Transaction management in multidatabase systems
Dimitrios Georgakopoulos · 1991
The concept of Multidatabase Systems (MDBS) was introduced to support applications that access data stored in multiple databases, controlled by autonomous and possibly heterogeneous Local Database Systems (LDBSs). The autonomy and the heterogeneity of the LDBSs causes several new problems in multidatabase transaction management that do not exist in other distributed database systems. The primary difficulty in enforcing global serializability in a multidatabase environment is due to the fact that transaction execution and serialization orders can be different. To determine the serialization order of global transactions the MDBS must take into account conflicts caused by local transactions. However, due to the autonomy of the LDBSs such information is not available. A serious problem in multidatabase recovery is that MDBS recovery actions constitute new transactions. This complication redefines the recoverability requirements in a MDBS. Most solutions proposed to deal with the above problems either allow incorrect results or place severe restrictions on global and local transactions. In the global serializability area the thesis contributes in several aspects. We first define a subclass of LDBSs which produce rigorous schedules where the execution order of transactions determines their serialization order. We also propose a family of concurrency control methods and prove that they guarantee global serializability without violating the autonomy of the local systems. The ticketing method assumes only local serializability while its refinements take advantage of the possible rigorousness of the LDBSs. In the area of multidatabase recovery, we introduce a multidatabase recoverability condition as the minimal requirement that can assure that multidatabase recovery preserves the consistency of a multidatabase system. We also describe a recovery mechanism that takes advantage of the local recovery in the LDBSs by minimizing the replication of recovery tasks. Another contribution of this thesis is the introduction of a new correctness criterion that is based on time. We show that chronological correctness captures temporal transaction dependencies in addition to other time-independent conflicts among transactions. We also propose a chronological scheduler which releases transaction isolation, takes into account transaction duration and allows implicit commitment of transactions.