Use of preferred preemption points in cache-based real-time systems

Time-critical applications require known worst-case execution times to ensure that system timing constraints are met. Traditional cache memory arrangements, however significantly impede the determination of tight upper bounds on these worst-case execution times (WCET). The difficulty comes in adequately predicting the cache miss ratio for a task in a preemptable multi-tasking environment. Caches thus increase the complexity of calculating WCET. To resolve this, caches have simply been excluded from WCET calculations. Each task must then be provided greater time in which to execute leading to lower throughput and performance. In this paper we present a cache management scheme that allows WCET calculations to more easily reflect the timing effects of caching. This is done through the appropriate selection of preemption points within a task's execution. The scheme focuses on the WCET component that is due to preemption overhead. An added benefit is a reduction in execution time of up to 10% for some tasks over traditional cache management.<
>