Effective Allocation of Database Buffers by Mining Users' Access Histories

The problem of database buffer management has extensively been studied for nearly three decades. In this paper, we explore the use of newly emerging data mining technology to tackle the traditional buffer management issue. In particular, we address the buffer size setting problem for distributed database systems. The main goal is to minimize physical I/O while achieving better buffer utilization at the same time. Different from the traditional buffer management strategies where limited knowledge of user access patterns is analyzed and used, our buffer allocation mechanism extracts knowledge from historical reference streams, and then determines the optimal buffer space based on the discovered knowledge. Simulation experiments show that the proposed method can achieve an optimal buffer allocation solution for distributed database systems.