Software fault-tolerance by design diversity, DEDIX : a tool for experiments

One way of reducing the effects of an error introduced during the design of a program is to use multiple versions of the program, independently designed from a common specification. If these versions are designed by independent programming teams, it is to be expected that a fault in one version will not have the same behavior as any fault in the other versions. Since the errors in the output of the versions will be different and uncorrelated, it is possible to run the versions concurrently, cross-checking their results at prespecified points, and mask errors. A DEsign DIversity eXperiments (DEDIX) testbed has been implemented at UCLA to study the influence of common mode errors which can result in a failure of the entire system. The layered design of DEDIX and its decision algorithm are described. The usage of the system and its application in an ongoing experiment are explained.