The paper presents an algorithm for the implementation of e-transactions that execute exactly-once, despite failures. The algorithm is based on an asynchronous replication scheme that tolerates unreliable failure detection and replica nondeterminism. The algorithm may vary dynamically between primary-backup and active replication schemes. A three-tier architecture is used in devising the algorithm.

The paper should interest all researchers and developers who are working on transaction processing in distributed systems. It is well organized with a clear description of the proposed algorithm, thus easy to read for anyone with a theoretical background in the field.