Power-aware load balancing in heterogeneous clusters

Minimizing energy consumption of computing systems is a complex problem with many parameters. Clusters have been studied from a performance perspective and more recently from a power consumption angle. Since clusters are usually composed of different hardware with different performance and energy consumption, we propose load balancing policies that take into account servers' computational power and power consumption. These policies are applied on a heterogeneous power-aware DVFS cluster running hard real-time CPU intensive tasks. Proposed policies are compared to the Earliest Deadline First policy. Simulation results show that these policies achieve higher energy savings and less task transfers at the same time.