Error Recovery in Multi-Version Software

Abstract In multi-version software, design faults in versions are tolerated through comparison of results from several diverse versions. Since all versions are likely to contain some design faults, it is necessary to have recovery and reconfiguration mechanisms that are able to recover these versions as they fail. Community Error Recovery, an error recovery algorithm for multi-version software, has been designed and implemented on the UCLA DEDIX system. It consists of two levels: the cc-point level is responsible for local error recovery and the recovery point level is responsible for global error recovery. Recovery of failed versions is achieved through the injection of "consensus" data values supplied by the decision algorithm of the MVS system. The provision of two levels minimizes both the disturbance to the system and the restrictions to the implementation of diverse versions. Markov models have been developed to evaluate the reliability of multi-version software running on DEDIX with and without Community Error Recovery.