Error detection based on execution-time monitoring

This paper examines if monitoring of task-execution times can be used to detect errors, so that undesired system behavior and failures of real-time systems can be averted early. To this end, the paper investigates if respectively how the temporal behavior of algorithms changes in the presence of errors that have been caused by hardware faults. We used software-implemented fault injection on a number of benchmark programs to create errors and performed runtime measurements for the altered benchmarks to check if the observed execution times are below or above minimum respectively maximum execution-time bounds of the code. Our results show that up to 70% of errors that are undetectable with other standard techniques can be detected with this simple execution-time monitoring method. The method thus provides an additional layer of protection against errors. It can be implemented with reasonable effort and overhead.