Modifying EDF for Uniform Multiprocessors

Scheduling real-time systems on multiprocessors introduces complexities that do not arise when using uniprocessors. If the processors do not all operate at the same speed, scheduling becomes even more complex. When scheduling using EDF on multiprocessors, breaking deadline ties in different ways can change the resulting schedule dramatically. We consider methods to resolve the ambiguities in EDF priorities due to coincident deadlines. We show that no optimal ambiguity resolver can be both on-line and priority-driven. When processor speeds differ, a job’s minimum required execution rate may also be considered in the scheduling decisions. We propose a modification of existing scheduling algorithms that considers processor speed and minimum required execution rate. We show that the resulting algorithm dominates EDF on identical multiprocessors and conjecture that it dominates EDF on uniform multiprocessors as well.