科研成果详情

We propose and analyse an efficient energy-aware application mapping algorithm for dynamic multicore systems. Computer systems integrate more and more processor cores in a single chip due to the limitation of frequency. Not only desktop and server processors, but also mobile processors are equipped with eight or even ten cores. Modern applications have increased focus on parallel computation due to the multicore trend. Mapping applications to the processor without optimization can lead to performance loss and system efficiency degradation. The energy efficiency of multicore systems is especially important for battery-driven devices. We investigate the energy consumption of data transmission in multicore chips. The consumption is divided into router energy and link energy. We discuss the total energy required for transferring data across nodes in a multicore processor. Based on the energy model and application model, we propose an energy-aware mapping algorithm which maps applications in two different ways. We evaluate the proposed algorithm with other algorithms by using synthetic and real applications. The synthetic workloads proved that the proposed algorithm can be executed dynamically with a negligible time overhead, while the energy consumption reduced significantly. Results from four applications show that the power consumption of the proposed algorithm has reduced by 20.2% and 13.4% compared with the first fit and incremental algorithms respectively.