Improved scaling simulation of the general reconfigurable mesh
The reconfigurable mesh (R-Mesh) has drawn much interest in recent years, due in part to its ability to admit extremely fast algorithms for a large number of problems. For these algorithms to be useful in practice, the R-Mesh must be scalable; that is, any algorithm designed for a large R-Mesh should be able to run on a smaller R-Mesh without significant loss of efficiency. This amounts to designing a ‘scaling simulation” that simulates an arbitrary step of an N×N R-Mesh on a smaller P×P R-Mesh in O(N 2/P 2 f(N, P)) steps; f(N,P) is a nondecreasing function representing the simulation overhead. The aim is to minimize this overhead, ideally to a constant.
In this paper, we present a scaling simulation for the general (unconstrained) R-Mesh. This simulation has an overhead of log N (smaller than the log PlogN/P overhead of the previous fastest scaling simulation), using a CREW LRN-Mesh (a weaker version of the General R-Mesh) as the simulating model; prior simulations needed concurrent write.
Unable to display preview. Download preview PDF.
- 4.Fernández-Zepeda, J.A., Trahan, J.L., Vaidyanathan, R.: Scaling the FR-Mesh under Different Concurrent Write Rules. World Multiconf. on Systemics, Cybernetics, and Informatics (1997) 437–444Google Scholar
- 9.Matias, Y., Schuster, A.: Fast, Efficient Mutual and Self Simulations for Shared Memory and Reconfigurable Mesh. 7th IEEE Symp. Par. Distrib. Processing (1995) 238–246Google Scholar
- 11.Pavel, S., Akl, S. G.: On the Power of Arrays with Optical Pipelined Buses. Int’l. Conf. Par. Distr. Proc. Techniques and Appl. (1996) 1443–1454Google Scholar
- 13.Trahan, J. L., Vaidyanathan, R.: Relative Scalability of the Reconfigurable Multiple Bus Machine. Proc. Workshop Reconfigurable Arch. and Algs. (1996)Google Scholar