1 IntroductionrnIn distributed systems, the performance of join query (mainly about one dimension join query) is one of the most necessary to be improved [1].Unlike traditional centralized database systems storing data locally, under distributed environment, data is partitioned into multiple blocks stored at separate physical nodes.For improving the performance of join query,sometimes, we need to maximize local execution of query plan to mitigate expensive network cost.Facing various data layouts and various combinations of operators, it is rarely possible to find local executing places for the whole plan.So it is usually required to split the complete plan into multiple fragments and search the local executing places per fragment.Current strategies generate local executing places by re-partitioning data on the fly [2, 3] or migrating data according to query history [4-6], but facing the problems of large negative effect to system performance when migrating data and not sharing the results of common fragments.