Application of Filters to Multiway Joins in MapReduce

Joining multiple datasets in MapReduce may amplify the disk and network overheads because intermediate join results have to be written to the underlying distributed file system, or map output records have to be replicated multiple times. This paper proposes a method for applying filters based on the processing order of input datasets, which is appropriate for the two types of multiway joins: common attribute joins and distinct attribute joins. The number of redundant records filtered depends on the processing order. In common attribute joins, the input records do not need to be replicated, so a set of filters is created, which are applied in turn. In distinct attribute joins, the input records have to be replicated, so multiple sets of filters need to be created, which depend on the number of join attributes. The experimental results showed that our approach outperformed a cascade of two-way joins and basic multiway joins in cases where small portions of input datasets were joined.