基于散列技术的并行流水线Join算法的设计与评价

Design and Evaluation of Hash-Based Parallel pipelined Join Algorithm

本文研究了一种新型的混合式通用并行结构上基于散列技术的并行连接算法的实现技术，并建立了一种性能分析模型。这种模型特别刻划了系统中三种主要资源ＣＰＵ、磁盘、互联网之间的操作重叠。我们以这一模型为基础深入分析了并行Ｈｙｂｒｉｄ散列连接算法（称为ＰＨＨＪ）和它的一种基于小关系复制技术的变形算法（称为ＰＨＨＪＲ）的性能及与系统结构之间的关系。我们还结合传统的位向量过滤技术提出一种用对称位向最和内外关系角色转换的技术，进一步提高并行连接算法的性能。

The implementation techniques for hash-based parallel join algorithm against a novel generic parallel architecture is studied and a performance evaluation model is established,which models the overlap of operations of the three major system resources(CPU,disk and interconnection network).In light of this model,the performance of the parallel hybrid hash join algorithm(called PHHJR)and its variation which replicates the smaller relation(called PHHJR) are intensively analysed.Based on symmmetry bit vector filter concept,a new technique for accelerating join operations is proposed and compared with the traditional bit vector filter.The analysis shows that the symmetry bit vector filter can outperform traditional bit vector filter.