Time Division Multiplexed Control of All-Optical Interconnection Networks
Circuit switching techniques are preferred in optical multiprocessor interconnection networks because they do not require any optical to electronic signal conversion to route messages. Thus, they can provide all-optical paths between message sources and destinations and the large communication bandwidth of optical signals is available to the nodes attached to the network. However, circuit switching techniques increase the complexity of managing the network, and thus may increase the communication delay. In a massively parallel processor, this added delay may affect the performance of a parallel program. In this chapter we model the performance impact of multiplexed circuit switched network management techniques. We show how program performance is affected by the choice of circuit switching techniques, by the multiplexing degree, and by the characteristics of the network.
KeywordsTime Slot Network State Communication Cost Optical Network Network Management
Unable to display preview. Download preview PDF.
- F. Cappello and C. Germain. Toward high communication performance through compiled communications on a circuit switched interconnection network. Proceedings of the First IEEE Symposium on High-Performance Computer Architecture, pages 44 - 53, Jan. 1995.Google Scholar
- C. Clos. A study of non-blocking switching networks. Bell Systems Technical Journal, 15 (1): 406 - 424, 1953.Google Scholar
- P. Dowd. Random access protocols for high-speed interprocessor communications based on an optical passive star topology. IEEE Journal of Lightwave Technology, x(6): 799 - 808, 1991.Google Scholar
- K. Hwang. Advanced Computer Architecture. McGraw-Hill, New York, NY, 1993.Google Scholar
- D. Lahaut and C. Germain. Static communications in parallel scientific programs. In PARLE '94 Parallel Architecture and Languages. IEEE, July 1994.Google Scholar
- R. Melhem. Time-multiplexing optical interconnection networks; Why does it pay off? In Proceedings of the 1995 ICPP Workshop on Challenges for Parallel Processing, pages 30-35. CRC Press, Aug. 1995.Google Scholar
- P. Prucnal, I. Glesk, and J. Sokoloff. Demonstration of all-optical self-clocked demultiplexing of TDM data at 250Gó/s. In Proceedings of the First International Workshop on Massively Parallel Processing Using Optical Interconnections, pages 106-117. IEEE, Apr. 1994.Google Scholar
- C. Salisbury and R. Melhem. Modeling communication costs in multiplexed optical switching networks. In 11th International Parallel Processing Symposium,pages 71-79, Geneva, Switzerland, Apr. 1997. IEEE.Google Scholar
- X. Yuan, R. Melhem, and R. Gupta. Compiled communication for all-optical TDM networks. In Supercomputing '96. IEEE, Nov. 1996.Google Scholar