Energy-Efficient System Design of Asymmetric Multiprocessor for Real-Time Streaming Applications

Abstract

In this paper, we propose a new methodology to design asymmetric multiprocessor (AMP) systems for real-time streaming applications under throughput and latency constraints, where streaming applications are specified by CSDF graph in order to explore an ample amount of parallelism inside. We first propose an approach to assign a suitable type of processor for each actor in CSDF-model streaming applications and derive a condition to quickly decide the energy-efficiency of the AMP designed according to the processor assignment. Then, based on those, we propose a polynomial algorithm to design AMPs for real-time streaming applications. We compare the designed AMPs with SMPs on the basis of energy dissipation and synthesized area. The results show that on average AMPs can gain 13% and 28% saving for energy dissipation and synthesized area, respectively. Furthermore, the proposed methodology provides as effective as the design space exploration based on genetic algorithm in terms of finding energy-efficient AMPs, but our proposed methodology can achieve several order of magnitude of speed-up in execution time.