Fairness Analysis of IEEE 802.11 Multirate Wireless LANs

Abstract

This paper investigates the issue of fairness in IEEE 802.11-based multirate wireless local area networks (WLANs). Distributed coordination function, which is the medium-access control (MAC) protocol used in 802.11 WLANs, provides equal long-term channel access probability to competing stations, irrespective of the time required in sending a frame. When equal-sized frames are used and channel conditions are similar, each station, regardless of its data rate, achieves the same throughput. Furthermore, the aggregate throughput is reduced to a level much closer to what one gets when all stations are of lower rate. This anomaly in the performance is a result of unfair channel time allocation for stations when they use multiple data rates. We consider provisioning of time-based fairness in which each station receives an equal share of the wireless channel occupancy time. We demonstrate that under time-based fairness, the throughput that a tagged node achieves in a multirate WLAN with nodes is identical to what it would achieve in a single-rate WLAN with nodes all at the same data rate as the tagged node. Furthermore, we show that under time-based fairness scheme, the ratio of throughputs per station corresponding to two different bit rates is directly proportional to the ratio of their bit rates. We analyze different mechanisms in achieving time-based fairness by using an analytical model. Using Jain's fairness index, optimal MAC parameters required in achieving maximum fairness between slow and fast stations are obtained. The impacts of these mechanisms on throughput of slow and fast stations are explored. We also consider the notion of proportional fairness in a multirate scenario and prove that it is equivalent to fair channel time allocation. Last, our investigation of an alternative fairness criterion also leads us to propose that the IEEE 802.11 MAC protocol should be redesigned with temporal fairness as a design objective in avoiding inefficiencies related to the performance anomaly.