A Program-and-User Based File Access Prediction Model

Abstract-In the research on external storage file access prediction algorithms, how to improve the predictive hit ratio and the degree of applicability becomes the major issue. This paper discusses several current prediction models including Program-based Last Successor (PLS) model, User-based Last Successor (ULS) model and Program-and User-based Last Successor (PULS) model, which both use program or user information to improve the predictive hit ratio, but all three models need a lot of file access history information which requires a long time to accumulate and have poor applicability. This paper presents a Program-and User-based Buffer Window (PUBW) model, which not only uses program and user information to improve the prediction hit ratio, but also uses the buffering mechanisms to improve its applicability. Our simulation results show that PUBW model achieves higher predictive hit ratio compared with the PLS and ULS model and has almost the same high degree of applicability as the Last Successor (LS) model. Our experiments show that the PUBW model is a useful and efficient file access prediction model.