一种并行XML数据库分片策略

A Data Placement Strategy for Parallel XML Databases

主要研究XML文档的并行数据分片策略,以便能够并行处理XML查询.为了描述XML数据分片,提出了媒介节点的概念.一组媒介节点的集合可以将一棵XML数据树分割成一棵根树和一组子树的集合:根树将在所有站点中复制;而子树集合则可以根据用户查询的工作负载被均匀地分片到各个站点中.对于同一棵XML数据树,会有很多种媒介节点的集合;而不同的媒介节点集合会产生不同的数据分片结果.然后,依据各个数据分片中的用户查询工作量是否均衡,来衡量一个分片的好坏.选择一组最佳的媒介节点集合是一个NP-hard问题.为了解决此问题,设计了一组启发式优化规则.基于这一思想,提出并实现了一种基于媒介节点的XML数据分片算法WIN(workload-awareintermediarynodesdataplacementstrategy).大量实验结果证明:WIN算法的性能要优于以往的并行XML数据分片策略.

This paper targets on parallel XML document partitioning strategies to process XML queries in parallel To describe the problem of XML data partitioning, a concept, intermediary node, is presented in this paper. By a set of intermediary nodes, an XML data tree can be partitioned into a root-tree and a set of sub-trees. While the root-tree is duplicated over all the nodes, the set of the sub-trees can be evenly partitioned over all the nodes based on the workload of user queries. For the same XML data tree, there are a number of intermediary nodes sets, and different intermediary nodes sets will generate different partitions. It can be evaluated if a partitioning is good based on the workload of user queries. It is obviously an NP hard problem to choose an optimal partitioning. To solve this problem, this paper proposes a set of heuristic rules. Based on the idea described above, this paper designs and implements an XML data partitioning algorithm, WIN, and the extensive experimental results show that its speedup and scaleup performances outperform the existing strategies.