Performability-driven adaptive fault tolerance
A.T. Tai · 2002
We introduce a performability-driven realization of adaptive fault tolerance. The object system for the investigation is a multiprocessor system in a real-time context accommodating parallel computation and two fault tolerance schemes, namely, recovery blocks (RB) and distributed recovery blocks (DRB). The optimality criterion is directly specified in terms of a preferability measure, based on which adaptive rules are generated to govern the utilization of RB and DRB during system's operation. The analysis shows that performability concepts and modeling methods enable us to manage the tradeoffs among system attributes in a formal manner and lead to effective realization of adaptive fault tolerance.>