PVU in VMware clusters: vMotion and the capacity question.

VMware is where PVU sub-capacity gets misunderstood most often, because virtualization makes capacity a moving target. A virtual machine running WebSphere might use four cores today, but if vMotion can relocate it across a cluster of large hosts, the relevant question for IBM is the maximum capacity the software has access to, not the slice it happens to occupy at one moment. Misjudging that boundary is how a modest deployment turns into a cluster sized finding.

What vMotion does to the count

vMotion and DRS allow a virtual machine to migrate between physical hosts. From a licensing standpoint, mobility means reachable capacity. If the IBM workload can move anywhere in a cluster, the eligible PVU can be assessed across the hosts it can reach rather than the single host it sits on. A four core virtual machine that is free to roam a cluster of several thirty two core hosts is not a four core licensing question, and treating it as one is the error that audits exploit.

How sub-capacity contains it

The control is ILMT combined with deliberate boundaries. ILMT measures the peak virtual capacity the IBM software actually reaches over time, and a correctly configured sub-capacity position lets you license that measured peak rather than the full physical cluster. But this only holds if the tool is deployed and reporting continuously across the cluster. Without that measurement, IBM is entitled to fall back to full capacity, and in a cluster full capacity means every core in every host.

• Reachable hosts, not occupied hosts, define the capacity at risk
• ILMT measures the peak virtual capacity the software actually uses
• Pinning or host affinity can constrain where the workload may run
• An uncontained cluster exposes every core to full-capacity charging

Pinning and host affinity as a licensing tool

One legitimate way to reduce reachable capacity is to constrain it. Host affinity rules or pinning can restrict an IBM workload to a defined subset of hosts, which shrinks the capacity IBM can count, provided the restriction is real and enforced rather than aspirational. The configuration has to match the claim. An affinity rule that exists on paper but is routinely overridden will not survive scrutiny.

Where audits find the money

The expensive VMware findings come from two situations: a cluster where ILMT coverage was incomplete so the peak could not be evidenced, and a cluster where workloads were free to roam with no boundary at all. In both cases IBM moves to full capacity across the cluster. The defensible position pairs complete ILMT measurement with documented affinity boundaries, so the capacity you license is the capacity you can prove.