On the processor utilization bound of the C=D Scheduling Algorithm

Abstract

Under semi-partitioned multiprocessor scheduling some (or most) tasks are partitioned to the available processors while the rest may migrate between different processors, under a carefully managed scheme. One of the best performing and practical to implement EDF-based semi-partitioned algorithms is C=D splitting. Under this algorithm, each migrating task always executes at the highest-priority on all but one of the processors that it uses. This arrangement allows for efficient processor utilisation in general, however no tilisation bound had been published so far for this algorithm. We address this situation by deriving the utilisation bound of 13/18 for a variant of C=D with the following constraint: at most one migrating task may utilise each processor. We also draw additional conclusions for the utilisation bound attainable under a C=D task splitting scheme in the general case.