Rollback recovery in real-time systems with dynamic constraints

Abstract

Rollback recovery is a backward error recovery technique for recovering from transient faults in computing systems. Real-time systems employing fault tolerance and reconfiguration generally have time-dependent (dynamic) constraints. The author presents a novel rollback point insertion strategy which evaluates the rollback conditions on-line. The technique minimizes both time and space overhead associated with rollback, thereby making it applicable to real-time systems with dynamic constraints. The algorithm presented attains a near-optimum solution in terms of the time spent in saving the states of the system. Details of the simulation conducted to validate the technique are also given. The simulation study has established that the degradation in performance due to using the proposed algorithms is insignificant and the precomputation time is very small for programs that can be represented by general acyclic graphs. On the other hand, for structured programs that can be represented by a simple sequence of tasks, the computation overhead is almost zero.<>