Data Stream Online Analytical Processing Technology

In recent years, data streams are becoming an increasingly important class of data type. Stream of data items differ from records stored in traditional DBMS in many ways, Therefore, several aspects of data processing and managements need to be reconsidered in the presence of data streams, offering a new research direction. Moreover, with years of research and development of OLAP technology, OLAP plays a more and more important role. The existing technology of OLAP is merely based on the finite, static, persistent, historical data. However , on-line analytical processing on data streams is based on the infinite, continuous, rapid changing stream items. Then the study of the technology of on-line analytical processing based on data streams is a challenging task.This paper mainly studies the technologies of on-line analytical processing of data streams. First, this paper investigates the controlling mehod of the size and updating technology of the sliding windows over data streams. They adapt well to the situation that data streams to be processed arrive at the multiple rate. This paper puts forward a novel technique to aggregate a data streams over sliding windows. The technique could effectively implement approximate query of detail data streams. Second, this paper investigates methods for regression analysis of time-series data streams. This research includes several following aspects: (1) A technology of building linear regression model of time-series data streams and the applications of the linear regression model. (2) A technology for aggregating the multiple linear regression equation of time-series data streams. Our experiments show the methods can effectively come into being regression model of time-series data streams, and rapidly fulfill the regression analysisof such data. Third, this paper proposes the time dimension with multiple levels of granularity. This paper defines a critical sub-model, and sets forth an apriority-based drilling approach. Based on the results, this paper presents a partially materialized multi-dimensional data model of data streams, which is based on the critical sub-models. Its performance shows the model can effectively make use of costly storage resource. In the end, based on the above algorithms, the paper proposes a general design of an online analytical processing based on data streams, and gives experiment results.