Simulation of Adversarial Scenarios in OMNeT++ - Putting Adversarial Queueing Theory from Its Head to Feet

Abstract

Adversarial models of traffic generation replace probabilistic assumptions by considering the deterministic worst-case. The framework of adversarial queueing theory (AQT) has discovered unexpected results on the stability of networks and has seen continuous research efforts over more than 15 years. So far, almost all AQT results have been derived analytically under simplifying but arguably harmless assumptions. However, as can be observed from recent work in AQT, the adversarial scenarios, in particular those that demonstrate instability, become more and more contrived and complex, thus lending themselves less and less to analytical tractability. While simulation seems like a good match for this problem, no available simulation model includes adversarial traffic generation. In this work, we introduce an OMNeT++ simulation framework for AQT as a tool to facilitate the study and development of instability examples. We validate the usefulness of AQT simulations in several use cases and, en-passant, discover some new insights into adversarial effects.

On the relevance of adversarial queueing theory in practice

“…Previous work has introduced transformations of adversaries, so that this requirement can always be satisfied [3,6]. Additionally, simulation results on the minimal size of initial configurations required for classical adversaries support the claim that already a small number of packets suffices [4].…”

“…Simulation has been employed to study adversarial scenarios before, most notably in [9], [11], and recently in [4], which all study unbounded queues. In contrast, our work is focused on quantifying loss in a finite buffer setting.…”

Maria Serna’s contributions to adversarial queuing theory

“…Returning to wired networks, another line of research has explored the relation between the theoretical results in aqt and real networks. For instance, Berger et al [17] introduced aqt elements in the OMNeT++ simulator, and were able to reproduce by simulation some of the analytical instability results presented in Section 3. They found interesting behaviors not present in the analysis, like the interplay among the initial system state, the injection rate, and some added randomness.…”