Rethinking Wireless MAC Architecture for Quality of Service Support – Design and Implementation Ajit Warrier and Injong Rhee Department of Computer Science North Carolina State University Abstract—Existing wireless MAC standards are known to have fairness problems. In this paper, we ask a more fundamental question: should fairness be a property of medium access, or in other words, are we placing the blame on the right place? We argue that fairness is a property of applications and must be implemented by applications. Medium access control should simply provide mechanisms for application-defined fairness poli- cies to be implemented effectively and efficiently. We argue that these mechanisms must provide separate control knobs for two main functions of MAC, namely, fairness control and contention resolution, in order to enable their independent evolution. Furthermore, they must be lightweight and their behaviors must be predictable to enable efficient and consistent implementations of fairness policies on top of these mechanisms. Existing MAC standards do not follow these guidelines. In this paper, we design a new MAC protocol, called Siren, adapted from an existing MAC protocol using these guidelines, and implement Siren in a real multi-hop wireless network. The efficacy of our design is demonstrated by implementing on top of Siren various popular fairness policies such as static priority, fair time sharing, proportional rate allocation, earliest deadline first (EDF) and proportional fairness and measuring their performance in our network testbed. I. I NTRODUCTION One of the frequently cited problems of current wireless networks is related to lack of fairness. The reported fairness problems of MAC include, to name a few, starvation [24], [11], [29], priority inversion [30], inequitable allocation of bandwidth [29], lack of QoS support [18] and multi-rate LAN unfairness [14]. The research community has come up with several solutions [24], [30], [18] to such fairness problems. The common thread among such solutions is that they often require altering the MAC protocol significantly to achieve the authors’ notion of “fairness”, in effect embedding a fairness policy into the MAC. Clearly, this makes them “point solutions”, wherein the proposed changes to the MAC for one solution are incompatible with those for other solutions. The underlying reason for this is the MAC-centric notion of fairness prevalent in the research community today. Evidence for this can also be seen in the design of several existing QoS-enhanced MAC protocols such as IEEE 802.11e and HIPERLAN/1 [5]. These MAC protocols either explicitly or implicitly implement some notions of fairness defined by their designers. For instance, IEEE 802.11e implicitly supports a fairness notion similar to proportional rate allocation among different priority classes where the exact allocation among the classes are hard to tract due to interaction with contention resolution (CR) and window backoffs, and HIPERLAN/1 explicitly supports a policy similar to EDF (earliest deadline first). The solutions to fix the fairness problems of existing MACs (especially IEEE 802.11) have the same problems. For instance, [15], [13], [28], [31] embed their solutions into MAC directly so that MAC is tuned to support only the notion of fairness that the authors define. Independent of whether such policies are adopted by de- signers explicitly or purely by chance due to some design and implementation artifacts, a MAC protocol embedded with a predefined notion of fairness makes it very difficult to support diverse fairness policies that application utility functions dic- tate. Furthermore, an embedded fairness policy which suits some applications may be unfair for different applications. These two observations, namely (a) the ever-growing list of point solutions to enforce different fairness policies requiring MAC protocol changes and (b) the difficulty in enforcing a fairness policy in a MAC which already has an embedded notion of fairness motivate us to take an altogether different approach to the problem of MAC architecture design. In this paper, we outline important design principles for a flexible and efficient MAC architecture designed to support application- level fairness policies; whatever those policies may be. In order to demonstrate that these principles are realizable and effective using off-the-shelf radios, we propose and implement a new MAC protocol called Siren on the MicaZ sensor radio chip interface. We then show the versatility of Siren by implementing various fairness policies including static priority, fair time sharing, EDF and proportional fairness on top of Siren in our network testbed. In particular, the proportional fairness in multi-hop networks has never been implemented in a real network. We provide its first practical implementation. Our main contributions in this paper are: 1) Identifying design principles for a flexible and efficient MAC architecture. 2) Proposing of a MAC architecture, Siren, which follows these principles. 3) Developing algorithms for the implementation of several different fairness policies on top of Siren, demonstrating its flexibility and efficiency. 4) Implementing Siren and these fairness policies on a conventional CSMA-based radio – the CC2420 and experimental results on a 30 node multi-hop sensor testbed. The rest of the paper is structured as follows: Section II