Locking policies in distributed databases

Abstract

In Distributed Databases the typical problems of Centralized Databases become more difficult. One of them is Concurrency Control. It can be summarized as follows. Users of the Database access it by executing transactions. Different transactions are executed concurrently therefore their actions interleave. Without proper control this interleaving may produce incorrect results, even if individual transactions are correct. The Concurrency Control process has to prevent these situations. There are several possible mechanisms for controlling concurrency, of which the most widely used is Locking. In this paper we examine and analyze Locking as a Concurrency Control mechanism for Distributed Databases. We define Distributed Locking Policies (methods for locking entities in Distributed Databases) and show how existing Policies for a Centralized Database generalize to the Distributed case. We also define a new category of Distributed Locking Policies, D-policies, into which these generalizations fall. An algorithm which determines whether all transactions of a given D-policy are guaranteed to produce only correct interleavings (are safe) is presented. The algorithm is efficient, even though testing an arbitrary set of transactions for safety is coNP-complete. However, we prove that optimal locking of transactions to satisfy the conditions tested by the algorithm is NP-hard even for a Centralized Database.