On Folded-Clos Networks with Deterministic Single-Path Routing

Folded-Clos networks, also known as fat-trees, have been widely used as interconnects in large-scale high-performance computing clusters. Although users often treat such interconnects as replacements of nonblocking crossbar switches that can carry out any permutation communication without contention, the networking capability of such interconnects without a centralized controller in computer communication environments is not well understood. In this article, we investigate nonblocking two-level folded-Clos networks with deterministic single-path routing, but no centralized controller, and establish the nonblocking condition. The results indicate that nonblocking two-level folded-Clos networks without a centralized controller are much more expensive to construct than the traditional nonblocking networks in the telecommunication environment. Practical two-level folded-Clos based interconnects are blocking. For such interconnects, we establish the lower bound for worst-case contention for permutations with any deterministic single-path routing scheme, show that existing routing schemes perform poorly in terms of worst-case contention for permutations, present a routing scheme that achieves the theoretical optimal, and empirically compare the performance of existing schemes with the optimal routing scheme. The techniques developed for two-level folded-Clos networks are further extended for the general fat-trees of any heights.

[1]  Eli Upfal,et al.  Efficient Algorithms for All-to-All Communications in Multiport Message-Passing Systems , 1997, IEEE Trans. Parallel Distributed Syst..

[2]  Torsten Hoefler,et al.  Adaptive Routing Strategies for Modern High Performance Networks , 2008, 2008 16th IEEE Symposium on High Performance Interconnects.

[3]  Mohan Kumar,et al.  On generalized fat trees , 1995, Proceedings of 9th International Parallel Processing Symposium.

[4]  Torsten Hoefler,et al.  Multistage switches are not crossbars: Effects of static routing in high-performance networks , 2008, 2008 IEEE International Conference on Cluster Computing.

[5]  Pedro López,et al.  Improving routing performance in Myrinet networks , 2000, Proceedings 14th International Parallel and Distributed Processing Symposium. IPDPS 2000.

[6]  Yuanyuan Yang,et al.  Wide-Sense Nonblocking Clos Networks Under Packing Strategy , 1999, IEEE Trans. Computers.

[7]  V. Benes On rearrangeable three-stage connecting networks , 1962 .

[8]  Mateo Valero,et al.  Oblivious routing schemes in extended generalized Fat Tree networks , 2009, 2009 IEEE International Conference on Cluster Computing and Workshops.

[9]  Fabrizio Petrini,et al.  k-ary n-trees: high performance networks for massively parallel architectures , 1997, Proceedings 11th International Parallel Processing Symposium.

[10]  Eitan Zahavi Fat-tree routing and node ordering providing contention free traffic for MPI global collectives , 2012, J. Parallel Distributed Comput..

[11]  Torsten Hoefler,et al.  ORCS : An Oblivious Routing Congestion Simulator , 2009 .

[12]  Xin Yuan,et al.  On LID assignment in infiniBand networks , 2007, ANCS '07.

[13]  Darren J. Kerbyson,et al.  Optimized InfiniBand TM fat-tree routing for shift all-to-all communication patterns , 2010, ISC 2010.

[14]  Rami G. Melhem,et al.  Oblivious Routing in Fat-Tree Based System Area Networks With Uncertain Traffic Demands , 2007, IEEE/ACM Transactions on Networking.

[15]  Romeo Rizzi,et al.  Edge-Coloring Bipartite Graphs , 2000, J. Algorithms.

[16]  Charles Clos,et al.  A study of non-blocking switching networks , 1953 .

[17]  Pedro López,et al.  Deterministic versus Adaptive Routing in Fat-Trees , 2007, 2007 IEEE International Parallel and Distributed Processing Symposium.

[18]  V. Benes,et al.  Mathematical Theory of Connecting Networks and Telephone Traffic. , 1966 .

[19]  José Duato,et al.  Deadlock-Free Routing in InfiniBand through Destination Renaming , 2001 .

[20]  Charles E. Leiserson,et al.  Fat-trees: Universal networks for hardware-efficient supercomputing , 1985, IEEE Transactions on Computers.

[21]  William J. Dally,et al.  Adaptive Routing in High-Radix Clos Network , 2006, ACM/IEEE SC 2006 Conference (SC'06).

[22]  B GibbonsPhillip ACM transactions on parallel computing , 2014 .

[23]  Charles E. Leiserson,et al.  Randomized Routing on Fat-Trees , 1989, Adv. Comput. Res..

[24]  Craig B. Stunkel,et al.  Adaptive source routing in multistage interconnection networks , 1996, Proceedings of International Conference on Parallel Processing.

[25]  Yeh-Ching Chung,et al.  A multiple LID routing scheme for fat-tree-based InfiniBand networks , 2004, 18th International Parallel and Distributed Processing Symposium, 2004. Proceedings..

[26]  Michael Lang,et al.  Optimized InfiniBandTM fat‐tree routing for shift all‐to‐all communication patterns , 2010, Concurr. Comput. Pract. Exp..

[27]  Xin Yuan,et al.  On Nonblocking Folded-Clos Networks in Computer Communication Environments , 2011, 2011 IEEE International Parallel & Distributed Processing Symposium.

[28]  Michael Lang,et al.  Optimization of infiniband for scientific applications , 2008, 2008 IEEE International Symposium on Parallel and Distributed Processing.