The join operation is one of time consuming operations in relational database systems. With rapidly decreasing cost of main memory, the hash based join technique as an alternative to traditional ones, namely nested loop and sortmerge joins has also been proposed in the past. In this thesis, we propose a new parallel hash join method that is robust in severe data skew and also works well when the amount of available memory changes dynamically. This work differs from other hash based join methods in that those methods assumes either static amount of main memory or uniform input data. We describe the various problems of the previous ones in the multiuser and multiprocessor environment, and present how to overcome these problems. We also show through performance comparison that the proposed method works better than others for various degrees of data skew and various amounts of main memory.
