Transaction optimization techniques

Replication introduces a tension between query optimization and remote access control in a distributed database system. If we view a transaction as a partially-ordered set of queries and updates, then factors that affect quorum selection for the fragments accessed by a transaction as a whole are currently orthogonal to factors that affect the replica selection during the planning of individual queries. Therefore, the two processes may act at cross-purposes to one another. Query optimization considers an individual query and selects a set of fragments that minimizes the computation and communication cost and allows computation to be pushed into the local site. Transaction management, on the other hand, selects quorums (sets of replicas to retrieve) based on replica availability and on mutual consistency constraints such as quorum intersection among write operations or between read and write operations. Thus, transaction optimization narrows the “optimal” solution space for the queries it contains Hence, transaction management should cooperate with query optimization to optimize transaction processing.

In this book chapter, we discuss why and how to optimize transactions. We present a novel transaction optimization strategy that integrates query optimization techniques with transaction management. The proposed strategy chooses quorums of maximum intersection, while minimizing a communication and/or computation cost function. It also attempts to maximize the number of up-to-date copies of read quorums, so as to maximize the optimization space of the individual queries.