Exploring pattern-aware routing in generalized fat tree networks

Abstract

New static source routing algorithms for High Performance Computing (HPC) are presented in this work. The target parallel architectures are based on the commonly used fattree networks and their slimmed versions. The evaluation of such proposals and their comparison against currently used routing mechanisms have been driven by realistic traffic generated by HPC applications. Our experimental framework is based on the integration of two existing simulators, one replaying an MPI application and another simulating the network details. The resulting simulation platform has been fed with traces from real executions. We have obtained several interesting findings: (i) contrary to the widely accepted belief, random static routing in k-ary n-trees (which is the default option for InfiniBand and Myrinet technologies) is not a good solution for HPC applications; (ii) some existing oblivious routing techniques can be very good for certain communication patterns present on applications, but clearly fail for some others and (iii) one of the proposed pattern-aware routing algorithms could be used to better utilize network resources and thus achieve higher performance, particularly for the case of cost-effective networks. Copyright 2009 ACM.