A decoupled multicomputer architecture with optical full interconnection
Decoupled architectures have been proposed in sequential processing to increase performance when memory access delays are a bottle-neck. Communication delays are expected to be the major bottle-neck in future large-scale parallel computers.
Although optical technology is often used to improve the throughput of each connection, such communication systems still have significant data transfer delays. We believe that providing more parallel connections, by increasing the density of topologies, is a more effective approach to reduce the communication bottle-neck.
We are currently evaluating the first prototype of a new decoupled multicomputer architecture supporting static allocation and static scheduling of data processing tasks. The decoupling of communication and computation in this prototype not only supports efficient allocation of data processing resources, but at the same time also allows efficient scheduling of data transfers on the interconnection system to minimize the effect of data transfer delays.
Unable to display preview. Download preview PDF.
- Acampora A.S., Karol M.J., Hluchyj M.G., “Multihop lightwave networks: A new approach to achieve terabit capabilities”, Proc. ICC'88, vol. 1, pp 1478–1484,1988.Google Scholar
- Bruno, J.L. and Downey, P.J., “Probabilistic bounds on the performance of list scheduling”, SIAM Journal on Computing, Vol. 15, No.2, May 1986.Google Scholar
- Clymer D. C., Goodman J.W., “Optical clock distribution to silicon chips”, Optical Engineering, Vol. 25, No. 10, October 1986.Google Scholar
- Dowd P.W., “High Performance Interprocessor Communication Through Optical Wavelength Division Multiple Access Channels”, Proc. 18th International Symposium on Computer Architecture, Toronto, Canada, May 1991.Google Scholar
- Fisher J.A., “The VLIW machine: a multiprocessor for compiling scientific code”, IEEE Computer, Vol. 17, No. 7, July 1984.Google Scholar
- Fried J.A., “Optical I/O for high speed CMOS systems”, Optical Engineering, Vol. 25, No. 10, October 1986.Google Scholar
- Frietman E. E. E, Ruighaver A.B., “An Electro-Optic Data Communication System for the Delft Parallel Processor”, Computer Architecture News, Vol. 15, No. 4, 1987.Google Scholar
- Ruighaver, A.B., “Hierarchical Programming and the DPP84”, Ph.D.Thesis, Technical University Delft, December 1986.Google Scholar
- Ruighaver, A.B., “A modular network for dense optical interconnection of Processing Elements”, Computer Architecture News, Vol 18, No. 2, June 1990.Google Scholar
- Ruighaver, A.B., “The Melbourne University Optoelectronic Multicomputer Project” Supercomputer, Vol. VIII, No. 6, November 1991, pp. 22–32.Google Scholar
- Smith J.E., Weiss S., Pang N.Y., “A Simulation Study of Decoupled Architecture Computers”, IEEE Transactions on Computers, Vol. C35, No. 8, August 1986.Google Scholar