Based On Accessing Frequency Dynamic Materialization Views In Data Warehouses

In data warehouse environment, OLAP (On-Line Analytical Processing) and OLDM (On-Line Analytical Mining) usually use concise results, which are computed from huge base data, to answer queries committed by users. Because of this point, Materialized views are of unprecedented importance in data warehouses. In data warehouses, materialized views are used to pre-compute and store aggregated data such as sums and averages, from which answers to queries are made through simple search or simple computation, consequently materialized views can remarkably speedup the response to queries committed. However, materialized views need space to store and need being maintained. As a result materializing all views is unpractical. There must be some strategy to select out what should be materialized, which is called MVS (Materialized Views Selection).There have been a lot of studies which focus on MVS, whereas they can not meet the stability and the robust character, which count a great deal in the practice of industry. The current data warehouse products can not soundly support materialized views auto-selection: What they adopt are static approaches which greatly disobey dynamic nature of OLAP and DSS. However, the future of data warehouse demands MVS more on its efficiency, tractability, validity and adaptable.This dissertation puts forward a dynamic materialization views approach which is based on views accessing frequencies. Compared to traditional static MVS approach, the dynamic approach described here is more adaptable, more efficient and more tractable. It is able to solve large-scale MVS problem.The approach materialize views in different phases and in different way according to their different accessing requirements, which reduces the complexity of a whole MVS problem, meanwhile, the effectivity of materialized views of a givenstore space is improved. An improved greedy algorithm, which is polynomial time, is run to select initial materialized views automatically. The range of this sub-problem can be controlled to acceptable running time. In the approach, the frequencies of views being accessing are taken into account as an important factor, which reflects query trend of users. Based on the views accessing frequency, a benefit model is constructed, through which the adjusting standard is figured out. According to the standard, materialized views set is adjusted to suit the changed query trend. Materialized views are made a difference between permanent ones and temporary ones, only the latter can be adjusted during query processing, which avoids the materialized views with steady users accessing frequencies are deleted. PWL (Prior Warning Line) is introduced to mark the status of occupied store space by materialized views. If the occupied space reaches PWL, measures are taken to prepare for coming adjusting, which can improve the efficiency in query processing. The dissertation produces a primary algorithm, which dynamically materialize and adjust views. In Oracle data warehouse environment, the algorithm is tested using 1GB data which generate by TPC-H benchmark. The experiments show that dynamically materializing and adjusting algorithm outperforms static view selection.This approach gets the adaptability, high efficiency and tractability, but if being integrated into data warehouse, further work should be done on cooperation work with other parts.