A non-intrusive checkpointing protocol

The authors address the problem of global consistency of a loosely coupled system of processes in the presence of failures. In particular, they present a checkpointing protocol that guarantees the existence of a globally consistent state from which the system can be restarted if and when process failures occur. The protocol itself is resilient to process failures and is unique in the degree of its noninterference with normal activities of the processes in the system.<<ETX>>