Conclusion

The impact of the relocation operation and CPU demand and capacity.

The left part of Figure 1 clearly reflects the CPU constrained situation for the Linux guest running the transactional workload, as visible through the comparatively high steal values. In this phase CPU demand is higher than the available processing capacity. This causes stolen” CPU capacity to appear to the guests, that is, the guests experience elapsed time intervals where their virtual CPUs appear stalled, because the real CPUs are needed for other guests or for work done by z/VM.

The center part of Figure 1 shows the impact of the relocation operation itself, while z/VM needs additional processing resources to perform the relocation process. At the end of the relocation phase the steal value would reach the maximum possible value of four CPUs for the duration of the quiesce phase. However, this is not visible from Figure 1 because during the quiesce time the guest is stopped, and thus Linux statistics data is not collected and thus not represented in Figure 1.

Once running on the otherwise unused target z/VM host, the relocated guest first needs a small amount of time to resume its workload processing, but then can make almost full use of its virtual processors, as reflected by the very low steal value.