Voting: A Paradigm for Adjudication and Data Fusion in Dependable Systems

Voting is used in realizing ultrareliable systems based on multichannel computation. In practical applications to date, the multiple computation channels have consisted of identical hardware circuits, which are expected to produce the same outputs, or diverse software modules, which may produce inexact or incomplete results. The challenge in hardware voting is the choice of adjudication points where voters must be inserted, and the associated synchronization and data throughput requirements. In software voting, algorithms for reconciling multiple inexact or incomplete results, and their time and code complexities, are important aspects of the problem. Chapter 4 deals with weighted voting algorithms and their hardware and software realizations, as a unifying methodology for a variety of voting schemes that have been found to be useful in practice. As in all algorithm design problems, correctness and performance of voting algorithms are key properties to evaluate and ascertain. We also show, through examples, that there is more to voting than the simple majority or plurality viewpoint heretofore dominating the dependable computing literature. It is argued that considering voting as a form of data fusion, and drawing from both mathematical studies of the anomalies in voting as well as sensor fusion work in signal processing, can be beneficial to further developments in the field of dependable computing.