Research On Query Language And Query Processing In DM_OLAP

Online Analysis Processing (OLAP) is the main technique of processing data in decision support system. In order to resolve problems of expressing and processing queries in DM_OLAP system, it is necessary to do research on query languages and the methods of query optimization in the environment of DM_OLAP. Based on these techniques, DM_OLAP can be a practical tool to be used in decision support system.After comparing Multi-Dimensional eXpression (MDX) with extensions to SQL, it is determined to design a kind of MDX language in DM_OLAP system, which is named DM_MDX. The fundamental grammar of DM_MDX is derived from the grammar of the MDX language, which has been implemented in SQL Server 2000 Analysis Services. To be suitable to multidimensional model and analysis functions in DM_OLAP, several units of the MDX grammar are modified, such as axis dimension, slice dimension, member, data source, feature property expression, aggregation function, etc. Finally, the grammar of DM_MDX is designed and implemented in normal form. With the grammar of DM_MDX, the DM_MDX complier is able to transform a DM_MDX query to a set of relational algebraic expressions. The complier is composed of syntactic analyzer and semantic transformer. The syntactic analyzer makes lexical analysis and syntactic analysis on the query, and sets up syntactic tree from the structure of the query. The semantic transformer makes semantic checks on syntactic tree and transforms the tree to a set of relational algebraic expressions, which are represented by the CPreQuery structure.As to the problem of query optimization, several methods including push-down, early grouping and hierarchical pre-grouping, which are used to optimize star queries, have been analyzed. In the kernel system of DM_OLAP, star queries are proceeding on two kinds of fact tables of star model, which are the base table and the cube. Two types of optimization plans on the base table are applied, including the plan of push-down and the plan of hierarchical pre-grouping. According to the features of the completed computed cube and the index structure in the kernel system, two types of plans on the cube are put forward by modifying the plans on the base table. Finally, a star query optimizer has been designed. The optimizer produces one of the best executive plans from a set of relational algebraic expressions submitted by the DM_MDX complier, and is composed of plan space, plan generator and cost estimator. This optimizer provides algebraic plans and executive plans derived from four types of optimization plans on the base table or the cube, and makes use of cost estimation functions to choose the best executive plan of minimum cost.