Switch design to enable predictive multiplexed switching in multiprocessor networks

Predictive multiplexed switching is a new approach for building interconnection switches for high performance parallel systems. This approach advocates sacrificing some link bandwidth in return for more efficient network control and simpler connection management. The main idea is to depart from the traditional packet and wormhole switching in favor of row data communication over established communication pipes (connections). The overhead of this circuit switching approach can be justified when established connections are repeatedly used before they are torn down. For this, we use multiplexing to allow multiple connections to share the same resources (links and switches), thus avoiding tearing down connections prematurely. The connection establishment overhead is further reduced by exploring communication locality and predictability in applications that exhibit these properties. We present the design of an interconnection system which is based on multiplexed switching and which establishes connections either reactively, in response to dynamically generated requests, or proactively, in response to compiler or application directives. A communication prediction component may be supported to reduce the network control overhead in applications that exhibit communication locality and predictability. The design is evaluated using hardware design, synthesis, and cycle-accurate simulation. Comparison with more traditional switching paradigms shows the potential of our predictive multiplexed switching approach.

[1]  Fabrizio Petrini,et al.  BCS-MPI: A New Approach in the System Software Design for Large-Scale Parallel Computers , 2003, SC.

[2]  Rami Melhem,et al.  A high speed scheduler/controller for unbuffered banyan networks , 1998, ICC '98. 1998 IEEE International Conference on Communications. Conference Record. Affiliated with SUPERCOMM'98 (Cat. No.98CH36220).

[3]  Rami G. Melhem,et al.  Algorithms for Supporting Compiled Communication , 2003, IEEE Trans. Parallel Distributed Syst..

[4]  Henry G. Dietz,et al.  Compiler Techniques for Flat Neighborhood Networks , 2000, LCPC.

[5]  Rami Melhem Time-Multiplexing Optical Interconnection Networks; Why Does it Pay Off? , 1995 .

[6]  Rami G. Melhem,et al.  Dynamic Reconfiguration of Optically Interconnected Networks with Time-Division Multiplexing , 1994, J. Parallel Distributed Comput..

[7]  Donald M. Chiarulli,et al.  Predicting Multiprocessor Memory Access Patterns with Learning Models , 1997, ICML.

[8]  Walid A. Najjar,et al.  Analysis of communications and overhead reduction in multithreaded execution , 1995, PACT.

[9]  Sudhakar Yalamanchili,et al.  Interconnection Networks: An Engineering Approach , 2002 .

[10]  Cécile Germain,et al.  Static Communications in Parallel Scientific Propgrams , 1994, PARLE.

[11]  Russell Tessier,et al.  ASOC: a scalable, single-chip communications architecture , 2000, Proceedings 2000 International Conference on Parallel Architectures and Compilation Techniques (Cat. No.PR00622).

[12]  Thomas R. Gross,et al.  Communication styles for parallel systems , 1994, Computer.

[13]  Nikitas J. Dimopoulos,et al.  Design and evaluation of communication latency hiding/reduction techniques for message-passing environments , 2000 .

[14]  William Thies,et al.  StreamIt: A Language for Streaming Applications , 2002, CC.

[15]  Timothy Mark Pinkston,et al.  A methodology for designing efficient on-chip interconnects on well-behaved communication patterns , 2003, The Ninth International Symposium on High-Performance Computer Architecture, 2003. HPCA-9 2003. Proceedings..

[16]  Kirk L. Johnson The impact of communication locality on large-scale multiprocessor performance , 1992, ISCA '92.

[17]  Ahmad Faraj,et al.  Communication Characteristics in the NAS Parallel Benchmarks , 2002, IASTED PDCS.

[18]  Thomas R. Gross,et al.  Compiler directed architecture-dependent communication optimizations , 1995 .

[19]  Stefanos Kaxiras,et al.  Coherence communication prediction in shared-memory multiprocessors , 2000, Proceedings Sixth International Symposium on High-Performance Computer Architecture. HPCA-6 (Cat. No.PR00550).

[20]  James R. Larus,et al.  Compiler-directed Shared-Memory Communication for Iterative Parallel Applications , 1996, Proceedings of the 1996 ACM/IEEE Conference on Supercomputing.

[21]  David J. Lilja,et al.  Characterization of Communication Patterns in Message-Passing Parallel Scientific Application Programs , 1998, CANPC.

[22]  Thomas R. Gross Communication in iWarp systems , 1989, Proceedings of the 1989 ACM/IEEE Conference on Supercomputing (Supercomputing '89).

[23]  F. Cappello,et al.  Toward high communication performance through compiled communications on a circuit switched interconnection network , 1995, Proceedings of 1995 1st IEEE Symposium on High Performance Computer Architecture.