Experimental assessment of timing verification techniques for AFDX

International audienceAvionics systems distributed on AFDX networks are subject to stringent real-time constraints that require the system designer to employ techniques and tools to guarantee the worst case traversal time of the network (WCTT) and thus ensure a correct global real-time behavior of the distributed functions. The Network Calculus (NC) is an active research area based on the (min,+) algebra, that has been developed to compute such guaranteed bounds, and has been for instance successfully used to certify the Airbus A380 AFDX backbone. Over the years, a number of traffic models and verification algorithms have been developed and integrated into the NC theory, and there are now many possibilities to choose from in the NC framework, each offering a specific trade-off with regard to accuracy (tightness of the bounds), computation time (e.g., linear or exponential complexity) and complexity of the code. Different techniques are often best suited at the different phases of the development cycle: research on NC theory, preliminary feasibility assessment, design space exploration, certification, etc. The objective of the paper is to provide an experimental assessment of the performances of different verification techniques on hundreds of realistic networks randomly generated, where in previous studies comparisons were done only on few examples. The software toolset used in this study is the freely available AFDX benchmark generator NETAIRBENCH and the RTaW-Pegase timing verification software