Multiple Omega Networks for Parallel Processing

  • José Eduardo Moreira University of Illinois at Urbana-Champaign / USP

Resumo


In this paper we propose the use of multiple Omega networks as an interconnection system for shared memory multiprocessors. This allows us to achieve a much higher bandwidth of communication, accommodating the needs of current high-performance processors, including those with multiple memory ports. We also obtain a very scalable system, by defining a processor-switch-memory building block that can be used in systems with processor count in the range of a few units to several thousands. The performance evaluation of multiple Omega networks is clone through a simple analytical model that allows us to compare their performance to a that of a single network, and investigate alternatives for processors with multiple memory ports. The results show that the performance (in terms of bandwidth and latency of communication) of systems with multiple networks is more stable with respect to variations in systems parameters such as number of processors and memory access rate, than that of systems with just a single network.

Referências

John B. Andrews, Carl J. Beckmann, and David K. Poulsen. Notification and Multicast Networks for Synchronization and Coherence. Journal of Parallel and Disrtributed Computing, (15), 1992.

Mark A. Franklin and Sanjay Dhar. On Designing Inteconnection Networks for Multiprocessors. In Proceedings of the 1986 International Conference on Parallel Processing, pages 208-215, St. Charles. Illinois, 1986.

William T. Hsu and Pen-Chung Yew. The Performance of Hierarchical Systems with Wiring Constraints. In Proceedings of the 1991 International Conference on Parallel Processing, pages 1:9-16, St. Charles, Illinois, 1991.

Young Man Kim and Kyungsook Y. Lee. Performance Analysis of Buffered Banyan Network under Nonuniform Traffic. In Proceedings of the 1989 International Conference on Supercomputing, pages 452-460. Crete, Greece, June 5-9, 1989.

J. Konicek et al. The Organization of the Cedar System. In Proceedings of the 1991 International Conference on Parallel Processing. pages I:49-56. St. Charles. Illinois. 1991.

Duncan H. Lawrie. Access and Aligament of Data in an Array Processor. IEEE Transactions on Computers. C-24(12):1145-1155. December 1975.

T. N. Mudge and B. A. Makrucki. Probabilistic Analysis of a Crossbar Switch. In Proceedings of the 9th Annual International Symposium on Computer Architecture, pages 311-319, Austin, Texas, April 26-29, 1982.

Janak H. Patel. Performance of Processor - Memory Interconnections for Multiprocessors. IEEE Transactions on Computers, C-30(10):771-780, October 1981.

Mahib Rahman and David G. Meyer. General Analytic Models for the Performance Analysis of Unique and Redundant Path Interconnection Networks. In Proceedings of the 1991 International Conference on Parallel Processing, pages I:584-591, St. Charles, Illinois, 1991.

K. A. Robbins and S. Robbins. Bus Conflicts for Logical Memory Banks on a Cray Y-MP type Processor System. In Proceedings of the 1991 International Conference on Parallel Processing, pages I:21-24. St. Charles, Illinois, 1991.

Honda Shing and Lionel M. Ni. A Conflict-Free Memory Design for Multiprocessors. In Proceedings of Supercomputing'91, pages 46-55, Albuquerque. New Mexico, November 18-22, 1991.

Howard Jay Siegel and William Tsun yuk Hsu. Computer Architecture - Concepts and Systems, chapter Interconnection Networks. North-Holland. 1988.

J. E. Smith, W.-C. Hsu, and C. Hsiung. Future General Purpose Supercomputer Architectures. In Proceedings of Supercomputing 90. pages 796-804, New York, New York, November 12-16, 1990.

J. E. Smith and W. R. Taylor. Accurate Modeling of Interconnection Networks in Vector Supercomputers. In Proceedings of the 1991 International Conference on Supercomputing, pages 264-273. Cologne, Germany, June 17-21, 1991.

Ted Szymanski and Chien Fang. Design and Analysis of Buffered Crossbars and Banyans with Cut-Through Switching. In Proceedings of Supercomputing '90, pages 264-273, New York, New York. November 12-16, 1990.

Peixiong Tang and Raul H. Mendez. Memory Conflicts and Machine Performance. In Proceedings of Supercomputing 89, pages 826-831, Reno, Nevada, November 13-17, 1989.

Chuan-Lin Wu and Tse-Yun Feng. The Universality of the Shuffle-Exchange Network. IEEE Transactions on Computers, C-30(5):324-332. May 1981.
Publicado
26/10/1992
MOREIRA, José Eduardo. Multiple Omega Networks for Parallel Processing. In: INTERNATIONAL SYMPOSIUM ON COMPUTER ARCHITECTURE AND HIGH PERFORMANCE COMPUTING (SBAC-PAD), 4. , 1992, São Paulo/SP. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 1992 . p. 195-212. DOI: https://doi.org/10.5753/sbac-pad.1992.22711.