Research And Design On Self-Maintainable Data Warehouse

A data warehouse stores materialized view over data from one or more sources in order to fast access to integrated data. These views need to be maintained in response to updates in the source data. This is often done using incremental techniques and re-computation from the scratch that must access data from underlying source. But in the data warehouse scenario,accessing source can be difficult,low efficient and expensive,sometimes data may be unavailable,since the source are distributed or heterogeneous. For these reasons,the problem of materialized views self-maintenance has received increasing attention. However,not all the views defined by relation and group operations are self-maintainable,we must use auxiliary relations in data warehouse in order to self-maintain views.In this thesis,first 1 research on the self-maintainability of simple views that only use one operation. Then I propose an efficient self-maintainable algorithm based on analysis of binary express of views. The algorithm includes three steps:1. deciding auxiliary relations in an up-down fashion according to simple views self-maintainability. 2. computing each node updates in a bottom-up fashion. 3. updating the materialized views and auxiliary relations.In the 4th chapter,I study some key questions involving bag scenario,distributed and parallel computing issues,data backup and restore and security control. In particular,for data restore,I use a different method with traditional log file one,which can decrease auxiliary space and improve maintenance efficiency on a large scale. In the last chapter,I implement a mini-type system using the self-maintainable algorithm. At the end of the thesis,I give the summary and point out further direction of research on view self-maintenance.