Early Regulation of Unresponsive Flows

The increasing deployment of unresponsive traffic can cause serious problems onthe Internet. By unresponsive traffic, we mean flows that do not reduce theirsending rate in response to congestion. The problems include unfairnessagainst competing traffic that employ congestion control and even congestioncollapse. In this thesis work, we propose router mechanisms to regulateunresponsive best-effort traffic. The goal of the proposed mechanisms is todrop undeliverable packets - packets that are dropped somewhere in the networkbefore they reach their destination - as close to the periphery of the networkas possible. The key ideas of our approach are: (1) edge routers keep trackof incoming flows and their arrival rates; (2) core routers use RED (RandomEarly Detection) for queue management and generate rate-limited source quencheson packet drops to advice sources to reduce their sending rates; and (3) edgerouters snoop on source quenches passing through them and use them to controlper-flow regulators. Regulators adjust their maximum sending rate using amultiplicative-decrease, additive-increase discipline. A decrease is triggeredby the arrival of a source quench; an increase is triggered for every estimatedround-trip time period. We examine the impact of these mechanisms for avariety of simulated network topologies and traffic patterns.