Static and dynamic task mapping onto network on chip multiprocessors

Abstract

Due to its scalability and flexibility, Network-on-Chip (NoC) is a growing and promising communication paradigm for Multiprocessor System-on-Chip (MPSoC) design. As the manufacturing process scales down to the deep submicron domain and the complexity of the system increases, fault-tolerant design strategies are gaining increased relevance. This paper exhibits the use of a Population-Based Incremental Learning (PBIL) algorithm aimed at finding the best mapping solutions at design time, as well as to finding the optimal remapping solution, in presence of single-node failures on the NoC. The optimization objectives in both cases are the application completion time and the network's peak bandwidth. A deterministic XY routing algorithm was used in order to simulate the traffic conditions in the network which has a 2D mesh topology. Obtained results are promising. The proposed algorithm exhibits a better performance, when compared with other reported approaches, as the problem size increases.