The Trajectory approach for AFDX FIFO networks revisited and corrected

We consider the problem of dimensioning real-time AFDX FIFO networks with a worst-case end-to-end delay analysis. The state-of-the-art has considered several approaches to compute these worst-case end-to-end delays. Among them, the Trajectory approach has received more attention as it has been shown to provide tight end-to-end delay upper bounds. Recently, it has been proved that current Trajectory analysis can be optimistic for some corner cases, leading in its current form, to certification issues. In this paper, we first characterize the source of optimism in the Trajectory approach on detailed examples. Then, we provide a correction to the identified problems. Two problems are solved: the first one is on the root cause of the underestimated time interval to compute delays of competing flows and a problem in the definition of the end-to-end delay computation. The second one is on the way that serialized frames are taken into account in the worst-case delay analysis.