Limited preemptive scheduling of non-independent task sets

Abstract

Preemption is a key factor against architectural coupling in concurrent systems. The whole verification process of real-time systems postulates composability in multiple dimensions, including time. As coupling wrecks composability, the design of real-time systems really needs preemption. However preemption effects complicate feasibility analysis or make it more pessimistic. Hence methods that limit preemptions without affecting feasibility are attractive. State-of-the-art approaches to limited preemption, however, do not treat resource sharing with the importance that it deserves. The placement of non-preemptive regions - and their interactions with shared resources - should not become a design problem, but rather stay as an implementation level feature that does not backtrack to the design space. In this paper we present a refinement to the state-of-the-art limited preemption model that addresses the interaction with resource sharing, and discuss a kernel implementation that uses run-time knowledge to warrant safe and efficient overlaps between critical sections and non-preemptive regions. Experimental results prove the effectiveness of the proposed solution.