Towards Unified Tool Support for Real-time Calculus & Deterministic Network Calculus

Real-Time Calculus (RTC) and Deterministic Network Calculus (DNC) both provide the tooling to derive worstcase performance bounds for characteristics of distributed and networked systems: nodal backlog and communication delay. In fact, both were derived from a common basis, yet, their respective evolution let them diverge in various aspects. These are also reflected in the strengths of current, separate tools for RTC and DNC. In this paper, we investigate a potential unification of their tool support. To that end, we provide insights on similarities and differences, suggestions to exploit and overcome them, as well as a benchmark showing first benefits of a unified tool. I. APPLICATION DOMAIN & CHALLENGE Knowledge about rutime performance is central to the operation as well as the certification of systems with real-time constraints. For formal verification, worst-case analyses are applied. These guarantee to deterministically bound the potential worst-case behavior of a system w.r.t. certain performance metrics. In distributed computing and communication systems, these are foremost the buffer demand of system components and worst-case execution or traversal times, i.e., delay. RealTime Calculus (RTC) [1] and Deterministic Network Calculus (DNC) [2] are two versatile methodologies that can compute such bounds. They both employ envelope functions that deterministically bound system behavior, i.e., worst-case service offerings and task or data arrivals, as their principal model. These so-called curves have been accompanied with operations. They manipulate curves to compute the aforementioned worst-case performance bounds [2], [3]. Verification of worst-case performance measures becomes ever more important for industries like avionics. For instance, generations of in-cabin network designs by Airbus, formerly EADS, have been analyzed with RTC (Heterogeneous Communications System [4]) and DNC (AFDX data networks [5]). Yet, these efforts look rather distinct due to the calculi’s different analysis approaches. We illustrate them on the prime freely available tools: the RTC toolbox [6] and the DiscoDNC [7]. The RTC toolbox aims at modular performance analysis. Its standard way of usage is via its Matlab frontend. It provides an abstraction of the calculus operations to components such as the Greedy Processing Component (GPC). These are crossed by the resources considered in the analysis. When crossing a GPC, available service becomes left-over service to be used by a subsequent component and data arrival constraints become data output constraints to be fed into another component. In This work is supported by a Carl Zeiss Foundation grant. fact, the analysis frontend thus requires to provide the order of calculus operations by interconnecting components correctly. The DiscoDNC [7], in contrast, automated the derivation of the order of the calculus’ operations. It takes a network model consisting of servers that is crossed by data flows – both deterministically modeled with service curves and arrival curves. The order of operations to apply to the curves is then derived by the DiscoDNC. The tool offers alternative derivations that consider different analysis principles. Depending on the analyzed network size, they can vastly increase the quality of end-to-end delay bounds as well as analysis run times [8]. In this paper, we investigate a potential unification of tool support for RTC and DNC.