Program behavior prediction and prepaging

摘要: The design and evaluation of a prepaging technique is presented. It consists in the preloading, at instants which page-faults occur, those pages whose use future most likely. These are determined as function an a-priori knowledge program behavior. We shall first present two tools for predicting that based on Markov memory reference transition model, second semi-Markov model page-fault transitions. Both methods successively experimented by paging algorithm allowing prepaging. Experimentation shows can significantly improve performance algorithm, especially when prediction transitions between page-faults. sensitivity to various working conditions tested. Implementation examined shown be feasible without special hardware devices. study completed influence global system performance. Some common cases architecture experimentation show non-negligible increase CPU utilization.