Data Placement and Query Processing Based on RPE Parallelisms

The basic idea behind parallel database systems is to perform operations in parallel to reduce the response time and improve the system throughput. Data placement is a key factor on the performance of parallel database systems. This paper proposes two data partition strategies to decluster XML documents with very large size, Path Schema based Path Instance Balancing (PSPIB) strategy, in which all path instances with the same path schema in a data tree are declustered evenly over all sites, and Node Schema based Node Round-Robin (NSNRR) strategy, in which all node objects with the same node schema in a data tree are declustered over all sites in a round-robin way. Accordingly, two query processing algorithms are proposed based on the two partition methods, Parallel Path Merge (PPM) algorithm and Parallel Pipelining Path Join (PPPJ) algorithm. The performance analysis and evaluation on the two data placement strategies and corresponding query processing algorithms are given in this paper.