Making Shared Caches More Predictable on Multicore Platforms ∗

In safety critical domains, the usage of multicore plat-forms has been hampered by problems due to interactions across cores through shared hardware. The inability to pre-cisely characterize such interactions can lead to pessimism in worst-case execution time analysis that is so great, the extra processing capacity of additional cores is entirely negated. In this paper, a new framework called $MANRTis proposed for dealing with such interactions in the context of shared caches. The major thesis of this paper is that the management of cache lines is a synchronization problem. $MANRTcontrols cache usage by using recently developed optimal real-time multiprocessor locking protocols in con-junction with page coloring. The idea is to associate a set of colors with each task and require it to lock its needed colors whenever it is invoked. Experiments are presented herein that show that $MANRTcan greatly lessen observed worst-case execution times and correspondingly improve schedu-lability.