A Self-Adaptive Database Buffer Replacement Scheme.

The overall performance of a database system is very sensitive to the buffer replacement algorithm used. However, the performance evaluation of database buffer replacement algorithms commonly assumes that database accesses are independent and the probability for each individual database record to be accessed is fixed. Due to these rigid assumptions, the results of performance evaluation are not always reliable. In this dissertation, we apply Simon\u27s model of information accessing to model database accessing frequencies. This approach relaxes the independent assumption, and since it also allows certain dynamic behavior in accessing frequencies; thus, it is more robust and preferable over the traditional artificial data approach. Furthermore, taking advantage of the conceptual similarity between the self-organizing linear search heuristics and the traditional buffer replacement algorithms, we propose a self-adaptive buffer replacement scheme that outperforms conventional database buffer replacement algorithms. The findings of our study can be further applied to many other computer applications, e.g. the more complex problem of archival storage design in larger database systems