Run-Time Computation and Communication Aware Mapping Heuristic for NoC-Based Heterogeneous MPSoC Platforms

摘要

The rapid increase in the complexity of real-life applications has led to the perpetual demand of refined architectural designs. Multiprocessor systems-on-chip (MPSoC) emerges as one of the possible solution for satiating such enormous computational needs. These MPSoCs are employed with Network-On-Chip (NoC) interconnect for power efficient and scalable inter-communication required between processors. Mapping parallelized tasks of applications onto these MPSoCs is the next gigantic problem, which can be done either at design-time or at run-time. However, design-time strategies may sometimes provide a more optimal mapping but they are restricted to predefined set of applications and seem incapable of run-time resource management. On the contrary, run-time mapping techniques overcome this limitation by determining the state of the platform and incorporating resource management before mapping. This paper describes a heuristic for run-time mapping of parallelized tasks of an application considering efficient computation, communication and resource utilization as the main parameters for optimization.