On Performance Improvement of Concurrent Applications Using Simultaneous Multithreaded Processors as NoC Resources

Network on chip (NoC) is a new design paradigm for building scalable multi-core/multiprocessors systems on chip (SoC) which are able to exploit the coarse grain parallelism of concurrent applications. Simultaneous multi-threaded processor (SMTP) is a superscalar processor architecture which adoptively exploits the coarse grain and the fine grain parallelism of applications, by simultaneously executing instructions from several thread contexts. In this paper, the authors make a case for using SMTPs as NoC resources and show that such a multiprocessor architecture provides better timing performances than a NoC with solely general-purpose processors (GP) or with multi-threaded processors (MTP). The authors have developed a methodology for task mapping and scheduling on a NoC with mixed SMTPs, MTPs and GPs resources, which aims to maximize the timing performances of concurrent applications and to verify their soft timing constraints. The methodology incorporates a timing analysis for estimating the response time of tasks executed in multithreaded mode of SMTPs. The experiments performed on synthetic benchmarks with average instruction parallelism of 200, have shown a maximum speedup of 2.91 for a 3times3 NoC with 9 SMTPs (4-issue, 4-multithreaded)