We analyze the fairness of decentralized, randomized MAC protocols and introduce two metrics for measuring short-term fairness. Unlike past efforts tha t studied fairness over long t ime scales, we focus on the distribution of accesses of different s tat ions to a shared channel over short t ime scales. Our first metric uses a slidingwindow analysis of packet traces, while the second is an analytical approach tha t uses the renewal reward theory of Markov chains to model the MAC protocol. • To see why shor t term fairness is impor tant , we examined typical packet sequence traces of multiple stations contending for a wireless channel using Lucent WaveLAN's CSMA/CA (Carrier Sense Multiple Access/ Collision Avoidance) protocol. We observed tha t each stat ion tends to monopolize the channel for several packet transmission times before it is captured by the other station, which in turn monopolizes the channel for a long time, even though at each point in t ime the other stat ion has packets waiting to be sent. An example of such a trace can be found in [Koksal00]. This behavior greatly degrades the performance of applications and t ranspor t protocols. For example, realt ime audio and video are delay-sensitive and perform be t te r when j i t ter is low. T C P performance degrades greatly when MAC protocols exhibit shor t term unfairness, as shown in Fig. 1. Here, acknowledgments (ACKs) sent from the receiver do not reach the sender in a t imely manner , resulting in large transmission bursts. Thus we observe a severe form of ACK compression. We analyze the reasons for such behavior by certain MAC protocols and formalize the notion of shor t term unfairness. We build on the intuition tha t a shor t term
No takes yet. Share an insight, caveat, or question.
Koksal et al. (2000) studied this question.