In systems which serve many users there is a need to respect some fairness rules while looking for the overall efficiency. This applies among others to network design where a central issue is how to allocate bandwidth to flows efficiently and fairly. The so-called max-min fairness is widely used to meet these goals. However, allocating the bandwidth to optimize the worst performance may cause a large worsening of the overall throughput of the network. In this paper we show how the concepts of mult-criteria equitable optimization can effectively be used to generate various fair and efficient allocation schemes. We introduce a multi-criteria model equivalent to equitable optimization and we develop a corresponding reference point procedure to generate fair and efficient bandwidth allocations. Our analysis is focused on the nominal network design for elastic traffic that is currently the most significant traffic of IP networks. The procedure is tested on a sample network dimensioning problem for elastic traffic and its abilities to model various preferences are demonstrated. ᭧

Abstract-Expanding demand on the Internet services leads to an increased role of the network dimensioning problem for elastic traffic where one needs to allocate bandwidth to maximize service flows and simultaneously to reach a fair treatment of all the elastic services. Thus, both the overall efficiency (throughput) and the fairness (equity) among various services are important. The Max-Min Fairness (MMF) approach, widely used to this problem, guarantees fairness but may lead to significant losses in the overall throughput of the network. In this paper we show how the concepts of multiple criteria equitable optimization can be effectively used to generate various fair resource allocation schemes. We introduce a multiple target model equivalent to equitable optimization and we develop a corresponding procedure to generate fair efficient bandwidth allocations. The procedure is tested on a sample network dimensioning problem and its abilities to model various preferences are demonstrated.