Promoting self-imposed end-to-end congestion control via a Sword of Damocles approach

Abstract

As the Internet evolves, traffic management has become a high-priority problem. Without a network centric traffic management scheme, severe fairness problems can arise. For instance, unfairness occurs with TCP traffic in the presence of careless or subversive misbehaving flows, or even in the presence of other TCP flavors if we consider inter-protocol unfairness. In this paper we propose a new AQM scheme, Stochastic Fair Traffic Management (SiFTM1), to help achieve fair bandwidth allocation among flows even in the presence of misbehaving ones. SiFTM relies on multi-level hashing to perform packet classification and on a random sampling mechanism to identify and punish misbehaving flows harshly. SiFTM improves the fairness index among TCP flows even with different TCP flavors. Through simulations we show that SiFTM outperforms other related schemes, especially in the presence of misbehaving flows. Furthermore, we argue that with the fair bandwidth allocation it provides, SiFTM deprives end systems of the incentive to behave unresponsively as a means to enjoying better services. It thus promotes rational behavior and self imposed network-friendly congestion control in the transport protocol.