An Evaluation of Multi-way Joins for Relational Database Systems

In database systems most join algorithms operate on only two inputs at a time. Research into joins on more than two inputs, called multi-way joins, has shown that the intermediate partitioning steps of a traditional hash join based query plan can be avoided. This decreases the amount of disk based input and output (I/Os) that the query requires. This work studies the advantages and disadvantages of implementing and using different multi-way join algorithms and their relative performance compared to traditional hash joins. Specifically, this work compares hash join with three multi-way join algorithms: hash teams, generalized hash teams and SHARP. The results of the experiments show that in some cases multi-way hash joins can provide a significant advantage over hash join but not always. The cases where hash teams and generalized hash teams have better performance is limited, and it does not make sense to implement these algorithms in a production database management system. SHARP provides enough of a performance advantage that it makes sense to implement it in a database system used for data warehousing.