A holistic comparison of static VC allocation versus dynamic VC allocation based NoC routers

Router architectures advocating the use of static as well as dynamic virtual channel allocation have been proposed in the literature. However, the comparisons between the two have primarily focused on simulation set-up, largely using synthetic benchmarks. In this work, we present an apple-to-apple comparison of the dynamic and static VC allocation based routers for network performance. We compare the static VC allocation and dynamic VC allocation based routers for real and synthetic benchmarks as well as their implementations. The performance of the two routers has been compared for different routing algorithms, topologies, packet sizes and input port configurations. We observe that for all synthetic traffic patterns, except uniform traffic, static VC allocation based router outperforms the dynamic VC allocation based router. We also observe that in almost all PARSEC benchmark applications, except Blackscholes and Bodytrack with DOR routing, the simplified EVA router outperforms the dynamic VC allocating router. In our ASIC implementation, we observe a clock frequency improvement of 20.2% and power reduction of 19.6% for the static VC allocation based router over the dynamic VC allocation based router.