Voluntary preemption: A tool in the design of hard real-time systems

Abstract

A new technique for scheduling hard real-time systems is proposed. We term this technique as voluntary preemption as it allows the user to determine the points at which the execution of a process can be preempted. There are several advantages of this technique. First of all, the internal structure of the processes can be used in determining the points where a process may be preempted. This permits, among other things, that critical sections are not preempted as well as identifying points at which a process should be preempted. Secondly, overhead cost of context switching can be taken into account realistically without sacrificing performance. Finally, since the overhead cost of context switching depends on the point (in the execution) where a process is preempted, this technique allows the preemption points to be chosen to reduce the overhead cost of context switching. These various considerations about where a process can be preempted, can be incorporated into an automatic tool to aid in the design of real-time systems.

We describe and illustrate this technique by analyzing in detail the performance of a voluntary preemptive scheduling algorithm for scheduling a classical hard real-time problem on a single processor with context switching costs. In addition we also obtain a tight load cutoff for this algorithm. The performance of the voluntary preemptive scheduling algorithm is good as its load cutoff matches that of an optimal scheduling algorithm when context switching costs are zero. This load cutoff theorem also provides a mechanism for scheduling a multiprocessor system by partitioning the problem into independent subproblems each of which can be scheduled on a single processor.

Part of this work done while at Odyssey Research Associates under U.S. Air Force, RADC contract F30602-86-C-0115