The total number and multiprocessor level of the VMs affect the VM2000 overhead:
Total number of VMs and total number of virtual CPUs
The more virtual CPUs are started on a real CPU, the less ideal the effect of the CPU caches is.
By means of CPU affinity (see "VM2000 scheduling (/390 servers)") VM2000 (/390 servers) reduces the number of CPU switchovers in which a virtual CPU is executed on a different real CPU than before.
CPU pools enable the number of active virtual CPUs per real CPU to be restricted.The VM2000 hypervisor overhead for managing the CPU queues on /390 servers depends directly on the number of virtual CPUs in a CPU pool.
The relation of the number of active virtual CPUs (of the guest systems with production load) to the number of real CPUs should be less than or equal to 3 in all CPU pools. The number of guest systems should be correspondingly low. When the performance requirements for the individual guest systems are lower, a larger number of VMs (up to 15) can also be configured. In all cases it must be checked whether the available main memory is sufficient for the guest systems required.Multiprocessor level of a VM (= maximum utilization of the VM)
The load of the virtual CPUs of a VM has an effect on the utilization level of the time slice and consequently on the frequency of the scheduling operations. A high utilization level of the time slice and a low frequency of scheduling operations are favorable for performance.
The multiprocessor level of a VM should be oriented according to the peak requirement of the guest system. As low a value as possible should be selected. If the load is low, virtual CPUs in the guest system should be detached.