Dynamic Load Balancing for Asymmetric Multi-core Architecture

Abstract

In Chaps. 3 and 4, we have introduced the techniques to improve the performance of memory-bound applications on multi-socket architecture. In this chapter, on the other hand, we introduce the scheduling techniques proposed to improve the performance of CPU-bound applications. On Symmetric Multi-Core (SMC) architecture in which all cores provide equal performance, traditional random work-stealing performs well. However, while single-ISA Asymmetric Multi-Core (AMC) architectures have shown high performance as well as power efficiency, current parallel programming environments do not perform well on AMC because they are designed for SMC architectures. The random task scheduling policies used in current parallel programming environments, such as work-sharing and work-stealing, can result in unbalanced workloads in AMC and severely degrade the performance of parallel applications. Essentially, it is a NP-hard problem to find the optimal task scheduling on an AMC architecture. In order to balance the workloads of parallel applications in AMC, in this chapter, we introduce an Asymmetric-Aware Task Scheduling (AATS) methodology.