Evaluating Adaptive Routing Performance on Large-Scale Megaﬂy Topology
TimeThursday, 19 November 20208:30am - 5pm EDT
DescriptionThe Megaﬂy topology has recently been proposed as an eﬃcient, hierarchical way to interconnect large-scale high-performance computing systems. Megaﬂy networks may be constructed in various group sizes and conﬁgurations, but it is challenging to maintain high throughput performance on all such variants. A robust topology-speciﬁc adaptive routing scheme, therefore, is needed to utilize the topological advantages of Megaﬂy. Currently, Progressive Adaptive Routing (PAR) is the best known routing scheme for Megaﬂy networks, but its performance is not fully known across all scales and conﬁgurations. In this research poster, with the use of trace-driven simulation experiments, we show that the current PAR scheme performs sub-optimally on Megaﬂy networks with a large number of groups. As better alternatives, we propose two new practical adaptive routing schemes, KAPR and KU-GCN, that can improve the communication performance of Megaﬂy at any scale and conﬁguration.