An evaluation of fair packet schedulers using a novel measure of instantaneous fairness Hongyuan Shi, Harish Sethu * , Salil S. Kanhere Department of Electrical and Computer Engineering, Drexel University, 3141 Chestnut Street, Philadelphia, PA 19104, USA Received 13 October 2003; revised 28 March 2005; accepted 5 April 2005 Available online 5 May 2005 Abstract A number of emerging Internet applications such as video-conferencing and live multimedia broadcasts from Internet TV stations rely on scheduling algorithms in switches and routers to guarantee performance and an acceptable level of quality of service. Fairness in packet schedulers is an intuitively desirable property with practical value; fair schedulers such as weighted fair queueing are a critical component of quality-of-service mechanisms that seek to guarantee end-to-end delay bounds, and thus provide end-to-end service differentiation. Popular measures of the fairness achieved by packet schedulers are based on bounds, such as the relative fairness bound which captures the maximum possible difference between the normalized services received by any two flows. In this paper, we argue that such measures do not capture the actual fairness achieved at most instants of time, and therefore, do not represent a true measure of the ability of a scheduler to successfully deliver end-to-end quality for real-time applications. In this paper, we borrow from the field of economics and propose a new measure of fairness based on the Gini index. This measure captures the instantaneous fairness of a scheduler and, unlike other measures based on bounds, also captures the fairness of the scheduler in its handling of flows during idle periods. We use this measure on real gateway traffic traces and present a simulation-based evaluation of several well-known timestamp-based and frame-based schedulers. We also present a qualitative analysis of the phenomena underlying the observed results. q 2005 Elsevier B.V. All rights reserved. Keywords: Resource allocation; Fair scheduling; Fairness; Max–min fairness; WFQ 1. Introduction A number of emerging Internet multimedia applications will rely on scheduling algorithms in switches and routers to guarantee performance and an acceptable level of quality of service. In a DiffServ framework with heterogeneous service classes, the scheduling algorithm achieves the desired quality of service for each class by determining the exact sequence in which packets should be transmitted. An intuitively desirable and also a practically important property of a packet scheduler is the fairness it achieves in the allocation of the bandwidth resources on the output link. A fair scheduler may also be used to provide strict delay guarantees required for interactive applications. Even with the over-provisioning of resources such as is typical in the Internet core, fairness in scheduling is essential to protect flows from other misbehaving flows triggered by deliberate misuse or malfunctioning software on routers or end- systems. Fair scheduling becomes especially critical in access networks, within metropolitan area networks and in wireless networks where the resource capacity constraints tend to be significantly limiting to high-bandwidth multi- media applications today. Fair schedulers have now found widespread implementation in switches and Internet routers [1,2]. Fair schedulers attempt to allocate bandwidth resources by a certain allocation scheme based on a particular notion of fairness. The most commonly used notion of fairness is max–min fairness based on the intuition that no flow should receive more service than its demand and that no flow with an unsatisfied demand should receive less service than any other flow [3,4]. The generalized processor sharing (GPS) scheduler is proposed as an ideally fair but unimplementable scheduler that exactly achieves the goal of this notion of fairness [5]. During each infinitesimal interval of time, Computer Communications 28 (2005) 1925–1937 www.elsevier.com/locate/comcom 0140-3664/$ - see front matter q 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.comcom.2005.04.002 * Corresponding author. Tel.: C1 215 895 5876; fax: C1 215 895 1695. E-mail addresses: shi@ece.drexel.edu (H. Shi), sethu@ece.drexel.edu (H. Sethu), salil@ece.drexel.edu (S.S. Kanhere).