A new paradigm for the design of dependable systems

A new general strategy for dependability procurement in digital systems that allows highly selective use of redundancy with low overhead is described. The proposed approach is based on attaching a dependability tag (d-tag) to each data object and updating pertinent d-tag values as the computation progresses. Normal operations on data objects tend to lower d-tag values, while comparisons and voting on redundant copies of the same result, possibly obtained by different resources and/or algorithms, work in the opposite direction. Judicious intermixing of dependability-lowering and dependability-raising operations (driven dynamically by the dependability requirements for various results) can lead to desired dependabilities for computation results with minimal cost. Following an exposition of basic concepts of the proposed method with the assumption of perfect (error-free) d-tags and operations, a program of research is outlined for dealing with erroneous d-tags, imperfect operations, and other extensions. Some benefits and applications of the proposed approach are discussed