Memory System Architecture for Real-Time Multitasking Systems

Traditionally, caches have been used to reduce the average case memory latency in computer systems. However, real-time systems using preemptive scheduling algorithms cannot make use of this reduced latency, as it is not deterministic. A real-time system only benefits from an improvement of the worst case performance. Caches do little to improve worst case performance because preemptions allow cached data to be overwritten at unpredictable times, leading to nondeterministic behavior. Real-time systems using execu- tion quantization can deterministically make use of a significant portion of the processor's resources despite the use of slow memory. Execution quantization allows deterministic caching by guaranteeing that tasks can execute for fixed length blocks of time once they start running, thereby controlling preemptions, and allowing cache restoration to occur predictably and efficiently. Thesis Supervisor: Jonathan Allen Title: Professor of Electrical Engineering and Comp..