A dominant predecessor duplication scheduling algorithm for heterogeneous systems

  • Authors:
  • Kuan-Chou Lai;Chao-Tung Yang

  • Affiliations:
  • Department of Computer and Information Science, National Taichung University, Taichung, Taiwan 40302;Department of Computer Science and Information Engineering, Tunghai University, Taichung, Taiwan 40704

  • Venue:
  • The Journal of Supercomputing
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposes a new duplication-based task scheduling algorithm for distributed heterogeneous computing (DHC) systems. For such systems, many researchers have focused on solving the NP-complete problem of scheduling directed acyclic task graphs to minimize the makespan. However, the heterogeneity of computational resources and communication mechanisms poses some major obstacles to achieving high parallel efficiency. This paper proposes a heuristic strategy called the Dominant Predecessor Duplication (DPD) scheduling algorithm, which allows for system heterogeneities and communication bandwidth to exploit the potential of parallel processing. This algorithm can improve system utilization and avoid redundant resource consumption, resulting in better schedules. Experimental results show that the system heterogeneities and program structures of applications affect scheduling performance, and that our presented algorithm is better able to avoid these problems than those presented in previous literature. Here, we show that our algorithm can be applied to design efficient distributed systems to overcome performance bottlenecks caused by system heterogeneities.