Self-Testing Fault-Tolerant Real-Time Systems

We propose a periodic diagnostic algorithm based on the testing model of computation for real-time systems. The diagnostic task runs on every processor of the system. When the task starts execution, all the processors are synchronized and will be doing the same operation at every step of the algorithm. Each processor performs a test of itself and generates a token which contains the test result. Then the token is passed to some neighboring processors to check if a failure has occurred. In our model, a faulty processor does not necessarily stop functioning and it may behave in erratic manners when checking the token of a processor it is assigned to test. We give the conditions under which all processor failures are detected in a torus interconnection network, where each processor is tested by a minimum number of processors.