Minimizing Data-Communication Costs by Decomposing Query Results in Client-Server Environments

Many database applications adopt a client-server architecture, in which data resides on a server that receives queries from a client. For each client’s query, the server often needs to transfer to the client a large amount of data that is an answer to the query. The communication network in these environments could become a bottleneck in the computation. In this paper we study how to minimize the communication costs of transferring answers to largejoin queries from server to client. We propose a novel technique that decomposes the answer into intermediate results, or views, which can reduce the redundancy in the answer. These views are transferred to the client and are used by the client to compute the final answer. There are several challenges in implementing this technique: (1) the number of possible plans to decompose the answers could be very large; (2) the technique requires an efficient algorithm to give an accurate estimate of the size of each view; and (3) many factors could affect the decomposition choice; one such factor is whether relevant data is cached on the client. Our extensive experiments on queries adapted from the TPC-H benchmark show that our technique can significantly reduce the communication costs of transferring answers to large-join queries. The extra steps used in our approach do pay off to reduce the total time of transferring the result of a query, when the result has a lot of redundancy.