Energy-Efficient Dynamic Scheduling on Parallel Machines

Energy consumption is a critical issue in parallel and distributed systems. Workflows consist of a number of tasks that need to be executed to complete an application. These tasks typically have precedence relationships that have to be observed during execution for correctness. DAGs (Directed Acyclic Graphs) can be used to represent many such workflows. The static algorithms to schedule for energy minimization under the deadline constraints are based on estimating worst case execution time for each task to guarantee that the application completes by a given deadline. During execution, many tasks may complete earlier than expected during the actual execution. This allows for adjusting the schedule for the tasks that have not yet begun execution to incorporate the extra stack. This has to be done with the dual goal of reducing the energy requirements while still meeting the deadline constraints. In this paper, we present a novel dynamic algorithm for remapping tasks for energy efficient scheduling of DAG based applications for DVS enabled systems. Our experimental results show that the combination of our dynamic assignment and dynamic slack allocation leads to significantly better energy minimization compared to not changing the static schedule and/or only performing dynamic slack allocation. Furthermore, its execution time requirements are small enough to be useful for a large number of applications.