HiperDispatch and Shared Logical Partitions

The z17 provides a higher level of synergy between the PR/SM Hypervisor and z/OS® software for managing logical core resource allocations, called HiperDispatch.

HiperDispatch is the true start of having z/OS software become aware of the topology of the machine, presented as a logical partition topology, to then provide dynamic affinity dispatching with regards to CP placement in the structure.

PR/SM has traditionally managed shared logical partitions as being horizontally polarized. That is, the processing weight for the logical partition is equally divided between all the online logical cores in the logical partition. In turn, the OS running in the logical partition is obligated to use all of those online logical cores equally in order to be assured of receiving its fair share of the physical core resources. HiperDispatch introduces a new, optional form of polarization for managing the shared logical cores of a logical partition called vertical polarization. z/OS running with HiperDispatch uses the logical partition topology information to decide how to group its logical cores to set up its work queues and to exploit the vertical configuration of logical cores to pack work into a smaller set of logical cores, optimizing processor cache usage.

With HiperDispatch there is a common understanding in the PR/SM Hypervisor and the z/OS software such that work can be concentrated on a smaller number of logical cores within the logical partition that reflects the actual assigned weight of the logical partition. With HiperDispatch enabled in a logical partition, z/OS will redispatch its tasks back to a smaller set of logical cores, and PR/SM in turn can dispatch those logical cores back to the same physical cores, optimizing use of the L1, L2, L3, and L4 caches. Work can still expand and flow into more logical processors dynamically should workload conditions warrant it.