HARDWARE VOTING OF TRANSPUTERS IN REAL - TIME nMR FAULT - TOLERANT SYSTEMS

The particular attraction of hardware voting is that a high degree of fault-tolerance should be achievable in a largely software transparent manner and with a minimal performance penalty. The proposals are particularly applicable to systems using multiple transputers, allowing both the natural parallelism of the application and the redundancy demanded by replication to be exploited in a consistent manner. An outline of the voter structure and its fault-masking capabilities is presented together with experimental performance figures from prototypes. Whilst some faults can be masked transparently by the hardware, it would be desirable to be able to use fault information acquired by the voters to allow the maximum degree of tolerance to be restored through reconfiguration. The paper discusses the nature of the errors which can be distinguished by the voter and the use of timeouts to help in this process.