU cores are maxed out,ie the SLO has failed.

In the WLM User's guide (A.03.02) it does give some example configuration files (par_usage_goal.wlm with par_usage_goal.wlmpar) that appear to achieve what I want but seem to refer to moving TiCAP cores between npars not within a npar.

Will these example files do the job?

David

[melvyn burnard]

should do the trick, but you may wish to take a look here:
http://h20338.www2.hp.com/hpux11i/cache/328169-0-0-0-121.html

[kavanagh_1]

Thanks for your reply Melvyn,

I had already viewed the WLM web site conf examples. The par_usage_goal.wlm and par_usage_goal.wlmpar are described there as well as in the WLM manual.

I cannot see any example which specifically enables TiCAP CPU cores within a npar.

You say it should do the trick. Guess I will just have to try it but I would like to be sure first.

Can anyone confirm ?

[Jonathan Fears]

Hello David. The way WLM 3.2 works is that is always tries to migrate cores between the nPars before activing temporary capacity. This was done with saving money in mind. If one nPar is busy, but another is not, then the cores from the nonbusy nPar can be used in the busy nPar without any additional cost being incurred by the customer.

However, there are ways to get around this using the WLM configuration files. If you gave each nPar a guaranteed minimum number of cores that totals the number of active cores on the complex (i.e. 16 total cores on complex, each nPar has hmincpu of 8 cores), WLM has no choice but to activate TiCAP for that nPar as no cores can be migrated. Also using the utilitypri keyword allows you to give failing SLOs the ability active TiCAP and which can not. This can be used in such a way as to allow one nPar to active TiCAP and the other not to active it.

Hope this helps.

[kavanagh_1]

Hi Johnathan,

Many thanks for your reply. Please could you clarify your first statement for me.

You state "The way WLM 3.2 works is that is always tries to migrate cores between the nPars before activing temporary capacity."

If, for example, I had 2 npars (0 and 1) each with 5 active and 1 inactive iCAP cores and then npar1 required more CPU resource, could WLM deactivate one standard core in npar 0 (assuming it was only needing the resource of 4 cores)and then activate one of the Icap cores in npar 1?

Regards

David

[Jonathan Fears · Accepted solution]

Hello again. Yes, the example you've given is exactly correct. nPar1 would get CPU cores from nPar0 if available. WLM would deactivate a core in nPar0 and active a core in nPar1. This would keep the number of active cores for the complex at 10 and bypass the need to use TiCAP to satisfy nPar1's SLOs.

However, if nPar0 was using all 5 of its active CPU cores, then the iCAP core in nPar1 could be activated with TiCAP to give it 6 active CPU cores. Now instead of the standard 10 active CPU cores, you now have 11 active cores (one of which is depleting your available TiCAP).

Like I said previously, WLM will always try to move CPUs between partitions before activating cores with TiCAP. If you have any other questions or need more clarification, please let me know.

Regards,
Jonathan

[kavanagh_1]

Hi Johnathan,

Thanks again for your reply. This is becoming much clearer. So I presume gWLM will take this a step further and manage the CPU cores over a number of server complexes. Is that correct?

Regards

David

[Jonathan Fears]

Hello David. gWLM still currently manages within a single complex (physical electrical box) as well. At this time there is no mechanism to migrate CPU cores between multiple complexes using iCAP (as is done now between partitions in the same complex). However, the advantage comes when you have a large number of partitions that need to be managed. With WLM, each partition needs its own configuration, etc. With gWLM, everything is centralized and done from the CMS which greatly simplifies configuration and management. Also if the Virtual Server Environment (VSE) is purchased, other products such as Capacity Advisor (helps with capacity planning) are available as well along with gWLM.

Regards,
Jonathan