Accounting for Cache Related Pre-emption Delays in Hierarchical Scheduling

Hierarchical scheduling provides a means of composing multiple real-time applications onto a single processor such that the temporal requirements of each application are met. This has become a popular technique in industry as it allows applications from multiple vendors as well as legacy applications to co-exist in isolation on the same platform. However, performance enhancing features such as caches mean that one application can interfere with another by evicting blocks from cache that were in use by another application, violating the requirement of temporal isolation. While one solution is to flush the cache after every application context switch, this can potentially lead to a degradation in performance. In this paper, we present analysis that bounds the additional delay due to blocks being evicted from cache by other applications in a system using hierarchical scheduling.