Design And Implementation Of A Semantic Data Vertical Fragmentation System

To improve the performance of database,fragmenting the data is becoming a common demand.Before taking the fragmentation strategy,the principal of fragmentation should be considered,which includes reconfigurability,disjoint and integrity.There are three kinds of fragmentation strategies,including horizontal,vertical and mixed fragmentation.Each strategy of these three works is suitable for different scenarios.The main topic in this thesis is vertical fragmentation.Vertical fragmentation is usually used to classify the hot spot data,which can reduce data access time cost and improve the performance.However,these algorithms for vertical fragmentation are always based on the attributes,which ignores the data semantics.The fragments could be better if the data semantics are considered.Therefore,a new system about vertical fragmentation based on data semantics is proposed and implemented here.The main goal of clustering is to find the relationship among datasets.Based on the study of common clustering algorithms,including partitioning methods,hierarchical methods,density-based methods and etc.,the strengths of each method are obvious.After the research of these classical algorithms,K-Prototypes is chosen to handle multivariate data types since there is no doubt that both K-Means and K-modes are not suitable for this data type.Through the comparison of different distance measurement methods,the Jaccard similarity coefficient is used to merge the similar clusters,so that the number of clusters equals the nodes of database.As the new algorithm was implemented in the experiment,the fragments were produced.The experiment between the new and normal algorithm was conducted,which proved the superiority of the new algorithm.Based on the new algorithm,the data vertical fragmentation system was designed and implemented,which could not only produce vertical fragments but also process original data and provide important charts.