Method and system for managing replicated data with enhanced consistency

Abstract

Method and system for managing transactions operating on replicated data items at multiple physical sites in a distributed database system. The transactions include local transactions each being processed at one of the physical sites and global transactions each being processed at multiple physical sites. Each transaction originates at one of the physical sites, and is permitted to read data items only at that physical site. In addition, only transactions originating at a physical site at which a primary copy of a data item is located are permitted to update that data item. A number of virtual sites are defined such that a given virtual site VS.sub.i.sup.j at a particular point in time includes a set S.sub.i.sup.j of all data items at a physical site s.sub.j that a given transaction T.sub.i has accessed from an initial operation up to that point in time. A replication graph is maintained to represent the relationships between the transactions and the virtual sites. The replication graph includes a first set of nodes for the transactions, and a second set of nodes for the virtual sites. A node for a transaction T.sub.i and a node for a virtual site VS.sub.i.sup.j are connected in the graph if a schedule of transaction operations includes a write operation by transaction T.sub.i on a replicated data item in the set of data items associated with virtual site VS.sub.i.sup.j. The replication graph is maintained as part of a concurrency-control protocol and the absence of cycles in the graph may be used to indicate whether a given schedule is globally serializable.