The ability to tolerate faults is critical in multicomputers employing
large numbers of processors. This paper describes a class of fault-to
lerant routing algorithms for n-dimensional meshes that can tolerate l
arge numbers of faults without using virtual channels. We show that th
ese routing algorithms prevent livelock and deadlock while remaining h
ighly adaptive.