Abstract: Energy consumption is a critical issue in heterogeneous parallel and distributed systems. Dynamic voltage scaling(DVS) is a powerful technique to achieve energy saving by slowing down the processor into multiple frequency levels, and DVS algorithms typically consist of the assignment of tasks and the allocation of the slack. While most research focuses on the allocation phase, we considered the relation between the slack time and the assignmented of the tasks, and presented a Tabu-based Scheduling algorithm for energy minimization on heterogeneous distributed system.The algorithm finds an initial schedule for the tasks represented by directed acyclic graph(DAG) , and improves it by using a Tabu search strategy. The total energy consumption of the system is reduced further by using the Tabu strategy to forbid the task to assign onto specific processors and create more idle time slices for the slack allocation phase. Simulation results indicate that our algorithm achieves substantial energy savings on computer systems.

Key words: Scheduling algorithm, Dynamic voltage scaling, Heterogeneous system, Energy minimization, Tabu search