Loevenich, Johannes F.Johannes F.LoevenichSergeev, AleksandrAleksandrSergeevLopes Rettore, Paulo HenriquePaulo HenriqueLopes RettoreLopes, Roberto Rigolin F.Roberto Rigolin F.Lopes2022-09-222022-09-222022https://publica.fraunhofer.de/handle/publica/42603110.1109/DRCN53993.2022.97580212-s2.0-85129673835The faster a tactical system can resume IP data flows after unplanned link disconnections, the more robust the system is to link disconnections. Therefore, this paper introduces an intelligent model composed of two agents to quantify the Time to Resume (TTR) IP data flows after unplanned radio link disconnections. The communication scenario assumes two or more mobile nodes connected via unreliable radio links at the edge of tactical networks. The mobile nodes are hosting robust tactical systems implementing cross-layer control loops, using proactive routing protocols, and so on. Given a system configuration, we start with the hypothesis that the TTR converges in distribution to a real-valued cumulative distribution function, thus quantifying the system robustness. Our hypothesis was verified by a series of experiments automated by the agents and executed in a VHF network composed of real military radios. We used our model to quantify the robustness of three IP data flows, namely, unicast, broadcast, and overlay. Thus, a comparative study of two overlay configurations was conducted, with unicast and broadcast as the baselines. The experimental results suggest that our model converges in distribution with respect to the TTR within a given confidence interval.enLink DisconnectionMulti-Agent SystemsReinforcement LearningSystem RobustnessTactical Networksconference paper