With increasingly widespread use of computer networks, and the use of varied technology for the interconnection of computers, congestion is a significant problem. In this report, we refine a scheme for congestion avoidance in networks using a connectionless protocol at a network layer.

In the part II of this report series, we proposed a scheme for congestion avoidance using a feedback from the network. The users of the network respond to to the feedback by reducing the amount of traffic injected into the network. The assumption made in part II was that all the users shared the same set of network resources. As a result, the simplified fairness criterion that we used was that all the users have the same window size.

In this report, we relax the assumption of having the same set of resources shared by the different users of the network. Users may share arbitrary sets of resources of the network. We define a more general fairness goal to achieve in the light of the relaxation of the assumption. We present a solution alternative to achieve this goal, by having the routers in the network selectively feedback the congestion information only to those users that are using more than their fair share of each individual router's resources.

We continue to maintain the need for the scheme to be distributed and adapt to the dynamic state of the network. The scheme converges to the 'optimal' operating point and is simple to implement. We also address the performance of the scheme under the transient changes in the network, and for a variety of other network conditions.

This is part IV of a series of reports on our work on congestion avoidance for connectionless networks.

Complete Paper in Adobe acrobat format (1,640,653 bytes)