Fairness of traffic controls for inelastic flows in the Internet

Dah Ming Chiu, Adrian Sai-Wah Tam
2007 Computer Networks  
In best-effort networks, fairness has been used as a criterion to guide the design of traffic controls. The notion of fairness has evolved over time, from simple equality to a form of equality modulated by the user's need (e.g. max-min and proportional fairness). However, fairness has always been defined on a per-user basis for a deterministic workload. In this paper, we argue that we must redefine the notion of fairness when we study traffic controls for the co-existence of elastic and
more » ... c traffics. Our results indicate that subjecting inelastic flows to fairness congestion control on a per-flow basis does not necessarily maximize the network's utility. Instead, inelastic flows may follow their own form of traffic control, such as admission control (without congestion control). At the aggregate level, our results indicate that it still makes sense to maintain a balance between elastic and inelastic traffic. In order to support our arguments, we develop a methodology for comparing different traffic controls for given utility functions and different workloads, both deterministic and stochastic. Index Terms-congestion control, admission control, fairness, utility maximization, non-convex utility function, stochastic traffic model friendly (like) controls, for the deterministic workload case in section VI and stochastic workload in section VII. Finally, the significance of this work and future directions are discussed in the concluding section. II. REVIEW OF PREVIOUS WORK Actually, some of the basic ideas we are espousing have been discussed at some length in a seminal paper by Shenker ten years ago [5]. Shenker pointed out the obvious benefits in merging different types of networks (i.e. data, voice and TV) into a single network. While over-provisioning can always satisfy the needs of such an integrated network, Shenker argued that it is more effective to introduce multiple services into the network to support the different applications. Most importantly, to measure how good a network is, he introduced the notion of utility maximization. Shenker elaborated on some different forms of utility function for different types of (e.g. elastic and inelastic) flows, and applied utility maximization to justify the role for admission control in the situation when the utility function is non-concave. Although [5] was written in the context of advocating Integrated Service [2], which pre-dated subsequent work on end-system admission control [12] and TCP-friendly congestion control [6], its analysis and discussion of network design goals, viz. the utility optimization framework, still applies to end-system based traffic controls. Before the utility maximization formulation, various performance metrics had been adopted in studying network traffic control algorithms. In addition to the obvious goals of maximizing the throughput and minimizing the delay, fairness was also adopted as an important goal [13], [14]. In a simple setting of a fixed number of flows sharing a common bottleneck, the simple notion of fairness corresponding to dividing bandwidth equally among competing flows seems particularly appealing. Both distributed algorithms (e.g. AIMD [13]) and centralized algorithms (e.g. fair queueing [15]) were proposed to implement fair traffic control. The emphasis of equality was also extended to the case when not all flows share the same path. In a general network topology with arbitrary flows, fair bandwidth allocation is first applied to the most limiting bottleneck, and iteratively to all bottlenecks, leading to the definition of max-min fairness. The theory of network utility maximization blossomed when Kelly et al applied it to create a fluid model of the Internet D
doi:10.1016/j.comnet.2006.12.006 fatcat:2pxli7mz45e2pisejjgwg5eisi