A Unified Approach to the Performance Evaluation of Multiprocessors
This paper introduces a unified approach to approximately compute the bandwidth of tightly coupled multiprocessors, working either synchronously or asynchronously. Several types of interconnection networks are considered, namely: crossbar, multiple-bus and multiple-bus with partial-busses. The analysis is carried out for both constant and exponential memory service times. Simulation runs are also included to validate the approximation used in the model. Although the analysis is asymptotic, the computations turn out to be quite accurate for moderate, or even small, system sizes.
KeywordsInterconnection Network Markov Chain Model Memory Module Multiprocessor System Service Completion
Unable to display preview. Download preview PDF.
- M. Valero, J.M. Llaberia, J. Labarta, E. Sanvicente and T. Lang, “A performance evaluation of the multiple bus network for multiprocessor systems”, Proc. 1983 ACM Sigmetrics Conf., Minneapolis, USA, pp. 200–206, Aug. 1983.Google Scholar
- D. P. Bhandarkar, “Markov chain models for analyzing memory interference in multiprocessor computer systems”, Proc. I Ann. Symp. Comp. Architecture, pp. 1–6, Dec. 1973.Google Scholar
- Y.-C. Liu and C.-J. Jou, Effective memory bandwidth and processor blocking probability in multiple-bus systems, IEEE Trans. Comput., vol. C-3B, pp. 761–764, Jun. 1987.Google Scholar