A dynamic approach for handling data skew problems in parallel hash join computation

Data skew can result in load imbalance in parallel hash join. We introduce a new term, solid data skew, and classify data skew into plane and solid categories in this paper. The existing algorithms consider plane data skew only thus not adequate to approach solid data skew problem. A dynamic approach for handling solid data skew problem is proposed. It assigns join subtasks to processors during runtime using a method based on the estimation of execution time required for the computation.<>