A zoned node interconnection for the optimisation of fat tree class topologies

Fat tree topologies have been extensively used as interconnection networks for high performance parallel systems, with their most recent variants able to easily extend and scale to accommodate different system sizes and requirements. While each progressive fat-tree topology includes some extra advancements compared to the original ones, these topologies do not fully address or resolve the issues that faces large scales systems. We proposed in this paper, a zoned node tree architecture that generalises most variants of the fat trees topologies as k-ary n-tree and XGFT. Furthermore, we can construct a replication of super nodes that generalises complex interconnections like dragonfly and PERCS. We also propose a simple routing algorithm called the sliced-addressing that works
for all the classes of the zoned nodes and prove with simulation that it outperforms the previous topologies of the fat tree family.