Using dynamic, full cache locking and genetic algorithms for cache size minimization in multitasking, preemptive, real-time systems

Abstract

Cache locking have shown during the last years their usefulness easing the schedulability analysis of multitasking, preemptive, real-time systems. Cache locking provides a high degree of predictability while system performance is maintained at a similar level to that provided by regular, highly unpredictable, non-locked cache. Cache locking may also be useful to reduce hardware costs by means of reducing the size of the cache memory needed to make a real-time system schedulable.This work shows how full, dynamic cache locking may help to reduce the size of the cache memory versus a regular cache. This reduction is possible thanks to a genetic algorithm that selects the set of instructions that have to be locked in cache to provide the maximum cache size minimization while keeping the system schedulable.