EBRES: uma estratégia inteligente de substituição de páginas para banco de dados

Abstract

Database management systems (DBMS) do not always guarantee that all data is available in the main memory. DBMSs perform a replacement algorithm to determine which data to replace between main memory and secondary storage media. This class of algorithms is constantly evolving and increasingly considers new aspects in replacement decisions, such as modern flash storage media. Most probably, we may be starting a new era of these algorithms with extensive advances in forecasting techniques. This article proposes the EBRES (Efficient Buffer Replacement with Exponential Smoothing) algorithm that considers several aspects in the decisions with the support of a simple low-overhead forecasting model to predict data accesses. During the experiments, EBRES maintained a balance in its performance in different scenarios tested and was able to protect frequent data during sequential operations.