Governing with Insights: Towards Profile-Driven Cache Management of Black-Box Applications

There exists a divide between the ever-increasing demand for high-performance embedded systems and the availability of practical methodologies to understand the interplay of complex data-intensive applications with hardware memory resources. On the one hand, traditional static analysis approaches are seldomly applicable to latest-generation multi-core platforms due to a lack of accurate micro-architectural models. On the other hand, measurement-based methods only provide coarse-grained information about the end-to-end execution of a given real-time application. In this paper, we describe a novel methodology, namely Black-Box Profiling (BBProf), to gather fine-grained insights on the usage of cache resources in applications of realistic complexity. The goal of our technique is to extract the relative importance of individual memory pages towards the overall temporal behavior of a target application. Importantly, BBProf does not require the semantics of the target application to be known - i.e., applications are treated as black-boxes - and it does not rely on any platform-specific hardware support. We provide an open-source full-system implementation and showcase how BBProf can be used to perform profile-driven cache management