Contention-aware reliability management scheme for parallel tasks scheduling in heterogeneous computing systems

摘要

Energy efficiency and high system reliability are two primary measurement in modern high-performance computing. Most recent studies pay too much attention on low energy consumption or execution time for parallel tasks scheduling. In addition, these approaches are proposed for the classic scheduling model. It is increasing recognized that contention model is more realistic and be of benefit to create accurate and efficient schedules. This paper presents a contention-aware reliability management algorithm for parallel tasks scheduling in heterogeneous computing systems. Extensive experiments are performed to evaluate the results. It is demonstrated that our algorithm significant improve the system reliability.