ANALYSIS OF MEMORY INTERFERENCE IN BUFFERED MULTI-PROCESSOR SYSTEMS IN PRESENCE OF HOT SPOTS AND FAVORITE MEMORIES

In this thesis, a discrete Markov chain model for analyzing memory interference in multiprocessors, is presented. Each memory module is either hot or favorite or neither of these two. Since the state space of such a Markov model for an N x M system, where N and M are respectively the number of processors and memory modules, becomes exorbitant for large N and M, we restrict our analytical solutions to 2 x M and N x 2 systems. The general case is analyzed using simulation. In all cases, the effective memory bandwidth, mean memory-queue-length and mean-waiting-time for a memory request are derived. A heuristic is presented, using a probabilistic model, which finds the number of hot modules beyond which there is hardly any bandwidth change. ACKNOWLEDGEMENTS I am immensely grateful to my advisor, Dr. Sajal K. Das, without whose unceasing support, encouragement and guidance this thesis could not have been written. The several months of fruitful discussions I had with him, helped me get a better under-standing of the subject and also imparted the spirit of perseverance. My association with him had been a very rewarding experience which I will always treasure. I am grateful to the members of my supervising committee, Dr. Weiping Shi and Dr. Farhad Sharokhi for their review and helpful suggestions. I am indebted to th