Globally Asynchronous Locally Synchronous Simulation of NoCs on Many-Core Architectures

We evaluate the applicability of many-core architectures for the simulation of networks on chips (NoC). Compared to the well established shared memory multi-core architectures, many-core architectures significantly differ not only in the number of processing elements but also in the on-chip communication architecture, the memory subsystem, and the computational performance of an individual core. Proven multi-core simulation approaches do not consider such architectural aspects and thus suffer limited performance when being applied to many-core architectures. To enable high performance simulation, we identify conceptual drawbacks of state of the art parallel simulation approaches and consequently propose a novel globally asynchronous locally synchronous (GALS) simulation concept suited for many-core architectures. Our results show that our GALS simulation approach yields a speedup of up to 2.3 over parallel discrete event simulation.