Fair bandwidth sharing among adaptive and non-adaptive flows in the Internet

Abstract
The problem of fair bandwidth sharing among adaptive (TCP) and non-adaptive (i.e. CBR-UDP) flows at an Internet gateway is considered. An algorithm that drops packet preventively, in an attempt to actively penalize the non-adaptive traffic that attempts to "steal" buffer space, and therefore bandwidth from the adaptive traffic flows, is presented. The algorithm maintains minimal flow state information and is therefore scalable. The performance of the algorithm is compared with other gateway algorithms and it is shown that, in the presence of non-adaptive traffic, it achieves a more balanced bandwidth allocation among the different flows. The behavior of a flow subjected to the given algorithm has also been analysed in detail.

This publication has 7 references indexed in Scilit: