A Distributed Mechanism for Identification and Discrimination of non-TCP-friendly Flows in the Internet
Résumé
This paper proposes the MUV (Misbehaving User Vanguard) algorithm for identification and discrimination of non TCP-friendly best-effort flows. The operational principle of MUV is to detect non TCP-friendly flows at the ingress router by comparing arrival rates to equivalent TCP-friendly rates, i.e. the arrival rate of a TCP flow having the same round-trip time and packet-loss probability. If a flow is identified as non TCP-friendly, its packets are marked as “unfriendly”. Core routers discriminate packets marked as unfriendly with RED-based drop-preference mechanisms. In order to measure the round-trip time and the packet-loss probability for the computation of a flow’s TCP-friendly rate, ingress router and egress router communicate via a simple protocol.
The MUV algorithm fits into the Differentiated Services Architecture of the Internet and can be considered scalable as it only requires per-flow state at ingress- and egress routers. We show by simulation that MUV is able to reliably identify and discriminate unresponsive flows and investigate its performance bounds regarding the identification of flows using non TCP-friendly congestion control algorithms.