基于负载平衡的并行JOIN关系划分粒度研究

Based on the analysis of the task sizes and the load, this paper discussps the granularity ofrelation spliting in the spliting phase taking acount of task load being less than average load,andprobes into the relationship between the granularity and load balancing. The minimum number ofbuckets is determined on the basis of relation spliting granularity,and the maximum number ofproduct tuples of each node is induced under the prerequisite for ensuring load balance in paralleljoin.

解决并行JOIN中数据偏斜的一种方法

1.引言 最近的研究表明,数据库系统也可以利用并行处理来提高更大、更复杂查询的效,因为关系操作是面向集合的操作,本身就具有一定的可并行特性,这为查询优化在选择并行取路径时提供了很大的灵活性。