一個以目標關係為基礎的分散聯結運算方式

Abstract

A target relation of a query is a relation which contains attributes of selected tuples to be outputted. In this paper, by identifying target relations, we divide a given tree query into two parts: one final query tree and zero or more non-final query trees. Since only the root of each of the non-final query trees will participate in the final query tree, we can apply a semi join program to fully reduce the size of the root of each of the non-final trees first. Therefore, we can reduce the data transmission cost for the final query tree. Moreover, when there is more than one non-final query trees, we can process them in parallel, which can shorten the query response time. Then, we apply join and semi join operations together to optimize the cost of the final query tree, Consequently, our target-relation-based approach not only can reduce the data transmission cost but also the response time. Moreover, the larger the number of non-final query trees is, the more reduction our approach can achieve. We show that the proposed approach to distributed joins can have better performance than other approaches which either apply semi joins before the join process or apply both joins and semi joins together.