Finite State Model and Compatibility Theory: New Analysis Tools for Permutation Networks
IEEE Transactions on Computers
Journal of the ACM (JACM)
Uniform theory of the shuffle-exchange type permutation networks
ISCA '83 Proceedings of the 10th annual international symposium on Computer architecture
The universality of various types of SIMD machine interconnection networks
ISCA '77 Proceedings of the 4th annual symposium on Computer architecture
IEEE Transactions on Computers
Hi-index | 0.00 |
In this paper we study the rearrangeability of multistage shuffle/exchange networks. Although a theoretical lower bound of (2 log2N - 1) stages for rearrangeability of a network with N = 2n inputs and outputs has been known, the sufficiency of (2 log2N - 1) stages has neither been proved nor disproved. The best known upper bound for rearrangeability is (3 log2N - 3) stages. We prove that, if (2 log2R - 1) shuffle/exchange stages are sufficient for rearrangeability of a network with R = 2' inputs and outputs, then, for any N R, 3 log2N - (r + 1) stages are sufficient for a network with N inputs and outputs. This result is established by setting some of the middle stages of the network to realize a fixed permutation and showing the reduced network to be topologically equivalent to a member of the Benes class of rearrangeable networks. We first characterize equivalence to Benes networks in set-theoretic terms and use this to prove equivalence of the reduced shuffle/exchange network to the Benes network. From the known result that 5 stages are sufficient for rearrangeability when N = 8, we obtain an upper bound of (3 log2N - 4) stages for rearrangeability when N ≥ 8. Further, any increase in the network size R for which the rearrangeability of (2 log2R - 1) stages could be shown, results in a corresponding improvement in the upper bound for all N ≥ R. In addition, due to the one-to-one correspondence that exists between the switches in the reduced shuffle/exchange network and those in the Benes network, the former network can be controlled by the well-known looping algorithm.