Adaptive Multi-level Explicit Congestion

Adaptive Multi-level Explicit Congestion

Abstract

In this paper we present to you, an extended Multi-level ECN, a new TCP congestion scheme, a concept proposed by Mukundan Sridharan, Arjan Durresi and Raj Jain of The Ohio State University, Columbus, Ohio. The Multi-level Explicit Congestion Notification (MECN) algorithm allows network operators to achieve high throughput with corresponding low delays. But MECN average queue, is sensitive to its parameter settings and its level of congestion, hence no guarantees can be given about delay. Delay being a major component of the quality of service, network operators would naturally like to have a rough estimate of the average delays in their congested routers. To achieve predictable average delays with MECN would require constant tuning of the parameters to adjust to current traffic conditions. The goal of this paper is to present a solution to the parameter tuning problem of the MECN. We compare the performance of the Adaptive MECN system with the Adaptive RED system using simulations on the NS-2 simulator. Based on simulations we find that Adaptive MECN performs better than Adaptive RED.

INTRODUCTION

End-to-end congestion control schemes continue to be one of the main determinants of the robustness of the Internet. Congestion remains the main obstacle to Quality of Service (QoS) on the Internet. Although a number of schemes have been proposed for network congestion control, the search for new schemes continues. [5] gives a survey of different congestion control schemes. But the winner for the time being seems to be RED/ECN class of algorithms. ECN was made a standard by the IETF in 2001 [6]. Hence it becomes imperative that the possibilities of utilizing the ECN framework to the fullest are explored. [1] proposed a new scheme called the Multi-level Explicit Congestion Notification (MECN), which works with the frame work of ECN, but uses the two bits allocated for ECN, in the IP to indicate four different levels of congestion, to the source. But just like RED [7], MECN's average queue is also sensitive to parameter settings and the level of congestion. This average queuing delay is a very important for QoS applications. So setting the parameters of MECN is very important and to maintain a constant delay at the routers is a must, to give any QoS guarantees to the end users. In this paper a proposal is placed for an Adaptive version of MECN, which sets its parameters automatically and adapts its maximum marking probability to main a constant queuing delay. We compare the performance of AMECN, with ARED and MECN and show that it performs better than the other schemes. In Section II, we give a brief introduction to the MECN protocol. In Section III, we present to you the concept of Wireless MCN (WMCN). In Section IV, we introduce the Adaptive Multilevel ECN protocol and site some guidelines on setting the parameters. In Section V, we present the conclusions.

Download full seminar papers At
http://www.enjineer.com/forum