DIMTOOL: A Platform for Determining Worst Case Latencies in Switched Queuing Networks Emanuel Heidinger ∗ † , Stefan Burger ∗ , Stefan Schneele ∗ , Alexander Klein † , Georg Carle † ∗ EADS Innovation Works Dept. IW-SI-CO D-81663 M¨ unchen, Germany Email: {emanuel.heidinger,stefan.schneele, stefan.burger}@eads.net † Technische Universit¨ at M¨ unchen Institut f¨ ur Informatik, I-8 Boltzmannstraße 3 D-85748 Garching b. M¨ unchen, Germany Email: {klein,carle}@net.in.tum.de Abstract—Performance evaluation in computer networks re- quires consistent traffic and topology models to deliver compa- rable results. In this work we present our platform that provides a consistent interface by encapsulating standard parameters as scheduling strategy, link speed, processing delay, and propagation delay. In contrast to existing solutions, we provide several perfor- mance estimation techniques within a single toolbox making the identified performance results comparable. The functionality of our developed toolbox is demonstrated by employing it to real- world scenarios in avionics, which is the Ethernet based Cabin Management System and the communication network inside the cabin server. I. I NTRODUCTION Automotive and aeronautics industry have to cope with evolved communication networks in vehicles and airplanes. However, there are good reasons to benefit from recent progress in the field of switched Ethernet. In aeronautics, a clear cost driver is the weight of the communication in- frastructure, and so, any additional redundant link results additional weight. Transmitting more traffic over the network while possibly merging different safety domains would bring clear advantage compared to current systems. In automotive industry, the cost drivers in terms of network technology are connectors, cable and their shielding, as well as the price for the network chips. The objectives of both means of transportation differ somewhat, but the intention is clear: less weight, lower price, higher bandwidth, and yet no loss in terms of safety. Certainly, these objectives influence each other and are difficult to fulfill. Often an existing and expensive cer- tification process does not encourage the introduction of novel communication technologies. In terms of safety, an important point that be addressed is the certification process, which differs in the certain areas of transportation. The evaluations in those processes are often hand-crafted or semi-automated. Our toolchain tries to fill that gap by providing a consistent platform for performance evaluation and calculation such that certification processes can be completed in shorter time. Figure 1 shows the aspects of our performance calculation toolbox. The task of performance estimation and calculation yielded several approaches in the last decades. There are discrete event simulations based on Monte Carlo studies, methods that aim at the calculation of queuing delays based on queuing theory as Proc Delay Type Max Mean Min Median Trans Prop Total Queue Aspects WC MIP NC Sim Performance Calculator Traffic Model Token Bucket Model Stochastical Model Fig. 1. Performance Calculation with DIMTOOL well as its alternative Network Calculus (NC). The calculated results often fall in the category of latency, memory utilization, and link utilization. To the best of our knowledge, these tools deliver results, that cannot easily be compared since they may assume different models or constraints. In this work we present the network calculation platform DIMTOOL, which provides such a consistent view on network performance calculations. This paper provides the following key contributions: In contrast to previous tools, we provide several performance estimators under one umbrella, thus making performance results of different methods transparent and comparable. We employ DIMTOOL to verify latency constraints in the switched aircraft cabin and show that the NC FIFO bound does not hold by Worse Case Simulation (WCS). In contrast, we did not observe the necessity of the non-FIFO bound with the established Monte Carlo method— even after several random runs. This paper is structured as follows: Section II shows related work in the field of performance calculators. Section III pro- vides an overview of DIMTOOL, its objectives, its architecture and the provided back-ends. The experimental results are discussed in Section IV. Section V summarizes the work and outlines further steps. II. RELATED WORK In this section we provide a comprehensive overview of network performance calculation solutions. Throughout this work, network performance calculators are classified by the employed technology. More precisely, we cover network sim- ulation based on Monte Carlo Simulation (SIM), NC analysis, VALUETOOLS 2012, October 09-12, Cargèse, France Copyright © 2012 ICST DOI 10.4108/valuetools.2012.250296