In this project, the hardware issues - error detection and recovery - of rollback recovery technique are studied. The main objectives of this research are (1) to design a concurrent error detection scheme to facilitate recovery in multitasking and multiprocessor systems, and (2) to develop a scheme for recovery data protection and study its impact on the rollback strategy. Detection of errors with low error latency is vital for successful recovery in real-time systems. In most systems, however, the hardware to detect errors is subsumed. A watchdog processor is used to detect errors and initiate recovery. The technique of control flow checking is used for the on-line detection of errors. The concept of cyclomatic complexity is used in order to obtain a controllable error latency. In order to protect the recovery data in rollback, a recovery cache is employed. A new model for rollback is formulated to take into effect the use of a recovery cache for recovery data protection. The impact of the cache on rollback is investigated by deriving an optimization model. Analysis and experimental studies with real programs will be conducted to validate the model and the methods used for rollback.
