Abstract :
Proposed for MPCs, RDT has already been proved to support better performances than most of the direct network, such as smaller diameter based on its vector routing algorithm, smaller degree, embedded mesh/torus topology, etc. It gives a network form method to form groups of networks with recursively structured torus/torus-like networks. A RDT structure can be formed with the method and its related parameters. After analyzing its performance by the average distance, the descending vector routing reducing redundant computations and the fault-tolerant routing algorithm are proposed. A simpler structure named RDT(2,2,1)/a are proposed for the parallel systems with no more than 5000 nodes and given its performance evaluation. Considering the structure features of the practical RDTs, they are suitable for the interconnection network of an NoC design. The design of laying out RDT(2,2, 1)/a is proposed and proved to be available
Keywords :
fault tolerant computing; multiprocessor interconnection networks; parallel processing; performance evaluation; telecommunication network routing; NoC design; descending vector routing; fault-tolerant routing algorithm; interconnection network; massively parallel computer; parallel system; recursive diagonal torus; torus-like network; Algorithm design and analysis; Computer networks; Concurrent computing; Fault tolerant systems; Multiprocessor interconnection networks; Network topology; Network-on-a-chip; Performance analysis; Performance evaluation; Routing;
