Contention-Aware Scheduling with Task Duplication

Abstract

Scheduling a task graph onto several processors is a trade-off between maximising concurrency and minimising interprocessor communication. A technique to reduce or avoid interprocessor communication is task duplication. Certain tasks are duplicated on several processors to produce the data locally and avoid the communication among processors. Most algorithms using task duplication are for the classic model, which allows concurrent communication and ignores contention for communication resources. The recently proposed, more realistic contention model introduces contention awareness into task scheduling by assigning the edges of the task graph to the links of the communication network. It is intuitive that scheduling under such a model benefits even more from task duplication. This paper proposes a contention-aware task duplication scheduling algorithm, after investigating how to use task duplication in the contention model. An extensive experimental evaluation demonstrates the significant improvements to the speedup of the produced schedules.