Energy-aware algorithms for task graph scheduling, replica placement and checkpoint strategies

Abstract

The energy consumption of computational platforms has recently become a critical problem, both for economic and environmental reasons. To reduce energy consumption, processors can run at different speeds. Faster speeds allow for a faster execution, but they also lead to a much higher (superlinear) power consumption. Energy-aware scheduling aims at minimizing the energy consumed during the execution of the target application, both for computations and for communications. The price to pay for a lower energy consumption usually is a much larger execution time, so the energy-aware approach makes better sense when coupled with some prescribed performance bound. In other words, we have a bi-criteria optimization problem, with one objective being energy minimization, and the other being performance-related.