Some results about the efficiency of simple mechanisms for the detection of microcomputer malfunctions

Abstract

This paper describes results obtained by simulation, in order to evaluate the efficiency of very simple failure detectors: watch dog, memory violation, attempts to execute invalid operation codes. Although detection cannot be immediate, it is shown that the detection time may be very short: some tenths of millisecond to some milliseconds would be realistic values. The mechanisms are examined in detail and conclusions can be formulated for the design of future CPUs. The behaviour of programs in presence of hardware failure has been examined and a model has been developed, which permits to predict the detection time associated to the most likely failures. Finally the environment in which this type of malfunction detectors will be experimented is briefly described: an electronic switching system, involving some hundreds of `self-detecting' microcomputers.