Definition and analysis of hardware- and software-fault-tolerant architectures

A structured definition of hardware- and software-fault-tolerant architectures is presented. Software-fault-tolerance methods are discussed, resulting in definitions for soft and solid faults. A soft software fault has a negligible likelihood or recurrence and is recoverable, whereas a solid software fault is recurrent under normal operations or cannot be recovered. A set of hardware- and software-fault-tolerant architectures is presented, and three of them are analyzed and evaluated. Architectures tolerating a single fault and architectures tolerating two consecutive faults are discussed separately. A sidebar addresses the cost issues related to software fault tolerance. The approach taken throughout is as general as possible, dealing with specific classes of faults or techniques only when necessary.<<ETX>>