IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 49, NO. 6, DECEMBER 2002 1251 Analysis and Simulation Methods for Performance Evaluation of a Multiple Networked Embedded Architecture Paolo Castelpietra, Ye-Qiong Song, Françoise Simonot-Lion, and Mondher Attia Abstract—This paper deals with the modeling and the validation of multiple networked embedded computer systems supporting in-vehicle applications. In this context, we developed a modular modeling and simulation technique. This approach allowed the development of reusable component models with clearly defined interfaces. The building of a whole application model is then obtained by the integration of these components, so that model construction and result analysis are made easy. In fact, thanks to the formal definition of components, interfaces, and composition rules, this step is automatically achieved. In the context of the CAROSSE project, we implemented this methodology in the Carosse-Perf tool. Herein, we will detail the methodology together with its modeling principles and the resulting definition of basic components and interfaces. Finally, we will apply it to a case study drawn from a PSA Peugeot-Citroën application. Index Terms—Computer network performance, local area networks, real-time systems, road vehicle electronics, simulation software. I. INTRODUCTION T ODAY’S car manufacturers integrate more and more microcontroller-based electronic component units (ECUs) in their cars in order to improve safety, performance, comfort, and to meet the more and more restrictive pollution standards. These ECUs are interconnected using different networks such as Controller Area network (CAN) [1], [2], Vehicle Area Network (VAN) [3], [4], Time Triggered Protocol (TTP) [5], J1850 [6], etc. Since real-time and dependability constraints are om- nipresent in such a distributed computer control system, main problems for a designer of new architectures are their validation (meeting of the constraints) and optimization according to specified criteria. This task is traditionally ensured by taking measurements on a prototype. However, with the continuing shortening of time to market and the determination to reduce the design costs, a software-based validation tool is preferable. The first way of doing this is analytically, but this means one should be able to establish an analytical model [7], [8]. Manuscript received April 5, 2001; revised July 13, 2001, September 10, 2001, and November 5, 2001. Abstract published on the Internet September 13, 2002. This work was supported by the French National Research program PREDIT 2 (1998–2000) under Grant 033. Partners are: LORIA, LIP6, and PSA Peugeot-Citröen. P. Castelpietra, Y.-Q. Song, and F. Simonot-Lion are with the LORIA French Research Laboratory in Computer Science, F-54516 Vandoeuvre-lès-Nancy, France (e-mail: castelpi@loria.fr; song@loria.fr; simonot@loria.fr). M. Attia is with the DRIA/SARA, PSA Peugeot-Citroën, F-78943 Vélizy- Villacoublay, France (e-mail: ma22@psinet.fr). Digital Object Identifier 10.1109/TIE.2002.804972 Considering the complexity of a real embedded system, such a model will necessarily be strongly simplified and will only provide oversized solutions. For instance, the holistic sched- uling approach as well as VOLCANO tool [9] introduced by Tindell and Clark [10] only allows us to evaluate the worst case end-to-end response time of distributed periodic tasks. Using this holistic scheduling approach, Song et al. [11] studied the end-to-end task response time for an architecture composed of several ECUs, interconnected by CAN. The same system using simulation-based analysis has also been studied giving more re- alistic performance measures [12]. A priori simulation approach, adopted in the CAROSSE project, is a validation technique complementary to the analytic one. The direct use of the general-purpose simulation platform (SES Workbench, 1 for example) is not judged suitable by in-vehicle embedded system designers since too much effort must be made in building the simulation model. In collaboration with PSA Peugoet-Citroën (carmaker) we specified a modeling methodology based on the modularity, where a certain number of basic components, their behavior and their interactions were defined. This methodology has been implemented through the development of a simulation tool called Carosse-Perf. It is composed, on the one hand, of a library of prebuilt component models for the SES Workbench simulation platform and, on the other hand, of a constructor that uses these models to obtain the whole model that will be simulated. This tool allows designers to easily build a simulation model of their new in-vehicle embedded systems and then to validate them. The validation is performed by verifying the meeting of the different time constraints. Although the approach is quite general, the module decompo- sition is specific to embedded architectures where the behavior of the software applications is assumed to be well known. This is why the Carosse-Perf tool can only be used to model and sim- ulate embedded applications. The remainder of this paper is organized as follows. Section II details the CAROSSE modular approach and Section III describes the Carosse-Perf tool. Section IV presents a case study, where the model is automatically generated using the Carosse-Perf tool. Finally, we conclude in Section V. II. CAROSSE METHODOLOGY We call operational architecture the result of the mapping of one software architecture onto one hardware architecture 1 SES Workbench is a product of HyPerformix Inc. (http://www.hyper- formix.com). 0278-0046/02$17.00 © 2002 IEEE