AIX WLM in an uncapped micropartition - what's 50% mean?

While not totally new to WLM, we've not actually put it in active mode before.

I have an uncapped micropartition.
4 Virtual Processors,  .4 entitlement.

I have two JVMs running on the box
I've assigned each to a class (Web1 and Web2)
Web1 is capped at 50% cpu
I want to guarantee Web2 to have 60%, if it wants it.
If web1 isn't doing anything, let Web2 use it all.
If they're both busy, Web2 gets 60%, then has to fight for the remaining 40.

Question has come up, this is 60% of what?  .4 processors? or 4 full processors?
Another way to ask, WEB1 is capped at 50%, have I limited it to .2? or 2?

WLM is configured as follows: (contents tweaked for readability)
Classes -------
System: tier   = 1
Default: tier   = 1
        description = "Standard Web Services"
        tier   = 1
        inheritance = "yes"
        authuser = "webadmin"

        description = "Websphere2 Services"
        tier   = 1
        inheritance = "yes"
        authuser = "webadmin"

Shares: ---------------
Web1: CPU    = 10
Web2: CPU    = 60

Limits: ---------------
System:  memory = 1%-35%;100%
        CPU    = 0%-50%;50%
        memory = 0%-35%;35%
        CPU    = 60%-100%;100%
        memory = 60%-60%;60%

Who is Participating?
woolmilkporcConnect With a Mentor Commented:
If you're familiar with inter-partition scheduling you will probably know that your partition as a whole is limited to the use of 4 physical processors at max (due to 4 virtual processors) and to 0.4 processors as a minimum (due to entitlement).

Thus the actual CPU consumption of the whole partition will oszillate between these two values as a function of "Uncapped weight" and scale of contention between partitions.

You can now easily calculate that the 60% will be the share of this total CPU consumption.
Given there were no contention between logical partitions (only between your WPARs)
Web2 would be able to consume 2.4 physical CPUs, Web1 the remaining 1.6 CPUs.

As scale of contention between LPARS grows the ratio between Web1 and Web2 will remain the same, but the absolute value wil diminuish (down to the guaranteed 0.24 and 0.16 physical CPUs in the worst case).

So if the question were  "60 guaranteed percent of what?" the answer would be "60 percent of the guaranteed 0.4 processors".


From what I understand the Percentage shown is always the percentage of the entitled capacity.   In your instance it would be .4.  However if your parameters are setup with the Desired, Maximum and Minimum Processing Units and the LPAR is uncapped the utilization of your processors will fluctuate.  The server running that you are viewing will change slightly but sill only report the percentage of the processing units it has access to.  Which changes depending on your allocations of your partitioning.

The mpstat and lparstat tools with the HMC console you should be able to properly determine what is actually being utilized as far as physical processors.  The best way to control the growth and access to more processing power is to properly identify and weigh the applications in priority of of business needs.

Hope this helps, micro-partitioning, as soon as I think I get it I end up more confused than when I started lol
TomuniqueAuthor Commented:
What your describing, especially when it gets to the HMC,  sounds like inter-partition priorities, than intra.

Cloud Class® Course: C++ 11 Fundamentals

This course will introduce you to C++ 11 and teach you about syntax fundamentals.

TomuniqueAuthor Commented:
Ok, that makes sense as everything above entitlement is no longer "guaranteed".
But the hard cap on Web1 at 50%...
Assuming no inter-partition contention, and Web2 being idle, it could use 2 physical cpus?

The "Absolute Maximum" of Web1 being 50% - if there are enough free CPUs in your managed system and if Web2 is idle Web1 can use 2 physical CPUs - well, aside from the kernel/hypervisor overhead, of course!

TomuniqueAuthor Commented:
Thanks for the information/education.
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.