A self-tuning page cleaner for DB2

Abstract

The buffer pool in a DBMS is used to cache the disk pages of the database. Because typical database workloads are I/O-bound, the effectiveness of the buffer pool management algorithm is a crucial factor in the performance of the DBMS. In IBM's DB2 buffer pool, the page cleaning algorithm is used to write changed pages to disks before they are selected for replacement. We conducted a detailed study of page cleaning in DB2 version 7.1.0 for Windows by both trace-driven simulation and measurements. Our results show that system throughput can be increased by 19% when the page cleaning algorithm is carefully tuned. In practice, however the manual tuning of this algorithm is difficult. A self-tuning algorithm for page cleaning is proposed posed in this paper to automate this tuning task. Simulation results show that the self-tuning algorithm can achieve performance comparable to the best manually tuned system.