A Dynamic Cache Reconfiguration Platform for Soft Real-Time Systems

Dynamic Cache Reconfiguration (DCR) significantly improves performance and energy consumption. However, implementing this technique in real-time systems is non-trivial, because it introduces unpredictability to the system. The few DCR approaches for soft real-time systems that have been proposed rely on offline preprocessing and exhaustive simulation. In this paper, a dynamic cache reconfiguration platform for soft real-time systems is proposed. Our platform enables the design of DCR algorithms in a soft real-time simulation environment and obtains statistics in terms of performance, energy consumption and real-time accuracy. We also developed two fully online DCR heuristics for soft real-time systems, which do not rely on any offline processing or simulation. Our results show that our heuristics significantly improve the energy consumption and performance of the cache while keeping a minor impact on timing constraints for most of our test applications.