Roll-Forward Recovery with State Estimation

We propose and analyze a novel roll-forward recovery scheme that is based on state estimation. In this solution, a dedicated monitoring component has the capability to predict future states of the physical environment by periodically analyzing the state of an associated monitored component. These predictions can serve either directly as restart states for a failed component or as facilitators of the recovery process. Our solution combines three key concepts: identification and handling of a component´s ground state, resilience of service users to occasional but bounded service outage, and state estimation. The main benefit of the proposed scheme is that it avoids replication and thus provides responsive recovery for components that lack the property of replica-determinism. The aim of this work is recovery of time-critical but not safety-critical components in mixed-criticality real-time systems. We show that the proposed recovery scheme has comparable error-detection and repair efficiency to roll-forward recovery with behavior-based checks introduced by Xu and Randell.