AN OPTIMISTIC APPROACH TO CONCURRENCY CONTROL IN DISTRIBUTED DATABASES

Abstract

In a distributed database management system, several users can concurrently access and modify the same data objects, which may result in a conflict. A concurrency control algorithm is required to control interleaving of conflicting transactions to maintain the database consistency and correctness. Locking and timestamping are two commonly used concurrency control methods. But both of them are suitable for environment, where conflicts occur very often, because they try to resolve conflicts even before starting of the execution of the transaction. In cases, where conflicts are rare, another class of algorithms, popularly known as optimistic approach, is found to be more suitable, because it allows the transactions to run to completion on a local copy. The conflicts are resolved only before transferring the results to the actual database. An optimistic voting algorithm has been implemented in this dissertation work. It incorporates optimistic approach in conventional voting (timestamping) algorithm to improve the degree of concurrency, and hence throughput. Since transactions are executed on a local copy and only the results are transferred to the actual database, transaction can easily be aborted by discarding the local copy in case of site-failure. Using this feature the proposed algorithm also incorporates the crash recovery scheme.