On the Impact of Runtime Overhead on Energy-Aware Scheduling

The real-time research community is often con- cerned with finding suitable assumptions to simplify the sch edu- lability analysis of current real-time systems. This inclu des sim- plified power models, negligible scheduling overhead, negl igible preemption cost, bounded cache misses and bus contention, etc. However, the actual behavior of a real application might be significantly different than that expected from a simplified system under the adopted set of assumptions. This paper investigates the impact that preemption cost might have on the energy consumption under a given energy-aware scheduling algorithm. A set of simulation experiments illustrate the influence of context switch and scheduling overhead on th e actual energy consumed in a given system. Results show that, in certain conditions, the penalty due to the runtime overhead might be as large as the amount of energy saved using aggressive DPM and DVFS scheduling techniques. In this context, limited preemptive scheduling is proposed as a possible solution for limiting the main sources of overhead to fully exploit the benefits of power saving features of current computer architectures.