MMAR:A deadlock recovery-based fault tolerant routing algorithm for mesh/torus networks

Abstract

In direct networks, such as mesh and torus, the switching capacity will increase as the number of components increase. But the fault probability of the network also increases with the increasing of components. This paper proposes a novel fault-tolerant algorithm, named as minimal misrouted adaptive routing (MMAR) which is based on true fully adaptive routing algorithm and deadlock recovery mechanism. Due to the high adaptability, MMAR can accommodate arbitrary shaped fault models using minimal number of virtual channels in each physical link. When encountering concave fault models, MMAR minimizes the length of the misrouted path by avoiding routing the message into the irrespective holes. Simulation results show that MMAR can work efficiently and achieve favorable performance.