SUPPORTING MULTIPLE ISOLATION LEVELS IN REPLICATED ENVIRONMENTS.

Resumen

Replication is used by databases to implement reliability and provide scalability. However, achieving transparent replication is not an easy task. A replicated database is transparent if it can seamlessly replace a standard stand-alone database without requiring any changes to the components of the system. Database replication transparency can be achieved if: (a) replication protocols remain hidden for all other components of the system; and (b) the functionality of a stand-alone database is provided. The ability to simultaneously execute transactions under different isolation levels is a functionality offered by all stand-alone databases but not by their replicated counterparts. Allowing different isolation levels may improve overall system performance. For example, the TPC-C benchmark specification tolerates execution of some transactions at weaker isolation levels in order to increase throughput of committed transactions. In this Ph.D. Thesis we identify the conditions under which replication protocols may manage multiple isolation levels transparently and prove that protocols that satisfy these conditions are correct [3]. To this end, we suggest a new Snapshot Isolation level definition which fits better with the dependency graphs used as a base representation of transaction conflicts [1]. This definition unifies the base SI definition and the generalised version [2] suggested in other works [1, 4]. We also modify the popular ROWAA-based replication scheme to support different isolation levels. As an example, we demonstrate how these extensions can be applied to specific protocols. The majority of the replication solutions under consideration require only minor changes to support multiple isolation levels, which may result in an improved degree of concurrency and minor transaction completion times for those transactions that can be executed in a relaxed isolation level. Our model is general enough to be applied to any database replication protocol. [1] A. Adya. Weak Consistency: A Generalized Theory and Optimistic Imple- mentations for Distributed Transactions. PhD thesis, Massachusetts Insti- tute of Technology, March 1999. [6] H. Berenson, P. Bernstein, J. Gray, J. Melton, E. O'Neil, and P. O'Neil. A critique of ANSI SQL isolation levels. In Intnl. Conf. on Manag. of Data (SIGMOD), pages 1-10, San José, CA, USA, May 1995. ACM Press. [10] J. M. Bernabé-Gisbert and F. D. Muñoz-Escoí. Supporting multiple isolation levels in replicated environments. Data And Knowledge Engineering, 79 - 80(0):1 -16, 2012. [19] S. Elnikety, F. Pedone, and W. Zwaenepoel. Database replication providing generalized snapshot isolation. In 24th Intnl. Symp. on Reliab. Distrib. Syst. (SRDS), pages 73-84, Orlando, FL, USA, October 2005. IEEE-CS Press.