Energy, Vol.125, 258-274, 2017
Energy-aware stochastic scheduler for batch of precedence-constrained jobs on heterogeneous computing system
The problem of optimal scheduling of precedence-constrained jobs as well as finding the Pareto-optimal sets for multi objective scheduling problem have been proven to be nondeterministic polynomial time(NP)-complete. The growing consumption of energy has compelled the researchers to consider energy consumption as an important parameter along with other parameters in multi-objective scheduling problem. Accordingly, many energy-aware precedence-constraints scheduling algorithms have been reported in the literature. Most of the algorithms have a limitation of treating this problem as a single objective optimization problem modelling with deterministic execution times rather than stochastic execution times. This work proposes energy-aware stochastic scheduler to schedule the batch of precedence-constrained jobs on dynamic voltage frequency scaling-enabled processors in order to optimize the energy consumption and the turnaround time. The execution and inter-communication times are stochastic which are drawn from independent probability distributions. A novel encoding for batch of precedence-constrained jobs, stochastic turnaround time and energy models are also proposed. Experimental results show that, compared with other algorithms, the proposed scheduler offers reduced turnaround time and reduced energy consumption. (C) 2017 Elsevier Ltd. All rights reserved.
Keywords:Heterogeneous computing system;Stochastic scheduling;Dynamic Voltage and Frequency Scaling;Precedence-constrained jobs;Multi-objective optimization;Energy;Turnaround time