Abstract: Hive, the database on Hadoop, enables more users to process relation data by providing sqllike interface. However, Hive does not provide an efficient approach for join, a common but expensive operator in Hadoop. In order to solve the performance of join, this paper proposes a novel hybrid algorithm, HJA, which can help to automatically choose the relatively better one among several methods, according to the current situation. Experiments results show that HJA can get best performance in most situations.

Key words: MapReduce, Hadoop, partition join, autotuning, Hive