throughput or fairness. An additional important consider ation is that in campus networks a user can obtain service from any of the APs in transmission range. Therefore, any design may need to carefully consider issues such as load- 85EE balancing of users across APs along with power control In chaotic networks, the infrastructure is controlled by multiple organizations, and, unfortunately, their priorities often conflict. For example, for a home network consisting of a single AP, the best strategy is to always transmit at maximum power, and there is no incentive to reduce power and, thus, interference. The results in the previous section show that such a "Max Power"strategy, when employed by multiple APs, will result in suboptimal network perfor- mance. This implies that while a single node can improve its Tx power(dBm) performance by increasing power, it can actually obtain bet- ter performance if it, and all of its neighbors, act socially and Figure 10: Minimum AP distance vs. Tx powe reduce their transmission power appropriately. This is anal (client= 10m gous to the tradeoffs between selfish and social congestion control in the Internet [10. While a node can improve per- formance by transmitting more quickly in the Internet, this the minimum distance between APs that can be support ga We draw a few key conclusions from Figure 10. Clear can result in congestion collapse and degraded performance for all. We believe that similar factors that drove the wide num supported density increases) dra- deployment of congestion control algorithms will drive the atically as the transmission power(in dBm) is decreased deployment of power control algorithms. We should note We also find that high aP density and higher loads require that an added side incentive for the deployment of auto- transmit power levels below 0 dBm. This is the lowest trans- matic power control is that it limits the propagation of an APs transmission which, in turn, limits the opportunity of mit power available from commercial hardware that we are malicious users eavesdropping on any transmission. aware of. Adding support for lower transmit power levels to wireless hardware would be a simple way of improving the Our work focuses on socially responsible power control al tensity of APs that can be supported. Secondly, the graph gorithms that would work well in chaotic environments. We an also be used to determine the upper bound on the power call such power control algorithms "socially responsible"to level that should be employed(x-axis) in order to achieve differentiate them from approaches that require global coor- dination across multiple access points(e. g for campus-wide a certain throughput, given a certain inter-AP distance(y- wireless networks). Our algorithms are targeted at individ- axis). Using a higher power level will typically not affect (i.e not decrease or increase) the performance for that node, but ual access points and clients, which behave in an altruistic will reduce performance for other, nearby nodes. This is the manner. agnostic to the actions of other aps and clients. Our algorithms could also work in campus scenarios. How- the next section. Finally we note that the highest densi- ever, we do not consider issues such a AP load-balancing ties require the use of very low transmission power, forcing which arise in such environments. We leave the extension of nodes to use a transmission rate under 11 Mbps. This sug our design to campus deployments for future work. gests that, when their traffic requirements are low, it may Note that while our algorithms are targeted at nodes be- advantageous if nodes voluntarily reduce not only their having in an altruistic manner, there are also practical con- transmission power but also their transmission rate since it siderations that make them more feasible than simply re- could increase the overall network capacity in very dense lying on the altruism of end users would. In particular networks, We will revisit this issue in Section 9 these algorithms are implemented not by end users, but by equipment vendors. From an equipment vendor's point-of view, reducing interference is beneficial. Moreover, regula- 6. DEPLOYMENT CHALLENGES tory mandates already limit transmit power, and could be Power control offers a simple but powerful technique for extended to require dynamic adjustment of transmit power reducing interference. The tradeoffs are obvious: reduc- in order to increase spatial reuse and potentially allow for ing the power on a channel can improve performance for higher transmit power limits which would clearly benefit ther channels by reducing interference, but it can reduce both end users and equipment vendors. Finally, as dis- he throughput of the channel by forcing the transmitter to cussed in Section 3.2, we find that new technology is quickly use a lower rate to deal with the reduced signal-to-noise ra adopted in chaotic networks, and that many users in chaotic tio. As a result, we must carefully consider the incentives dor implemented intelligent transmit power control could&. networks do not change factory default settings. Hence, ven that users may have for using such techniques. In practice, deployed relatively quickly and would be widely adopted the incentives for using power control are complex and we have to distinguish between the techniques that are applica- ble to campus deployments and chaotic wireless networks. In campus environments, there are a number of APs un 7. TRANSMISSION POWER AND der the control of a single organization. This organization RATE SELECTION in a position to do power control in each cell in a way that ow power adaptation affects both optimizes some global network metric, e. g. total network network-wide and individual user throughput, we ran exper-1 10 100 1000 -20 -18 -16 -14 -12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20 Tx power (dBm) minimum AP distance (meters) 0.1 0.3 0.5 0.7 0.9 1.1 Load Mbps 2 Mbps 11 Mbps 5.5 Mbps Figure 10: Minimum AP distance vs. Tx power (dclient = 10m) We draw a few key conclusions from Figure 10. Clearly the minimum distance between APs that can be supported decreases (i.e. maximum supported density increases) dra￾matically as the transmission power (in dBm) is decreased. We also find that high AP density and higher loads require transmit power levels below 0 dBm. This is the lowest trans￾mit power available from commercial hardware that we are aware of. Adding support for lower transmit power levels to wireless hardware would be a simple way of improving the density of APs that can be supported. Secondly, the graph can also be used to determine the upper bound on the power level that should be employed (x-axis) in order to achieve a certain throughput, given a certain inter-AP distance (y￾axis). Using a higher power level will typically not affect (i.e. not decrease or increase) the performance for that node, but will reduce performance for other, nearby nodes. This is the basis for one of the power control algorithms discussed in the next section. Finally we note that the highest densi￾ties require the use of very low transmission power, forcing nodes to use a transmission rate under 11 Mbps. This sug￾gests that, when their traffic requirements are low, it may be advantageous if nodes voluntarily reduce not only their transmission power but also their transmission rate since it could increase the overall network capacity in very dense networks. We will revisit this issue in Section 9. 6. DEPLOYMENT CHALLENGES Power control offers a simple but powerful technique for reducing interference. The tradeoffs are obvious: reduc￾ing the power on a channel can improve performance for other channels by reducing interference, but it can reduce the throughput of the channel by forcing the transmitter to use a lower rate to deal with the reduced signal-to-noise ra￾tio. As a result, we must carefully consider the incentives that users may have for using such techniques. In practice, the incentives for using power control are complex and we have to distinguish between the techniques that are applica￾ble to campus deployments and chaotic wireless networks. In campus environments, there are a number of APs un￾der the control of a single organization. This organization is in a position to do power control in each cell in a way that optimizes some global network metric, e.g. total network throughput or fairness. An additional important consider￾ation is that in campus networks a user can obtain service from any of the APs in transmission range. Therefore, any design may need to carefully consider issues such as load￾balancing of users across APs along with power control. In chaotic networks, the infrastructure is controlled by multiple organizations, and, unfortunately, their priorities often conflict. For example, for a home network consisting of a single AP, the best strategy is to always transmit at maximum power, and there is no incentive to reduce power and, thus, interference. The results in the previous section show that such a “Max Power” strategy, when employed by multiple APs, will result in suboptimal network perfor￾mance. This implies that while a single node can improve its performance by increasing power, it can actually obtain bet￾ter performance if it, and all of its neighbors, act socially and reduce their transmission power appropriately. This is anal￾ogous to the tradeoffs between selfish and social congestion control in the Internet [10]. While a node can improve per￾formance by transmitting more quickly in the Internet, this can result in congestion collapse and degraded performance for all. We believe that similar factors that drove the wide deployment of congestion control algorithms will drive the deployment of power control algorithms. We should note that an added side incentive for the deployment of auto￾matic power control is that it limits the propagation of an AP’s transmission which, in turn, limits the opportunity of malicious users eavesdropping on any transmission. Our work focuses on socially responsible power control al￾gorithms that would work well in chaotic environments. We call such power control algorithms “socially responsible” to differentiate them from approaches that require global coor￾dination across multiple access points (e.g., for campus-wide wireless networks). Our algorithms are targeted at individ￾ual access points and clients, which behave in an altruistic manner, agnostic to the actions of other APs and clients. Our algorithms could also work in campus scenarios. How￾ever, we do not consider issues such a AP load-balancing which arise in such environments. We leave the extension of our design to campus deployments for future work. Note that while our algorithms are targeted at nodes be￾having in an altruistic manner, there are also practical con￾siderations that make them more feasible than simply re￾lying on the altruism of end users would. In particular, these algorithms are implemented not by end users, but by equipment vendors. From an equipment vendor’s point-of￾view, reducing interference is beneficial. Moreover, regula￾tory mandates already limit transmit power, and could be extended to require dynamic adjustment of transmit power in order to increase spatial reuse and potentially allow for higher transmit power limits which would clearly benefit both end users and equipment vendors. Finally, as dis￾cussed in Section 3.2, we find that new technology is quickly adopted in chaotic networks, and that many users in chaotic networks do not change factory default settings. Hence, ven￾dor implemented intelligent transmit power control could be deployed relatively quickly and would be widely adopted. 7. TRANSMISSION POWER AND RATE SELECTION In order to characterize how power adaptation affects both network-wide and individual user throughput, we ran exper-