Efficient Admission Control for Enforcing Arbitrary Real-Time Demand-Curve Interfaces

Server-based resource reservation protocols (e.g., periodic and bandwidth-sharing servers) have the advantage of providing temporal isolation between subsystems co-executing upon a shared processing platform. For many of these protocols, temporal isolation is often obtained at the price of over-provisioned reservations. Other more fine-grained approaches such as real-time calculus permit a precise characterization of the resources required by a subsystem via demand-curve interfaces. However, an important, unsolved challenge for subsystems specified by such interfaces is the development of efficient enforcement techniques to guarantee temporal isolation between the subsystems. Admission control algorithms can be used in this regard to ensure that the cumulative subsystem demand never violates the demand-curve specified by the interface. In this paper, we address the challenge by designing admission controllers for complex, arbitrary demand-curve interfaces and proposing enforcement techniques. First, we propose an exact algorithm and show that its complexity is infeasible for long-running systems. To address this drawback, we then design an approximation algorithm and associated enforcement techniques to handle unpredictable execution times. We validate, via simulations, that our approximate approach is significantly more efficient than the exact approach with only minor decrease in the accuracy of the admission controller.