Off-line analysis techniques for hard real-time systems are all based on the assumption that we can estimate the worst-case execution time of the different tasks executing in the system. In the traditional cyclic-executive schedulers, execution time limits were enforced for each task by the scheduler. Unfortunately, in concurrent hard real-time systems such as those using the tasking model defined in Ada, no bound on the execution time of tasks is enforced, which may result in a system timing malfunction not detected by the analysis techniques. In this paper we explore the implementation of execution time clocks within the task scheduler, and we describe methods to detect execution time overruns in the application, and to limit their effects. We also discuss the use of execution time clocks to enhance the performance of sporadic server schedulers implemented at the application level.