Modeling Optimal Effective Bandwidth of Page-Mode Memory for Stream-Oriented Computations

AbstractMemory system performance models have traditionally assumed that individual modules are insensitive to the sequence cf access requests. For current memory components, this assumption is not correct. Furthermore, performance models generally assume a stochastic sequence of reference. For stream-oriented computations, such models of reference are not suitable. This paper derives the optimal effective memory bandwidth achieved by stream-oriented computations at a single module of such memory components