A new method for handling data skews in relational joins on graphics processing units

We present a new method for handling data skews in the database operation, relation joins. Recently, Graphics Processing Units (GPU) has become popular among high performance computing research and is used to accelerate the execution of database operations. Various parallel algorithms are developed in the past years with emerging of multi -core processors. Coordinate Module Distribution (CMD) is originally designed for clusters, we have immigrated it onto GPU to improve the performance of relational joins. We find that using one GPU card, our method runs between hash join and sort-merge join, with hash join having the best performance. But our method runs close to hash joins and faster than sort-merge joins when processing relations with data skew.