中文摘要：

我们的学习在处理引擎(DQPE ) 的分布式的质问的提高的体系结构介绍一个新奇分布式的询问计划精炼阶段。质问计划精炼由可重用的混合式查询发货产生潜在地有效的分布式的质问计划(RAQS ) 来临。途径以预处理时间的成本改进响应时间。如果开销不能被质问结果用法补偿， RAQS 是不再有利的。因此，一个全球成本估算模特儿被雇用得到合适的操作员：RR_Agg， R_Agg，或 R_Scan。为在分布式的质问处理与聚合函数再使用询问的结果的目的，缓冲的一个多水平混血儿看法(HVC ) 计划被介绍。计划保留缓冲的部分火柴和混合式查询结果的优点。由我们的解决方案，有分布式的 TPC-H 询问的评估平均显示出重要改进响应时间。这篇文章的联机版本(做 i：10.1007/s11390-008-9190-3 ) 包含增补材料，它对授权用户可得到。

英文摘要：

Our study introduces a novel distributed query plan refinement phase in an enhanced architecture of distributed query processing engine （DQPE）. Query plan refinement generates potentially efficient distributed query plan by reusable aggregate query shipping （RAQS） approach. The approach improves response time at the cost of pre-processing time. If the overheads could not be compensated by query results reusage, RAQS is no more favorable. Therefore a globM cost estimation model is employed to get proper operators： RR_Agg, R_Agg, or R_Scan. For the purpose of reusing results of queries with aggregate function in distributed query processing, a multi-level hybrid view caching （HVC） scheme is introduced. The scheme retains the advantages of partial match and aggregate query results caching. By our solution, evaluations with distributed TPC-H queries show significant improvement on average response time.