Abstract

In this paper, we discuss employing a server in a router, the architecture of which is based on policy, for congestion management of IP traffic in Diffserv networks. We exploit the congestion information and employ that information to manipulate the network level parameters to adapt the traffic according to the current congestion state of the network. The congestion monitors interact with the policy server, which, depending on the network state, decides the policy/rules that should be enforced by the Diffserv network. The implementation of the selected policies typically leads to accepting, remarking or dropping the IP traffic packets entering the network. We have also indicated how these modules could functionally be used to generate and store appropriate rules based on policies for congestion management. Since the system will now be rule based, there is a need for policing action to monitor traffic and enforce these rules when congestion sets in, and we propose mobile agents accomplishing this job. Index Terms: Quality of service, IP Networks Congestion Management, Self-similar input, Policy based rules, Mobile Agents.

Abstract

In an IP network, if the source rates are increased beyond the service rates of the routers, then queues of packets waiting to be routed at the buffers, build up and exceed the buffering capacity of these routers leading to packets getting dropped. This results in low throughput and congestion collapse. In such networks, an AQM mechanism manages queue lengths in buffers and enables the end-systems to react to such losses by reducing their packet rate, avoiding severe congestion. Random Early Detection (RED) is one of the first AQM mechanisms to be used to avoid congestion in this manner. In this paper, the existing Normal and Gentle RED algorithms of Floyd as well as the justification for the proposed modified exponential RED algorithm have been discussed along with the results obtained on the functioning of the algorithms. Functioning of the algorithm proposed has also been tested using ns2 Simulator.