Bus based parallel computers: A viable way for massive parallelism
- First Online:
Abstract
In most distributed memory MIMD multiprocessors, processors are connected by a point-to-point interconnection network. Since interprocessor communication frequently constitutes serious bottlenecks, several architectures were proposed that enhance point-to-point topologies with the help of multiple bus systems so as to improve the communication efficiency. In this paper we study global communication on parallel architectures where the communication means are constituted solely by busses. We focus on the hyperpath and the hypergrid architectures, which are the bus-based versions of the well used point-to-point linear and grid interconnection networks. Using (hyper) graph theoretic concepts in order to model inter-processor communication in such networks, we developed a new tool called simplification in order to give extremely efficient algorithms for gossiping (all-to-all or total exchange communications).
Preview
Unable to display preview. Download preview PDF.
References
- 1.A. Aggarwal, Optimal bounds for finding maximum on array of processors with k global busses, IEEE Transactions on Computers 35 (1986) 62–64.Google Scholar
- 2.C. Berge, Hypergraphs, (North Holland, 1989).Google Scholar
- 3.J-C. Bermond, J. Bond and C. Peyrat, Interconnection networks with each node on two busses, in Parallel Algorithms and Architectures, eds. M. Cosnard, Y. Robert, P. Quinton and M. Tchunte, (North Holland, 1984) 155–167.Google Scholar
- 4.J-C. Bermond and F.O. Ergincan, Bus interconnection networks, Technical Report No. 93-56, Université de Nice — Sophia Antipolis, September 1993.Google Scholar
- 5.L.N. Bhuyan and D.P. Agrawal, Generalized hypercube and hyperbus structures for a computer network, IEEE Transactions on Computers 33 (1984) 323–333.Google Scholar
- 6.G. Cybenko, D.W, Krumme, and K.N. Venkataraman, Gossiping in minimum time, SIAM Journal on Computing, Vol.21, no. 1, pp. 111–139, Feb 1992.MathSciNetGoogle Scholar
- 7.A. Ferreira, A. Goldman vel Lejbman, and S. W. Song. Towards simple massively parallel systems: Bus based parallel computers and new communication patterns. Reserach Report 93-18, Laboratoire de l'Informatique du Parallélisme, Ecole Normale Supérieure de Lyon, France, June 1993.Google Scholar
- 8.C.M. Fiduccia, Bused hypercubes and other pin-optimal networks, IEEE Transactions on Parallel and Distributed Systems 3(1) (1992) 14–24.CrossRefGoogle Scholar
- 9.P. Fraigniaud and E. Lazard, Methods and problems of communication in usual networks, Rapport de Recherche LIP 91-33, LIP/ENS-Lyon, October, 1991, to appear in Discrete Applied Mathematics.Google Scholar
- 10.H. Jiang and K.C. Smith, PPMB: a partial-multiple-bus multiprocessor architecture with improved cost-effectiveness, IEEE Transactions on Computers 41 (1992) 361–366.CrossRefGoogle Scholar
- 11.S. Olariu, J.L. Schwing and J.Y. Zhang, On the power of two-dimensional processor arrays with reconfigurable bus systems, Parallel Processing Letters 1 (1991) 29–34.CrossRefGoogle Scholar
- 12.I.D. Scherson, Orthogonal graphs for the construction of a class of interconnection networks, IEEE Transactions on Parallel and Distributed Systems 2(1) (1991) 3–18.CrossRefGoogle Scholar
- 13.T. Szymanski, Graph-theoretic models for photonic networks, Proceedings of the New Frontiers: A Workshop on Future Directions of Massively Parallel Processing, McLean, VA, 1992, 85–96.Google Scholar
- 14.B. Wilkinson, On crossbar switch and multiple bus interconnection networks with overlapping connectivity, IEEE Transactions on Computers 41 (1992) 738–746.CrossRefGoogle Scholar