A Multi-objective GRASP Algorithm for Joint Optimization of Energy Consumption and Schedule Length of Precedence-Constrained Applications
English
Pecero, Johnatan[University of Luxembourg > Faculty of Science, Technology and Communication (FSTC) > Computer Science and Communications Research Unit (CSC) >]
Bouvry, Pascal[University of Luxembourg > Faculty of Science, Technology and Communication (FSTC) > Computer Science and Communications Research Unit (CSC) >]
2011 IEEE Ninth International Conference on Dependable, Autonomic and Secure Computing (DASC)
IEEE
510-517
Yes
No
International
978-0-7695-4612-4
IEEE Ninth International Conference on Dependable, Autonomic and Secure Computing - DASC
December, 2011
Sydney
Australia
[en] Green Computing ; Distributed Computing ; Scheduling ; Energy Optimization ; Green-IT
[en] We address the problem of scheduling precedence-constrained scientific applications on a heterogeneous distributed processor system with the twin objectives of minimizing simultaneously energy consumption and schedule length. Previous research efforts on scheduling have focused on the minimization of a quality of service metric based on the completion time of applications (e.g., the schedule length). Recently, many researchers are working on the design of new scheduling algorithms that consider the minimization of energy consumption. We report a new scheduling algorithm accounting for both objectives. The new scheduling algorithm is based on a multi-start randomized adaptive search technique (GRASP framework) that adopts Dynamic Voltage Scaling technique to minimize energy consumption. This technique enables processors to operate in different voltage supply levels at the cost of sacrificing clock frequencies. This multiple voltage implies a trade-off between the quality of the schedules and energy consumption. Therefore, the new proposed approach is designed as a multi-objective algorithm that simultaneously optimize both objectives. Simulation results on a set of real-world applications emphasize the robust performance of the proposed approach.