Network partitions in distributed databases

Abstract

The technology of distributed databases is based on two other technologies which have developed a solid foundation: computer networks and centralized databases. A distributed database designer should therefore address database-related as well as network-related issues. Provision has to be made for the detection of and recovery from failure of the database as well as the network components. One type of failure resulting from distribution is network partitions. During a network failure the sites of the distributed database are subdivided into two or more disjoint groups. These groups continue processing newly submitted transactions without any knowledge of the operations at the sites in the other partition groups. Since data may be replicated, the various copies of a data item may be updated inconsistently at the different partition groups. When handling continued transaction processing during network partitions, a trade-off is made between the level of availability and the consistency provided by the system. Pessimistic partition processing strategies ensure database consistency by reducing availability. The token lists pessimistic partition processing strategy, which utilizes a reliable broadcasting facility, proved to be the best for the system under consideration.