Research On The Storage And Aggregation Optimizing Methods Of Multidimensional Data

The traditional databases are the main information sources of data warehouses; data warehouses provide an integrated data environment for Online Analytical Processing (OLAP), Decision Support System (DSS) and Data Mining (DM). Organizing and managing the data efficiently is one of the keys of implementing data warehouses. This thesis studies it deeply on the aspects of data warehouses' concept model and OLAP implementation.Multidimensional data model is a basic aspect in the research field of data warehouses. After summarizing and analysing the existing data models, a data warehouses' multidimensional data model is proposed using classification method and the corelative concepts of data cube are defined in this thesis, the model is powelful enough for modeling complex data structure of data warehouse, which establishes theoretical foundation for the storage and aggregation optimizing methods of multidimensional data.The logic organization mode of multidimensional data is one of the keys of OLAP implementation, this thesis summarizes the two organizing ways of multidimensional data - relational mode and array mode thoroughly, and places emphases on the researches of array mode, including the storage structure of multidimensional data, the construction methods of multidimensional arrays, the compressing methods of sparse arrays, the principles of dividing arrays into chunks and the access methods of chunk arrays, and also this thesis realizes a storage instance of array mode based on the above theoretical analyses.One means of improving the performance of OLAP is to compute multidimensional aggregations efficiently. This thesis summarizes the main optimizing methods of computing aggregations, on which the corelative concepts are formally defined, furthermore, this thesis emphasizes the research of optimizing methods of array mode and proposes an aggregation algorithm - PartCube algorithm, it makes use of optimizing methods including Small-parent, amortize-scans and Cache-results, and also it establishes the minimum spanning tree based on search lattice. If memory is insufficient, PartCube divides array into parts and computes each separately,after all parts have been accomplished, PartCube merges the intermediate results into integrated aggregations. The analysis shows that this algorithm can make the best use of memory and reduce I/O times, so it has high efficiency in computing aggregations.At the end of this thesis, the researches are summarized and the future work is presented.