A multiobjective genetic algorithm to improve power and performance of heterogeneous multiprocessors
Jagpreet Singh, Mohit Kumar Pandey, Ekansh Katiyar, Rohit Tulasyan, Vijay Gupta, Nitin Auluck · 2016
The concern of reducing power consumption and the schedule length (or makespan) has added new levels of complexities to the scheduling of task graphs on multiprocessors. Recent researches in the field have obtained considerable improvement in power consumption with a comparable makespan by exploring a large search space with mixed integer linear programming (MIP). However, MIP based solutions are applicable only to smaller instances of task graphs. In this paper, we propose a multi-objective genetic algorithm (P2Gen) that uses dynamic voltage/frequency scaling (DVFS) to perform Power (P) and Performance (P) aware scheduling. With DVFS, tasks are made to run on low voltages, which decreases their computation power. However, it also increases their execution costs and hence, is optimized with a genetic algorithm. In the first phase, P2Gen optimizes the makespan whereas the second phase uses DVFS with multiple tolerance levels of the obtained makespan to optimize both the objectives. To enhance the convergence of P Gen as well as exploring a variety of population, each task is run by integrating the maximum and minimum voltage on a processor instead of iterating through all the voltage levels. The results demonstrate that P2Gen is able to balance power and performance better than other algorithms.