Stochastic Link-Fault-Tolerant Routing In A Torus
Seiya Kuramochi, Yu-on Chatchakan, Kousuke Mouri, Keiichi Kaneko · 2018
It is impossible to operate a parallel system that contains a large number of elements ignoring the existence of faulty elements. In fault-tolerant routing in a parallel system, the method to make each non-faulty vertex store the compressed fault information (limited global information) is effective. The torus is the most popular topology adopted for interconnection networks of parallel systems. However, because of its complexity, any effective method that utilizes the limited global information has not been proposed. In this paper, we propose a stochastic fault-tolerant routing method in tori, which can tolerate the link and/or vertex faults. We can expect that our result achieves drastic improvement regarding fault-tolerant routing in many parallel systems.