Systematic and design diversity - Software techniques for hardware fault detection

Abstract

For the detection of hardware operational faults in most safe systems static redundancy is used. Thus, in the most simple case we have the well known Duplex System. If design fault detection is required, design diversity in the software has to be used, too. We suggest the combined utilization of so called systematic diversity and design diversity in a time-redundant system instead of the structural redundant Duplex System. For this purpose two diversly designed and systematically transformed variants of an application program are executed sequentially on the same processor. We call this new approach a Virtual Duplex System. In this paper we investigate the safety of a Virtual Duplex System. We propose the use of software diversity techniques (i.e. systematic diversity) to detect nearly all hardware faults in this system. Transient faults are effectively detected through the time redundancy and permanent faults by the new software diversity approach. In addition software design faults and even compiler-, library-, operating system- and underlying hardware design faults can be detected. The proposed software techniques are either new or never considered systematically for the detection of hardware faults in a general purpose system environment with design diversity. As an example the new systematic diversity technique ‘simple register permutation’ was applied on different application programs by means of a simple heuristic. The technique was evaluated experimentally by injecting permanent hardware faults with the fault injection tool ProFI and measuring the safety of Virtual Duplex Systems. The results are compared to systems that do not use special fault detection (Simplex Systems) and Virtual Duplex Systems that use pure design diversity. The experiments show that even by simple systematic diversity most permanent hardware faults are detected.