Optimizing application mapping algorithms for NoCs through a unified framework

This paper presents a preliminary PhD research towards developing a framework to evaluate and optimize application mapping algorithms for Network-on-Chip architectures. Several such algorithms have been proposed for mapping the threads of a parallel application on a NoC architecture. However, the performance of those algorithms is evaluated only on some specific NoC designs. A unified approach for evaluating such algorithms allows a better comparison of their performance and can potentially lead to some optimizations. The proposed framework is intended to be flexible so that the algorithms can be tested on different NoC designs. To this end, a scalable and flexible Network-on-Chip simulator is proposed. Some preliminary results obtained with this simulator are presented, too. They show the flexibility of this simulator and that it is feasible for addressing the application mapping problem in a unified manner.