Flexible and Transparent Fault Tolerance for Distributed Object-oriented Applications

This report describes an approach enabling automatic structural recon gurations of distributed applications based on con guration management in order to compensate for node and network failures. The major goal of the approach is to maintain the relevant application functionality after failures automatically. This goal is achieved by a dedicated system model and by a decentralized recon guration algorithm based on it. The system model provides support for redundant application object storage and for application-level consistency based on distributed checkpoints. The recon guration algorithm detects failures, computes a compensating con guration, and realizes this new con guration. The report emphasizes exibility in the sense of adaptable levels of fault tolerance, as well as transparency in the sense of fully-automatic reaction to failures.