Considerations on Software Diversity on the Basis of Experimental and Theoretical Work

Diversity has been proposed as a method of achieving software fault-tolerance. Several versions of a program are written on the basis of the same specification and executed in parallel. Most of the past considerations about reliability improvement through multiversion programming depended on the assumption that independently developed programs would also fail independently. Recent work, however, has presented convincing arguments both in an experimental [KNI86] and a theoretical [ECK85] approach leading to the conclusion that the independence assumption does not hold in the general case. Therefore an analysis of reliability of diverse programming has to include the effect of dependent programming errors.