Fault-tolerant routing in 2D tori or meshes using limited-global-safety information

Abstract

A limited-global-safety-information-based metric called local safety is proposed to handle fault-tolerant routing in 2D tori (or meshes). Sufficient conditions for existence of a minimum feasible path between the source and destination is presented based on local safety information in a 2D torus network. An efficient heuristic function is defined to guide fault-tolerant routing inside a 2D torus network. Unlike the conventional methods based on the block fault model, our method does not disable any fault-free nodes and fault-free nodes inside a fault block can still be a source or a destination, which can greatly increase throughput and computational power of the system. Techniques for avoidance of deadlocks are introduced. Extensive simulation results are presented.