A decentralized termination protocol

The smallest recoverable unit of work in a distributed database system is a tran saction. Whenever site failures leave the processing of a distributed transaction in a (potentially) unsafe state, a termination protocol is invoked to restore the database to a safe state enabling operational sites to proceed with future tran sactions. In this paper we propose one such termination protocol and sketch a proof of its correctness. The protocol is an example of a decentralized protocol, where each site assumes an equal and symmetric role. The proposed protocol is resilient to all combinations of site failures that do not partition the network.