Distributed database management system recovery from network partitioning

Abstract

The purpose of this thesis is to analyze the operations of a distributed database management system under network partitions, review a number of existing methods proposed to deal with this problem and to present an alternate approach that will allow multiple operating partitions upon network partitioning. When a network that supports a distributed database with redundant data becomes partitioned, each partition may function separately. Due to this, independent updates at each partition may cause inconsistencies to arise. At network reconnection time such divergent data, in particular copies of the same data in different partitions have to be reconciled. There is no known general method for doing so. Existing solutions are often unacceptable because system availability is reduced. Two recently proposed methods that allow continuous operation of multiple partitions may work for certain applications but are not general enough.