Title :
Maximizing Throughput on a Dragonfly Network
Author :
Jain, Nikhil ; Bhatele, Abhinav ; Xiang Ni ; Wright, N.J. ; Kale, Laxmikant V.
Author_Institution :
Dept. of Comput. Sci., Univ. of Illinois at Urbana-Champaign, Urbana, IL, USA
Abstract :
Interconnection networks are a critical resource for large supercomputers. The dragonfly topology, which provides a low network diameter and large bisection bandwidth, is being explored as a promising option for building multi-Petaflop´s and Exaflop´s systems. Unlike the extensively studied torus networks, the best choices of message routing and job placement strategies for the dragonfly topology are not well understood. This paper aims at analyzing the behavior of a machine built using a dragonfly network for various routing strategies, job placement policies, and application communication patterns. Our study is based on a novel model that predicts traffic on individual links for direct, indirect, and adaptive routing strategies. We analyze results for individual communication patterns and some common parallel job workloads. The predictions presented in this paper are for a 100+ Petaflop´s prototype machine with 92,160 high radix routers and 8.8 million cores.
Keywords :
multiprocessor interconnection networks; parallel machines; telecommunication network routing; telecommunication network topology; Petaflop prototype machine; application communication pattern; dragonfly topology; interconnection network; job placement policy; routing strategy; supercomputer; throughput maximization; Adaptation models; Bandwidth; Network topology; Predictive models; Routing; Throughput; Topology; dragonfly networks; job placement; modeling; prediction; routing;
Conference_Titel :
High Performance Computing, Networking, Storage and Analysis, SC14: International Conference for
Conference_Location :
New Orleans, LA
Print_ISBN :
978-1-4799-5499-5
