EECS 298-11: Special CAD Seminar Friday, May 3, 1996, 11am 400 Cory Hall, Hughes Room Models for Performance Prediction of Coherence Protocols Sinisa Srbljic AT&T, San Mateo, CA sinisa@geoplex.com Dept. of Electrical & Computer Engineering University of Toronto, Toronto, Ontario, Canada sinisa@eecg.toronto.edu Faculty of Electrical Engineering and Computing University of Zagreb, Zagreb, Croatia sinisa@zemris.fer.hr In this presentation, a family of analytical models for predicting the performance of parallel applications under various cache coherence protocol assumptions will be presented. These models can be used to determine which protocols are to be used in order to improve the performance of the applications, and also to dynamically determine when to change protocols in the case of hybrid protocols. The models apply equally well to tightly- coupled multiprocessor systems and to loosely-coupled distributed systems (such as networks of workstations or Internet based systems). Our models are unique in that they lie between a large body of theoretical models that assume independence and an uniform distribution of memory accesses across processors (or nodes of distributed system), and a large body of address-trace oriented models that assume the availability of a precise characterization of interleaving behavior of memory accesses. The former are not very realistic, and the latter are not suitable for compile-time and run- time usage. In contrast, our models use a set of parameters that characterize the access behavior of applications well, and can be obtained with advanced compiler technology. The accuracy of our models is assessed on parallel applications from SPLASH and SPLASH-2 benchmark suites. The results obtained by the models guide many decisions in designing cache coherence protocol for NUMAchine multiprocessor (University of Toronto, Canada). As part of this study, we also show the potential advantage of using dynamic hybrid protocols. Upcoming seminars: May 8: Ganesh Gopalakrishnan, Univ. of Utah May 10 (Friday 11AM): Dhiraj Pradhan, Texas A & M Univ.