Energy-Efficient Task Clustering Scheduling on Homogeneous Clusters

Clusters provide powerful computing performance is at cost of huge energy consumption. Scheduling a parallel application with a set of precedence-constrained tasks on cluster is challenging because of high communication cost. Although task duplication based scheduling algorithm is applied to minimize communication overhead, most of them only consider scheduling lengths, however completely ignoring energy consumption of cluster. Based on this consideration, we propose a novel Energy-Performance Balanced Task Duplication based Clustering Scheduling algorithm (EPBTDCS for short) in homogenous clusters which can significantly saving energy by judiciously shrinking communication energy consumption when assigning parallel tasks to computing nodes. This algorithm not only reduces energy dissipation in cluster without significantly degrading system performance, but also gets an optimal scheduling with a simple and loose condition. We conducted extensive experiments based on real-world SPEC fpppp and Sparse Matrix Solver parallel tasks applications running on a simulated cluster. By comparing with task duplication-based scheduling (TDS for short) and non-duplication-based scheduling (MCP for short) algorithms to prove our algorithm can save energy consumption greatly.