Head-of-line blocking avoidance in Slim Fly networks using deadlock-free non-minimal and adaptive routing

摘要

Interconnection network performance is a key issue in HPC systems and datacenters, especiallyas theirnumberof endnodes grows, to cope with applicationneeds.Thenetwork topologyandtherouting algorithm are important factors for performance and cost. Topologies such as fat-tree orDragonfly were proposed to maximize network performance while reducing network resources.One of the most promising topologies is Slim Fly, which offers high network bandwidth assuringlow network diameter. However, adversarial traffic and/or congestion situations may degradeSlim Fly's performance dramatically. Non-minimal routings, such as Valiant or UGAL, can mitigatethe former problem while queuing schemes can handle the latter one. In this paper, weproposed a combined mechanism to provide Slim Fly network with both non-minimal routing andqueuing schemes by using several virtual networks to guarantee deadlock freedom. Each virtualnetwork consists of a set of virtual channels to store packets separately according to a mappingpolicy. This diminishes the interaction among traffic flows, thus reducing head-of-line blocking.The results obtained from a simulation-based evaluation show that our proposal enhances theperformance in all the traffic cases, in contrast to other mechanisms whose performance drops incertain scenarios.