A Unified Implementation of Concurrency Control and Recovery

Transaction management in shared databases is generally viewed as a combinationof two problems, concurrency control and recovery. These two problems areusually considered as orthogonal and, consequently, the correctness criteriaderived from them are incomparable. Recently, a unified theory for concurrencycontrol and recovery has been proposed along with a unified correctnesscriteria called prefix-reducibility. Developing efficient protocols foraccepting prefix-reducible executions was left as an open problem. In thispaper, we extend this theory to facilitate the design of efficient dynamicprotocols that accept prefix-reducible executions. Furthermore, we develop aunified implementation of the log and lock managers required to support thisnew class of schedulers.