Policy-independent real-time operating system mechanisms for timing error detection, handling and monitoring

Abstract: Most research focusing on timing errors deals with scheduling policies that avoid the errors. Since many of the policies are based on estimated worst-case execution times for each task, reliability is a function of the accuracy of the estimates. As a result, many hard real-time systems are implemented with the dangerous assumption that due to correct design and testing, a missed deadline will never occur. We have designed novel policy-independent mechanisms for detecting and handling timing errors, and for monitoring real-time tasks. The detection and handling requires less than 1 microsecond overhead per reschedule operation, and has a latency approximately the length of one context switch for handling an error. The monitoring mechanism uses 6 microsecond per context switch, and requires only 1 Kbyte of memory per 32 processes in the system. 1