Fault-tolerance for communicating multidatabase transactions

We propose a framework that can be used for the implementation of reliable multidatabase system communication protocols. The framework provides transactions on shared write-once objects, termed communication objects, that can be recovered after failures. Also concurrency is supported by recoverable processes. The realization of this framework that can be provided by means of a distributed coordination kernel, is focused upon. Strategies for communication objects that belong to different reliability classes are presented. The application of this framework for MDBS transaction processing is shown and the failure behavior of a MDBS communication protocol supporting global serializability through a global two-phase-commit is discussed.<>