Offline replication and online energy management for hard real-time multicore systems

Abstract

For real-time embedded systems, energy consumption and reliability are two major design concerns. We consider the problem of minimizing the energy consumption of a set of periodic real-time applications when running on a multi-core system while satisfying given reliability targets. Multi-core platforms provide a good capability for task replication in order to achieve given reliability targets. However, careless task replication may lead to significant energy overhead. Therefore, to provide a given reliability level with a reduced energy overhead, the level of replication and also the voltage and frequency assigned to each task should be determined cautiously. The goal of this paper is to find the level of replication, voltage and frequency assignment, and core allocation for each task at design time, in order to achieve a given reliability level while minimizing the energy consumption. Also, at run-time, we find the tasks that have finished correctly and cancel the execution of their replicas to achieve even more energy saving. We evaluated the effectiveness of our scheme through extensive simulations. The results show that our scheme provides up to 50% (in average by 47%) energy saving while satisfying a broad range of reliability targets.