Implementing Multiple Task Modular Redundancy

This paper presents a characterization of a new class of fault-tolerant systems based on the concept of task diagnosability (or equivalently f/(n-l)-diagnosability). Task diagnosability can be achieved on systems with disjoint comparison testing graphs, and this property can be used to validate correct task results without resorting to an auxiliary procedure for processor diagnosis that is prone to fail for intermittent or transient faults. Issues of task allocation and synchronization as well as implementation are considered