Managing Best Eeort Traac: Red

Best eeort traac in the Internet is managed by end to end congestion control mechanisms whose limitations are known. The continuous Internet growth leads to more and more applications that don't \play the rules" of the end-to-end congestion control scheme. Simple mechanisms inside the routers, refered as random early detection (RED), provide a more eecient congestion control. It can also be used to protect fair applications from unfair ones. RED can be deployed progressively as it does not require any modiication of the applications using the classical Internet congestion control scheme. The Internet has been primarily designed and will probably be used in the forthcoming years to carry best eeort traac. To allow eecient managing of best eeort traac, congestion control mechanisms have been deened. These mechanisms apply on a end-to-end basis, and are implemented in the transport protocol, TCP. Among them is the well-known additive increase, multiplicative decrease 5]. First, we describe two aspects of problems that exist with the TCP congestion control scheme (its imperfection and the impact of possibly non compliant sources) and present the approaches that are proposed to solve them. We then show how the RED mechanisms can achieve a much more eecient congestion control. We discuss the way to used them (in association with simple scheduling techniques) to prevent non compliant sources to deprive compliant ones. Finally, we discuss the advantages and drawbacks of using explicit congestion notiication instead of packet drops.