FTMP—A highly reliable fault-tolerant multiprocess for aircraft

FTMP is a digital computer architecture which has evolved over a ten-year period in connection with several life-critical aerospace applications. Most recently it has been proposed as a fault-tolerant central computer for civil transport aircraft applications. A working emulation has been operating for some time, and the first engineering prototype is scheduled to be completed in late 1979. FTMP is designed to have a failure rate due to random causes of the order of 10^-10failures per hour, on ten-hour flights where no air-borne maintenance is available. The prefered maintenance interval is of the order of hundreds of flight hours, and the probability that maintenance will be required earlier than the preferred interval is desired to be at most a few percent. The design is based on independent processor-cache memory modules and common memory modules which communicate via redundant serial buses. All information processing and transmission is conducted in triplicate so that local voters in each module can correct errors. Modules can be retired and/or reassigned in any configuration. Reconfiguration is carried out routinely from second to second to search for latent faults in the voting and reconfiguration elements. Job assignments are all made on a floating basis, so that any processor triad is eligible to execute any job step. The core software in the FFMP will handle all fault detection, diagnosis, and recovery in such a way that applications programs do not need to be involved. Failure-rate models and numerical results are described for both permanent and intermittent faults. A dispatch probability model is also presented. Experience with an experimental emulation is described.