NDSMMV——一种多维数据集物化视图动态选择新策略

NDSMMV——A New Dynamic Selection Strategy of Materialized Views for Multi-Dimensional Data

物化视图的选择策略是数据仓库研究的重要问题之一.通过深入研究提出了一种多维数据集中物化视图动态选择的新策略——NDSMMV,包括候选视图生成算法CVGA、物化视图选择算法IGA、物化视图调整算法MAMV和物化视图动态调整算法DMAMV.CVGA基于多维数据格生成候选视图集,对候选视图数量进行压缩以减少后续算法的视图空间搜索代价和时间复杂度;IGA基于视图查询、视图维护和存储空间三元评价标准在候选视图集上进行物化视图的选择;MAMV基于物化视图选择过程已选视图的收益变化情况对物化视图进行进一步调整以提高查询的响应性能;DMAMV定时地判断查询视图类型分布是否变化来决定是否进行物化视图的动态调整,从而避免了物化视图集的"抖动".理论分析和实验结果表明该策略是有效可行的.

The selection strategy of materialized view is one of the important issues of data warehouse research. Its goal is to elect a group of materialized views, which could cut down the cost of the query greatly on the basis of the limited storage space. The cost model is proposed at first. Then, a new dynamic selection strategy of materialized views for multi-dimensional data （NDSMMV） is presented, which is composed of four algorithms. CVGA （candidate view generation algorithm）, IGA （improved greedy algorithm）, MAMV （modulation algorithm of materialized views） and DMAMV （dynamic modulation algorithm of materialized views）. CVGA generates the candidate view set based on multi- dimensional data lattice, which reduces the number of candidate views to decrease the space search cost and time consumption of the following algorithm. IGA selects materialized views taking account of view query, view maintenance and space constraint. MAMV modulate the materialized views according to the change of the materialized view profit, which improves the capability of querying materialized views. DMAMV uses the sample space to judge whether it is necessary to change the view set which can avoid sharp dither. The comparative experiment indicates that NDSMMV operates more effectively than BPUS and FPUS in the respect that CVGA reduces the amount of views beforehand. IGA selects the materialized views quickly, MAMV modulates the materialized views accurately, and the query expense decreases further with the modulation of the DMAMV on line, which validates the efficiency of NDSMMV.