The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|Optimized routing and process mapping for arbitrary network topologies|
(Presentation - presented in Tokyo, Japan, Jun. 2012, Tokyo Institute of Technology )
AbstractThe network topology is one of the most important parameters of large-scale systems. In this talk we discuss supporting techniques for arbitrary topologies with regards to performance and energy consumption. First and foremost, efficient deadlock-free routing strategies are crucial to the performance of large-scale computing systems. We demonstrate a novel routing strategy based on the single-source-shortest-path routing algorithm and extend it to use virtual channels to guarantee deadlock-freedom. We show that this algorithm achieves low latency, high-bandwidth with only a low number of virtual channels. We implemented the proposed algorithm in InfiniBand's Open Subnet Manager and compared the number of needed virtual channels and the bandwidth of multiple real and artificial network topologies to established practice. Application benchmarks showed an improvement of up to 95%. In addition, mapping application communication topologies to underlying network topologies is as important as optimized routing. We demonstrate an efficient and fast new heuristic which is based on graph similarity and show its utility with application communication patterns on real topologies. Our mapping strategies support arbitrary heterogeneous networks and show significant reduction of congestion on torus, fat-tree, and the PERCS network topologies for irregular problems, respectively. Our efficient topology mapping strategies are shown to reduce network congestion by up to 80%, reduce average dilation by up to 50%, and improve benchmarked communication performance by 18%. Those two techniques show a path towards new, potentially irregular network topologies for upcoming extreme-scale systems.