Student:
Supervisor: Yuefan Deng (Stony Brook University)
Abstract: We propose a scheme to construct a family of large and high-performance interconnection networks that are scalable, low-radix, minimum diameters. These networks, whose diameters grow linearly as their sizes grow exponentially, are generated by using the Cartesian products of smaller optimal networks of minimum diameters. For the smaller base networks, we design the vertex-balanced routing algorithm by considering the forwarding pressure at each vertex. Comparative benchmarks on a Beowulf cluster show significant improvement in performance after using the new routing algorithm. Each node of the new network generated from base graphs with low-diameter can also sustain balanced forwarding loadings if we apply optimal routing algorithms to the base network. Simulation results for larger networks show that the optimal routing algorithms achieve the gain of communication performance.
ACM-SRC Semi-Finalist: no
Poster: PDF
Poster Summary: PDF
Back to Poster Archive Listing