Optimistic Partitioned Operation in Distributed Database Systems (Network Partitioning, Replicated Data, Consistency, Recovery, Fault-Tolerance)

147 p.Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1986.Network partitioning is a serious threat to the availability of replicated data in distributed database systems. To provide high data availability during partitioning, an optimistic strategy for transaction processing will allow transactions to be processed in any partition. Conflict detection and resolution will then be performed when partitions are merged. Recently, an automatic conflict detection and resolution method based on serialization graphs has been proposed in the literature. It has been argued that the graph-theoretic method can be attractive for certain applications.In this thesis, several problems related to the design and implementation of fault-tolerant transaction processing mechanisms that employ the graph-theoretic method are studied. A framework for the detection and synchronization of failures and recoveries is proposed. A major component of the framework is a new partition management protocol designed especially for optimistic partitioned operation. Several crash/merge recovery algorithms are developed and their implementation issues are discussed. This thesis also presents a simple validation algorithm that may be suitable for environments where transaction conflicts are rare.U of I OnlyRestricted to the U of I community idenfinitely during batch ingest of legacy ETD