Promised Consistency for Rollback Recovery

Checkpointing protocols usually rely on the constitution of consistent global states, from which the application can restart upon a failure. This paper proposes a new characterization and technique to build a recoverable state, aiming at relaxing the constraints and overhead. , for Promised consistency, is proposed as such a recovery condition on a global state. A key idea is to use promised events: place holders forcing any restart to reach an actual global state of the first execution. A preliminary contribution is a formal treatment of potential causality, studying its impact on recoverability and determinism