Dynamically Adaptive Prepaging for Effective Virtual Memory Management

Demand prepaging, an extension to the widely employed method of demand paging, is a concept designed to reduce page faults in a system implementing virtual memory by prefetching pages speculated to be referenced in the near future in addition to pages that have already been referenced. Doing so exploits high disk bandwidths while attempting to avoid high disk latencies. Although studies have shown that demand prepaging is generally not beneficial, other more recent studies have been able to demonstrate significant page fault reductions using dynamic prepage parameters as opposed to static [4]. There are several major parameters associated with demand prepaging however, and to date there are no studies which evaluate the performance of a completely dynamic set of these parameters. In this paper, I propose and evaluate a Dynamically Adaptive Prepaging (DAP) scheme in which all major demand prepaging parameters are dynamically modulated to changes in reference stream trends and phases. My proposed DAP policies attempt to reduce page faults by exploiting high disk bandwidths. I evaluate my proposed system through simulation, using a custom trace-driven simulator developed in C++. Various uni-programmed memory reference traces were simulated. My evaluation shows a potentially significant reduction in page faults for the simulated traces