Slack allocation algorithm for energy minimization in cluster systems

Energy consumption has been a critical issue in high-performance computing systems, such as clusters and data centers. An existing technique to reduce energy consumption of applications is dynamic voltage/frequency scaling (DVFS). In this paper, we present a novel algorithm called EASLA for energy aware scheduling of precedence-constrained applications in the context of Service Level Agreement (SLA) on DVFS-enabled cluster systems. Due to the dependencies among tasks and makespan extension, there may be some underused slacks. The main idea of the EASLA algorithm is to distribute each slack to a set of tasks and scale frequencies down to try to minimize energy consumption. Specifically, it first finds the maximum set of independent tasks for each task, and then iteratively allocates each slack to the maximum independent set whose total energy reduction is the maximal. Randomly generated graphs and two real-world applications are tested in our experiments. The experimental results show that our scheduling algorithm can save up to 22.68% and 12.01% energy consumption compared with the GreedyDVS and EvenlyDVS algorithms respectively in homogeneous environments, and 12.33% energy consumption compared with the EES algorithm in heterogeneous environments.