Experimental comparison of voting algorithms in cases of disagreement

Voting algorithms are used to mask erroneous results from redundant subsystems (variants) in fault tolerant systems. While voting algorithms are well known and widely used, the authors are not aware of any detailed investigation of voter behaviour in the presence of multiple errors. The results of an experimental evaluation of seven voting algorithms in a variety of simulated error scenarios are reported. A software error injection approach is used to simulate multiple errors in a triple modular redundant configuration. The Majority and Plurality voters produce the lowest number of catastrophic errors, in these tests. The Median voter produces the largest number of correct results, but also produces the largest number of catastrophic errors. The Three Domain voter results show a compromise between the large number of correct results identified by the Median voter and the small number of catastrophic results produced by the Majority voter.