Performance Evaluation of Rollback-Recovery Techniques in Computer Programs

Rollback-recovery is a backward error-recovery technique used to recover from temporary faults in database and process-control systems. Rollback in process-control systems is generally constrained by deadlines, thereby requiring a dynamic insertion of rollback points. This is in contrast with the rollback-recovery in database systems where rollback points are inserted at equidistant intervals. This paper develops a simple model based on a semi-Markov process to study the performance of rollback-recovery strategies. Using this model, the mean program completion time is obtained for both database and process-control systems when rollback-recovery is implemented. The analytic results obtained by the semi-Markov model are compared with the simulation results by means of extensive computer simulation.