Share Email Print

Proceedings Paper

Optical communication networks for multiprocessor computers
Author(s): Vladimir P. Torchigin; D. S. Strebezh
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Nowadays massively parallel computer systems consisted of many thousands of processor elements (PE) are reality. Their main component is so called communications network (CN) that provides for data exchanging between PEs. CN represents most of the cost of the computer, most of the power dissipation, most of the wiring etc. CN consists of communications processors (CP) or routers. Each PE has its own CP. CPs are connected by transmission lines in some regular way determining CN topology. To use the advantages of broadband optical transmission lines most efficiently it makes sense to take an approach when an optical computer consists of a number of identical virtual PEs, i.d. presents massively parallel computer1. One actual optical PE implements N virtual PEs , where N=T/T, T is the delay time in propagating signals between all-optical gates, Tis the period of their switching. N is usually equal to iO-iO. Time interval T is divided into N identical time-slots. Each of the virtual PEs operates in its own time-slot. The number of PEs coincides with the number of timeslots. Thus the data transmission between different PEs reduces to the data redistribution between time-slots. We have drawn the conclusion that N virtual CPs can be obtained on the basis of one actual CP being consisted of the above-mentioned optical gates. CNs of various topologies making possible a necessary trade-off of the number of gates against the transmission speed can be easily designed. The total number of transmission lines in the optical CN is N times less than it is in the similar electronic CN. The most simple an optical CN of 2D grid type consisted of n columns and N/n rows of PEs has been considered before1 . There is also a universal time-slot interchanger that is the time domain equivalent of a multiple input, multiple output spatial interconnection network2. There are 2n-1 optoelectronic switches connected in series for universal time-slot permutation of serial 1-bit array of size N=21 time-slots. Two basic features of the interchanger ought to note. First, the optoelectronic switches are controlled by electrical signals generated by special electronic generator. This factor limits the decrease of value of time-slots and prevents the taking full advantage of very high potential of broadband of optical channels. The electronic gates used in the generators could be used instead of the optoelectronic switches also. Delay lines could be replaced by electronic shift registers. Second, there is no self-routing. The permutation is defined by some external electronic generator. These features limit the possibilities of optical multiprocessor computer.

Paper Details

Date Published: 12 July 1993
PDF: 9 pages
Proc. SPIE 1806, Optical Computing, (12 July 1993); doi: 10.1117/12.147892
Show Author Affiliations
Vladimir P. Torchigin, Scientific Computer Ctr. (Russia)
D. S. Strebezh, Scientific Computer Ctr. (Russia)

Published in SPIE Proceedings Vol. 1806:
Optical Computing
Andrey M. Goncharenko; Fedor V. Karpushko; George V. Sinitsyn; Sergey P. Apanasevich, Editor(s)

© SPIE. Terms of Use
Back to Top