On the effects of background selftests on system performance

Systems which perform built-in tests to provide continuous monitoring of system reliability are of increasing interest. To minimize the test overhead it is appropriate to divide the entire system test into several subtests, which are executed during the idle times of the system, i.e., in background mode. In this paper a model using renewal theory is described to determine the probability distribution function and the mean of the duration of an entire system selftest. Finally, we evaluate the effects of selftesting on system performance, expressed by the delays occurring and the decreased throughput of user tasks. Our analysis allows us to compare the performance effects of various test policies. At the end of the paper we deal with optimal subtest lengths.