关键词: 数据预取, 蚁群规则, 惯性规则

Abstract: In this report, we explore a new approach toward intelligent caching and prefetching for data query of DBMS. First we abstract the data query statement into query patterns which consists of four units. We also consider the real query parameters which can be used to build real query from the query pattern. Base on the query pattern and the real query parameters, we develope two intelligent prefetch algorithms to fit two kinds of demand in data query. The first algorithm bases on ant-group rule. It can be used to predict the future query with highest probability. Our experiments show that in contrast to the substantially large number of queriescoming of the special application to the database system, the number of patterns of these differentqueries are quite limited. We take into consideration the query pattern and the historytrace of query reference when predicting future query and develop the second algorithm base on inertia rule which uses BP network to trace the history of user query. It is more fit for the multi-application situation than the previous. Our simulation shows the inter-query locality is highly query pattern dependable under single-application situation and the inertia rule has more flexibility under multi-application situation.

Key words: prefetch, ant-group rule, inertia rule