Fault-tolerant routing algorithms based on optimal path matrices

Presents a new concept - optimal path matrices (OPMs) - for fault-tolerant routing on hypercube multicomputers. OPMs stored on each node of a hypercube hold the fault information and indicate whether there is an optimal path from the node to a destination. Two fault-tolerant routing algorithms based on OPMs are proposed in order to maintain the matrices and to route messages from sources to destinations. One is a routing algorithm based on OPMs, which are filled with pre-collected information through information exchange between neighbors. It can easily establish a path for a message, and the length of the path is no greater than the Hamming distance between the source and the destination of the message plus two. The other routing algorithm is a modified depth-first search algorithm, which adopts a dynamic learning strategy to fill OPMs during normal message transmission and can find almost all the optimal paths. The memory overhead is n² words on each node of an n-dimensional hypercube