Achieving Efficiency without Sacrificing Model Accuracy: Network Calculus on Compact Domains

摘要

Messages traversing a network commonly experience waiting times due to sharing the forwarding resources. During those times, the crossed systems must provide sufficient buffer space for queueing messages. Network Calculus (NC) is a mathematical methodology for bounding flow delays and system buffer requirements. The accuracy of these performance bounds depends mainly on two factors: the principles manifesting in the NC flow equation and the functions describing the system. We focus on the latter aspect. Common implementations of NC overapproximate these functions in order to keep the analysis computationally feasible. However, overapproximation often results in a loss of accuracy of the performance bounds. In this paper, we make such compromising tradeoffs between model accuracy and computational effort obsolete. We limit the accurate system description to functions of a compact domain, such that the accuracy of the NC analysis is preserved. Tying the domain bound to the algebraic operators of NC instead of the operational semantics of components, allows us to directly apply our solution to algebraic NC analyses that implement principles such as pay burst only once and pay multiplexing only once.