Bag-of-Tasks Load Balancing on Power-Aware Clusters

Abstract

Energy preservation in computing systems is very important nowadays. A large percentage of the workload submitted to large-scale systems is bag-of-tasks (BoT) applications. Each BoT is a collection of independent tasks that do not communicate with each other. This type of tasks is used in astronomy, data mining and image processing. BoT load balancing has been studied in the literature from a performance perspective. In this paper, we study BoT load balancing from an energy efficiency perspective. We propose two load balancing mechanisms that are evaluated through simulation in a heterogeneous cluster environment. BoTs and high-priority real-time tasks are both submitted to the cluster, in order to represent more accurately the workload submitted in large-scale systems. Extensive simulation experiments show that by applying the proposed load balancing mechanisms, significant energy savings can be achieved with small performance degradation.