Energy-aware scheduling for workflow applications on distributed systems using an evolutionary approach and DVFS-techniques
Co-authorBernabe Dorronsoro and Pascal Bouvry
Task scheduling problem for workflow applications has been one of the fundamental issues in parallel and distributed computing. It has been extensively studied in recent years addressed to different efforts, traditionally, to minimize the application completion time and to treat time complexity. Nowadays, there is an important interest to reduce energy consumption in distributed computing systems due to performance, environment and operating cost issues. In this talk we address the problem of scheduling workflow applications on distributed computing systems with the objectives of minimizing both the finishing time and the energy consumption. We provide a scheduler based on an evolutionary algorithm which adopts dynamic voltage and frequency scaling (DVFS) to minimize energy consumption. DVFS has been increasingly integrated into many recent commodity processors as an efficient power management technology. DVFS enables these processors to operate with different voltage supply levels at the expense of sacrificing clock frequencies. In this context, this multiple voltage facility implies a trade-off between the quality of schedules and energy consumption. Therefore, we provide a multi-objective evolutionary approach based on a new cellular Genetic Algorithm (cGA). The main characteristic of the cGA algorithm is that the population is structured in a two dimensional toroidal mesh to achieve a good exploration/exploitation trade-off on the search space. This improves the capacity of the algorithm to solve complex problems as the energy-aware scheduling problem. The extensive simulations conducted as a part of this work verified that the performance of our algorithm is very compelling in terms of both application completion time and energy consumption.