Runtime Diversity against Quasirandom Faults

Abstract

Complex software based systems that have to be highly reliable, are increasingly confronted with fault types whose corresponding failures appear to be random, although they have a systematic cause.  This paper introduces and defines these "quasirandom" faults.  They have certain inconvenient common properties such as their difficulty to be reproduced, their strong state dependence and their likelihood to be found in operational systems after testing.  However, these faults are also likely to be detected or tolerated with the help of diversity in software, and even low level diversity which can be achieved during runtime is a promising means against them.  The result suggests, that runtime diversity can improve software reliability in complex systems.