Recovery of fault-tolerant real-time scheduling algorithm for tolerating multiple transient faults

Abstract

The consequences of missing deadline of hard real time system tasks may be catastrophic. Moreover, in case of faults, a deadline can be missed if the time taken for recovery is not taken into account during the phase when tasks are submitted or accepted to the system. However, when faults occur tasks may miss deadline even if fault tolerance is employed. Because when an erroneous task with larger execution time executes up to end of its total execution time even if the error is detected early, this unnecessary execution of the erroneous task provides no additional slack time in the schedule to mitigate the effect of error by running additional copy of the same task without missing deadline. In this paper, a recovery mechanism is proposed to augment the fault-tolerant real-time scheduling algorithm RM-FT that achieves node level fault tolerance (NLFT) using temporal error masking (TEM) technique based on rate monotonic (RM) scheduling algorithm. Several hardware and software error detection mechanisms (EDM), i.e. watchdog processor or executable assertions, can detect an error before an erroneous task finishes its full execution, and can immediately stops execution. In this paper, using the advantage of such early detection by EDM, a recovery algorithm RM-FT-RECOVERY is proposed to find an upper bound, denoted by Edm Bound, on the execution time of the tasks, and mechanism is developed to provide additional slack time to a fault-tolerant real-time schedule so that additional task copies can be scheduled when error occurs.