Research Of Summary Queries Based On Database

Recently, database has been applied widely in many fields, and more requirements with variety and effective are proposed for data query processing. Several aggregate methods, such as MAX, MIN, SUM, COUNT, are easy to be obtained in the data query. But the precise results of the queries are not enough to fully express the joint or spatial distribution of data. Data analysts often require more insight into the data distribution. In many applications including financial analysis, collection of real-time signal in sensor network, traffic monitoring and management, Web access log analysis, busisness trading etc.Because of its infmiteness and real-time of data arrived, all of original data can not be stored directly for user, consequently, precise query results can not be obtained. In fact, in many of the decision analysis, query processing does not require precise results in the data query strictly, most of them often require to approximate query processing based on those summaries of ocuuping less memory, such as quantile, sketch, wavelet etc, it is called the Summary query, and then helping decision makers make decisions quickly. Consequently, query processing and analysis based on the summaries brings unique opportunities but also great challenge, it has the higher value in the research of database application. This paper just works on the research about query processing and analysis technology based on quantile.This paper study that how to approximate query based on quantile, and bring an in-depth research about multi-attribute queries probolem of Summary query, and project the algorithm for multi-value objects of KNN (K Nearest Neighbor) based on quantile. The experiment shows that it is efficient and effective. The works of this paper are as the following:(1) Considering the problem of one query attribute of Summary query. This paper proposes algorithms for KNN based on quantile, which are called Q-KNN and GQ-KNN, and solve the problem of multi-attribute Summary query.(2) Some Experiments are conducted based on both real and synthetic datasets, which shows Q-KNN and GQ-KNN techniques perform quite efficiently and scalable.