On the Energy Optimization for Precedence Constrained Applications using Local Search Algorithms

We investigate the problem of scheduling precedence constrained applications on a distributed heterogeneous computing system with the aim of minimizing schedule length and reducing energy consumption. We present a scheduling algorithm based on the best-effort idea that promotes local search algorithms and dynamic voltage scaling to reduce energy consumption. The final goal is to maintain a given performance while minimizing energy use. The proposed approach first uses a list-based scheduling algorithm to find near-optimal solutions for schedule length, then local search algorithms with dynamic voltage scaling are applied to reduce energy consumption. However the algorithm it's not allowed to deteriorate the schedule length computed by the best-effort algorithm. We discuss simulation results obtained with sets of real-world applications that emphasize the interest of the approach.