IC5Notice: We have upgraded developerWorks Community to the latest version of IBM Connections. For more information, read our upgrade FAQ.
  • 1 reply
  • Latest Post - ‏2012-05-30T10:52:11Z by asmcleod
1 Post

Pinned topic ITUAM reporting on virtual and physical cores used - help needed

‏2012-05-26T18:01:04Z |
I appreciate your input on the following:
I am new to ITUAM but need to know if this can be done.

I see standard ITUAM reports with "cpu hours". However, what I would like to see
is simply number of virtual cores assigned to active VMs per day (over a
months interval for example so that I can determine a monthly average, and per tenant). Is
that possible? If so, which product skills are needed for this?

If this is possible then my next question would be: if I have a report (in
one resource group) with a number of virtual cores, and I divide this by
the overcommitment factor, then I have the number of physical cores that were needed
to run the number of virtual cores.
Is this reasoning correct? Would you automate this in ITUAM also - if so which product skills
are needed for this?

Reason for this question: I am working with a customer
where I have to translate back to physical cores because of the agreed
price model.
I appreciate your help - I hope you can comment on this.

Thanks, Menno.
Updated on 2012-05-30T10:52:11Z at 2012-05-30T10:52:11Z by asmcleod
  • asmcleod
    97 Posts

    Re: ITUAM reporting on virtual and physical cores used - help needed


    Have you looked at the VMNUMCPU (Number of Processors in the Virtual Machine) resource provided by TUAM's VMware data collector? Sounds like it provides the number of underlying processors/cores, instead of the amount of CPU time consumed.

    You could load that metric with a daily collection job (use a rate value of zero), then at end of month, run a custom job which extracts and averages VMNUMCPU for the month.

    Regarding overcommitment: though I'm not a VMware expert, I presume your logic would work for a given interval.

    Note that in both cases, averaging over a monthly interval sounds like any changes during the month might get diluted by the volume of data; that is, if one week had a higher overcommitment factor than the other three for a month, using monthly totals probably averages out that variability.

    Do you really need a physical core pricing model? Many clients use VMware and other virtualization technologies to isolate their customers from the details and minutia of the underlying hardware; they're happy to derive a CPU time rate, whereby TUAM's normal aggregation methods reflect each VMware guest's proportional use of the VMware host machine. This simplifies the billing model for the customer.

    Or, if you're using a service request tool such as IBM's Tivoli Service Automation Manager (TivSAM), those tools can report virtual service deployments in terms of instances, number of CPUs requested, etc. - that still allows the service provider to manage the underlying HW resources yet bill the customers on a service basis, which sounds close to what you're asking?