Comparison of MAC protocols for 802.11 based long ...

Comparison of MAC protocols for 802.11 based long distance networks Sandra Salmer´on-Ntutumu, Javier Sim´o-Reigadas

Rabin Patra

Universidad Rey Juan Carlos, Madrid, Spain Email: [email protected]

University of Berkeley California, USA Email: [email protected]

Abstract WiFi based Long Distance (WiLD) networks have been successfully deployed in rural areas of developing countries, thus providing Internet access and VoIP telephony services to previously isolated regions. However, it is well known that the standard CSMA/CA MAC protocol used by WiFi is not well suited for long distances. To achieve better performance, some researchers have proposed replacing the standard CSMA/CA MAC with a TDMAbased MAC, while others have tried to optimize timing parameters in the IEEE 802.11e EDCA standards. This paper is an experimental comparison of the two approaches for point-to-point links. We measure the throughput and delay in various test scenarios by varying the distance of point-to-point links in a controlled environment created using a wireless channel emulator. As far as we know, this is the first experimental comparison between a TDMA-based MAC solution over 802.11 with the IEEE 802.11e EDCA standard optimized for long distances. Finally, we illustrate the tradeoffs between using these two approaches under different conditions.

I. I NTRODUCTION Most people living in rural or remote areas of developing countries around the world have no access to affordable communication solutions. Traditional approaches based on telephone, cellular, satellite or fiber have turned out to be economically unfeasible because of low population density, low demand and lack of infrastructure such as reliable electricity [1]. For example in Africa, even when cellular or satellite coverage is available in rural regions, bandwidth is extremely expensive (e.g. satellite bandwidth is about US$3000/Mbps per month) [2]. Even WiMax [3], which is seen as a potential broadband alternative, requires a sufficient user density to amortize the cost of the base station, is so far too expensive for rural areas. In addition this, most of these solutions focus on licensed spectrum and carrier-based deployment, which limits their usefulness to the kind of ”grass roots” projects typical for developing regions [4]. Due to its cost-effectiveness, IEEE 802.11 [5] (WiFi) technology has seen widespread use beyond its original purpose for short-range indoor networks [6]. WiFi operates in the ISM (Industrial, Scientific, Medical) band which is license-free in many parts of the world, it is already well established, and has achieved low-cost mass production. Despite the promise of WiFi networks as a low-cost solution for long distance point-to-point networks (referred to as WiLD networks or Wireless in Long Distance), the real-world performance of such networks while using the standard 802.11 MAC protocol is often abysmal [7]. For example, on a 60 km point-to-point link in Ghana, with TCP traffic in both directions, we measured a throughput of only 600 Kbps, whereas the raw throughput on the link should have been over 6 Mbps. The main reasons behind this poor performance are certain shortcomings of stock 802.11 protocol that manifest only in WiLD environments. In particular, the 802.11 link-level recovery mechanism causes low utilization and the use of CSMA/CA at long distances results in more frequent collisions. In addition, nodes with multiple radios are faced with inter-link interference. Researchers have proposed two principal approaches to overcome these protocol shortcomings - a) optimizing 802.11 parameters using the extensions proposed in the 802.11e [8], [9] (referred to as WiLD-EDCA) and b) using time division multiplexing [10] (refereed to as WiLD-MAC). The first approach takes advantage of additional configuration parameters available in the new 802.11e EDCA standard to change 802.11’s underlying timing parameters like slot time, acknowledgment timeout and congestion window size to minimize collisions and increase link utilization even at long distances. While this solution substantially

improves throughput and delay over for long distance point-to-point links, it could still potentially suffer from interlink interference in multihop networks. The second approach proposes to replace 802.11’s CSMA/CA based MAC protocol by a time division access based MAC protocol (WiLD-MAC [10] and 2P [11]). While this solution eliminates collisions and achieves substantial throughput improvements for long distance point-to-point links, it suffers from higher delay (from longer time slots) Since it also completely abandons the 802.11 standard, deploying WiLD-MAC requires modifications at all the nodes of a network. This paper aims to understand the tradeoffs between these two techniques by comparing them with respect to multiple metrics including performance and complexity of implementation. We evaluate the throughput and delay characteristics of both WiLD-EDCA and WiLD-MAC under a single point-to-point link including an indoor wireless channel emulator testbed to evaluate the two approaches. Our results show that WiLD-EDCA is has better throughput for short and medium distances links (up to 40 km). At longer distances, WiLD-MAC improves performance for both point-to-point and multihop links, but it suffers from higher delay (from longer time slots). As far as we know, this is the first experimental comparison between TDMA based WiLD-MAC and IEEE 802.11e EDCA on a real hardware test-bed. The outline for the rest of the paper is as follows. We start by discussing the underlying shortcomings in Section II. Then we present both the approaches to improve performance of long-distance wireless in Sections III and IV. We present the results of our comparison of the two approaches in Section VI. Finally we discuss the implications and design guidelines in Section VII. II. C HALLENGES

WITH LONG DISTANCE WIRELESS

In this section, we first provide a brief overview of WiLD (Wireless in Long Distance) networks. We then briefly describe the key shortcomings in the default 802.11 MAC protocol that lead to poor end-to-end performance in WiLD networks. A. WiLD Networks: An Introduction The IEEE 802.11 standard, was designed for wireless broadcast environments with many hosts in close vicinity competing for channel access. Wireless radios are half-duplex and cannot listen while transmitting; consequently, a CSMA/CA (carrier-sense multiple-access/collision avoidance) mechanism is used to reduce collisions. Unlike standard WiFi networks, WiFi-based Long Distance (WiLD) networks can have long-distance point-to-point links ranging from 10-150 Km (unamplified links have been achieved up to 382 Km [12]). To achieve long distances in single point-to-point links, nodes use directional antennas with gains as high as 30 dBi, and may use high-power wireless cards with up to 600 mW of transmit power. Additionally, in multi-hop settings, nodes have multiple radios with one radio per fixed point-to-point link to each neighbor. if required each radio can operate in different channels. Some real life deployments WiLD networks include the Digital Gangetic Plains project [13], the Aravind network for telemedicine [14], the Akshaya network [15] for e-governance, the CRCnet project [16] and the different deployments done by Enlace HispanoAmericano de Salud (EHAS) Foundation [17] in Colombia (Department of Cauca) [18] and Peru (Department of Cuzco) [19]. B. Problems with unmodified WiFi As the 802.11 MAC protocol was not designed for long distance operation, real WiLD links show very poor endto-end performance. The key reasons for this are: (a) collisions at long distances; (b) inefficient link-level recovery and (c) inter-link interference. Collisions at long distances: The standard 802.11 protocol uses a CSMA/CA channel-access mechanism, in which nodes listen to the medium for a specified time period before transmission of a packet, thus ensuring that the channel is free. However, when using long links, it is possible that a node starts transmitting a packet unaware of another packet transmission at the other end. As a result, when the propagation delay increases, the probability of loss due to collisions also increases [8].

Inefficient link utilization: The 802.11 MAC uses a simple stop-and-wait protocol, with each packet independently acknowledged. Upon successfully receiving a packet, the receiver node is required to send an acknowledgment (ACK) within a tight time bound (ACKTimeout). If the ACK is not received at time, the sender has to retransmit. This mechanism has two main drawbacks: • As the link distance increases, propagation delay increases as well, and the sender waits during a longer time for the ACK to return. This decreases channel utilization. • If the time taken for the ACK to return exceeds the ACKTimeout parameter, the sender will retransmit unnecessarily and, thus wasting bandwidth[8]. As a result, we observe decreasing channel utilization as the distance, i.e. propagation delay of a link, increases. Multiple-link interference: Another important source of errors is the interference between adjacent 802.11 links operating in the same channel or in overlapping channels. This is because directional antennas also have sufficiently high gain (4–8 dBi) side lobes [13] in addition to the main lobes. Consequently, signal strength of packets transmitted from high-power radios on a node overwhelms any packet reception on other local radios [20]. In addition, due to the carrier-sensing MAC is not feasible to using simultaneous transmission, since radios can hear each others transmission, causing one of the radios to backoff. In a multihop network where each of these radios transmits over point-to-point long distance links to independent receivers, these effects lead to suboptimal throughput. III. W I LD-EDCA: L ONG

DISTANCE WIRELESS USING

IEEE 802.11 E EDCA

The IEEE 802.11e standard [21] includes EDCA (Enhanced Distributed Channel Access) as an evolution of the DCF (Distributed Coordination Function) in standard 802.11. It supports four access categories (AC) that may have different channel access probabilities at the MAC level. The ith access category has its own transmission queue characterized by a set of access parameters: AIF SNi (Arbitrary Inter-Frame Space Number), CWmin,i (Minimum Contention Window), CWmax,i (Maximum Contention Window) and T XOPi (Transmission Opportunity). A station willing to transmit has to wait for the channel to be idle during a period AIF Si = SIF S + AIF SNi .σ where σ is the SlotTime in the backoff process. Then, it uses a contention window of CWi slots, where CWi is a uniform random variable that takes a value in the range [0,CWi − 1] for the ith access category. The upper limit of CWi is increased exponentially each time the transmission is unsuccessful, starting at CWmin,i and ending at CWmax,i . If the channel is sensed busy at any time during the contention window, the countdown freezes, but is resumed once the channel has been sensed idle for a duration of AIF Si . When the countdown finishes, the station transmits and waits for an ACK that has to be sent by the receiver after a SIFS (Short Inter-Frame Space) time interval. If the ACK arrives before the ACKTimeout, the transmission is considered successful, otherwise the packet might be retransmitted. This process may be repeated several times until the maximum retransmission limit is reached. We can solve the most important problems of 802.11 MAC: (a) collisions at long distances and (b) low utilization (mentioned in Section II) by correctly adjusting several parameters of IEEE 802.11e EDCA. Some of these parameter adjustments are allowed by the standard while some others are non-standard adjustments. The last problem: (c) interlink interference, can be solved by using non-overlapping channels, assuming that we are using a relatively low number of nodes in our networks [18], [19]. Non-standard adjustment of 802.11 parameters: It has been shown in previous work [8], [22] that there are two essential timing parameters that need to be adjusted in plain 802.11 at long distances: ACKTimeout and SlotTime. The ACKTimeout needs to be increased in order to avoid the repeated retransmissions of frames when propagation time is too long. On the other hand, SlotTime must be increased to twice the propagation time for distances longer than 3 km (for a slot time equal to 20 µs) or 1.35 km (for a slot time equal to 9 µs) in order to guarantee that two stations that listen to each other may only collide if they transmit in the same slot. These adjustments are a slight violation of the standard but they are feasible with many available commercial WiFi systems, and compatibility with legacy WiFi stations can still be maintained. For the rest of this paper, the following adjustments for WiLD-EDCA are assumed:

(

σstd 2dmax c · σstd 2 c α = αstd − σstd + σ σ =

(1)

where σ is the optimal value for SlotTime, σstd is the standard value for SlotTime (20µs or 9µs depending on the case), dmax is the maximum distance between two stations that can collide in a BSS (Basic Service Set), c is the speed of the light, α is the optimal value for ACKTimeout, and αstd is the standard value for ACKTimeout [23]. Optimizing 802.11 EDCA parameters: In addition to the 802.11 timing parameters, the 802.11e EDCA extension offers control to adjust additional timing parameters. Recent work [9], [24]. has shown how two of these parameters: CWmin and TXOP limit, can be optimized for WiLD networks. If we increase the SlotTime parameter linearly with distance as explained above, it is straight forward to see that the CSMA/CA protocol wastes more time waiting to begin transmissions. It can be shown that the optimum CWmin,i changes with the distance, and there is an optimal value for CWmin,i called CWopt , that maximizes the throughput and minimizes the delay. This parameter allow us to optimize total throughput even while increasing the SlotTime parameter. In addition, 802.11e allows the station that wins the contention for accessing the channel to hold the channel upto the TXOP time limit, so that it can perform multiple packet transmissions. This is useful for increasing efficiency of the MAC protocol and for improving performance of WiLD-EDCA in terms of effective throughput and delay in the network. Finally, it is worth mentioning that the IEEE 802.11e EDCA standard includes a Block-ACK scheme that may be used in order to occupy the channel more efficiently by reducing the number of ACK transmissions. Multiple packets are transmitted in a block, separated by just the SIFS time, and the whole block is acknowledged by a single ACK frame called block ACK. Note that this mechanism is essentially similar to the bulk-ACK mechanism proposed for WiLD-MAC in the next Section IV. Although this mechanism is not yet implemented in real 802.11e EDCA hardware, recent IEEE 802.11n products include it. IV. W I LD-MAC: L ONG DISTANCE

WIRELESS USING

TDMA

As we saw in Section II, the key problems with 802.11 are (a) low utilization, (b) collisions at long distances, and (c) inter-link interference. The WiLD-MAC proposal [10] addresses the problem of low utilization by use of bulk packet acknowledgments. To mitigate loss from collisions at long distances as well as inter-link interference, WiLD-MAC replaces the standard CSMA/CA MAC with a TDMA-based MAC protocol. Bulk ACKs: The stock 802.11 stop-and-wait protocol is replaced by a sliding-window based flow-control approach in which the receiver transmits a bulk acknowledgment (bulk ACK) for a whole window of packets. The bulk ACK is generated as an aggregated acknowledgment for all the packets received within the previous slot. This way, a sender can rapidly transmit a burst of packets rather than wait for an ACK after each packet. The bulk ACK can be either piggybacked on data packets sent in the reverse direction, or sent as one or more stand-alone packets if no data packets are ready. Each bulk ACK contains the sequence number of the last packet received in order and a variable-length bit vector ACK for all packets following the in-order sequence. Here, the sequence number of a packet is locally defined between the pair of end-points of a WiLD link. Like 802.11, the bulk ACK mechanism does guarantee perfect reliability but has a maximum number of retries for every packet. Upon receiving a bulk ACK, the sender can choose to advance the sliding window skipping unacknowledged packets if the retry limit is exceeded. In practice, WiLD-MAC support different retry limits for packets of different flows. The bulk ACK mechanism introduces packet reordering at the link layer, which may not be acceptable for TCP traffic. To handle this, WiLD-MAC also provides in-order packet delivery at the link layer. Synchronous operation: With multiple adjacent directional links at a node, WiLD-MAC imposes synchronization among them. The nodes use a largely interference-free mode of operation termed as Simultaneous Synchronized Operation (SynOp) (proposed by Raman et al. [11]), where they either transmit simultaneously (SynTx), or receive simultaneously (SynRx). A simple loose time synchronization mechanism is used where during each time slot along each link, the sender acts as the master and the receiver as the slave.

Loss recovery: In WiLD-MAC, retransmissions or forward error correction (FEC) or a combination of both are used to deal with losses. A retransmission based approach can achieve the loss-bound q with minimal throughput overhead but at the expense of increased delay. An FEC based approach incurs additional throughput overhead but does not incur a delay penalty especially since it is used in combination with TDMA on a per-slot basis. A. Implementation WiLD-MAC is implemented as a set of driver modifications to control 802.11 parameters in combination with thin software layer that sits above the standard 802.11 Atheros (MADWiFi [25]) driver. Driver Modifications: WiLD-MAC needs to disable the following 802.11 MAC mechanisms: link-layer association in Atheros chipsets using the adhoc-demo mode, link layer retransmissions and automatic ACKs by using 802.11 QoS frames with WMM extensions set to the no-ACK policy and CSMA by turning off the Clear Channel Assessment (CCA) in Atheros chipsets. Scheduling layer: The single-link and inter-link synchronization using TDMA, sliding window flow control, and packet reordering for in-order delivery and the loss recovery mechanisms are implemented using the Click modular router [26] framework. Using kernel taps, WILD-MAC creates fake network interfaces. Then it uses Click to intercept packets sent to these virtual interfaces and modify them before sending them on the real wireless interface and vice versa. Packet queuing in the wireless interface causes variability in the time between the moment Click emits a packet and the time the packet is actually sent on the air interface. Thus, the propagation delay between the sending and the receiving click modules on the two hosts is not constant, affecting time slot calculations. Fortunately, this propagation delay is predictable for the first packet in the send slot, when the hardware interface queues are empty. In the current implementation of WILD-MAC, only the first packet in a slot is timestamped, and it is used to adjust the receive slot at the peer. If this packet is lost, the receiver’s slot is not adjusted in the current slot, but since the drift is slow this does not have a significant impact. V. E XPERIMENTAL S ETUP : M ETHODS

AND

T OOLS

We use a PropSim C8 wireless channel emulator for single link experiments where we can vary the length of the link as long as we want. In addition, the wireless nodes were enclosed in an anechoic portable camera Rhode & Schwartz CMU-Z10/Z11 to isolate them from external interference and also to ensure that the wireless radios communicate only through the emulated channel. This setup allows repeatable results by keeping link conditions stable during the whole experiment. For all the emulator experiments, we assume that the received signal strength is high enough to not cause bit errors. We use Atheros 802.11 a/b/g radios CM9 brand for all our experiments on 266 MHz x86 Geode single board computers running Linux kernel 2.4.26 loaded with MADWiFi [25] wireless driver and the WiLD-MAC kernellevel software. We use iperf [27] to generate UDP traffic for all the experiments. We added additional hooks in both MADWiFi and iperf to allow for accurate measurement of delay and throughput for our experiments. The configurations for the different scenarios are summarized in Table 1. To measure the mean end-to-end delay of the packets, we use the hardware TSF timer on Atheros wireless radios which can provide measurements that are accurate to a few microseconds. We use the ad-hoc mode of association which ensures that the TSF clocks on both the end-points are synchronized. At the sender, before sending a packet, iperf reads the TSF time from the wireless driver and inserts it into the packet. Another timestamp is inserted by the MADWiFi wireless driver before enqueuing the packet into the hardware queue. Similarly on the receive side, both MADWiFi driver and iperf record timestamps of the received packet. iperf server at the receiver then records all these timestamps are then printed out in a log file. These timestamps can now be used to calculate both application-level and driver-level delay for WiLD-EDCA and the WiLD-MAC protocols. The key steps of this process are illustrated in Figure 2. This measurement technique is similar to [28], where the authors request an interrupt after each successful transmission of packet to record the transmission time. Traffic tests were performed by injecting bidirectional traffic with short PLCP headers for a duration of 60 seconds with RTS/CTS disabled. Each traffic test was conducted 10 times and the results were then averaged to provide

Parameter PHY layer mode Channel rate Packet size Type of traffic PLCP header RTS/CTS Experiment time Queue length Distances [Km] Number of stations Offered load EDCA parameters (AIFSN, CWmin , CWmax , TXOP[ms])

Value 802.11b 11 Mbps 1472 bytes UDP short No 60 sec Driver = 50 packets [0-42] km and [0-102] km in 3 km steps 2 Nodes 9 Mbps Delay validation: AC BE=(2,32,1023,0) CWminopt variation: AC BE=(2,[4,8,16,32],1023,0) TXOP variation: AC BE=(2,CWopt ,1023,[0,20,40,60])

WiLD-EDCA WiLD-MAC WiFi parameters

Other parameters

AC BE = [2, CWopt , 1023, 60] Slotsize = 20ms Number of retries = 10 DIFS=2 CWmin =31 CWmax =1023 TXOP=0 SlotTime, ACKTimeout increasing with distance

Fig. 2. Steps involved in measuring delay in our experimental setup.

Fig. 1. Parameters for the experimental setup to compare WiLDEDCA and WiLD-MAC

statistical significance. The packet size was kept constant at 1472 bytes (1500 bytes minus 20+8 bytes for IP and the UDP header). We measured the throughput in saturation conditions with UDP traffic. We choose UDP as it gives us a more accurate estimate of the actual load in the network (no retransmissions at transport layer) and it also provides us the upper bound for the throughput possible with TCP. We also made sure that any loss on the wireless link is recovered by the MAC layer – either WiLD-EDCA or WiLD-MAC. For WiLD-EDCA, we only use class of traffic i.e. the best effort class (AC BE). A. Validation of setup We first validate our methodology for measuring throughput and delay by comparing our measurements with predictions of the theoretical model proposed by Tinnirello & Choi [29] for IEEE 802.11e EDCA. This model analytically determines the throughput and delay that can be achieved on a saturated link as we increase it’s distance. To compare with our setup with real radios, we measure the throughput and delay of directional UDP traffic in saturation on our wireless channel emulator as we increase the distance of the emulated link (by increasing the propagation delay). Figures 3 and 4 show the comparison of throughput and delay respectively. We can observe that the throughput measurements from the emulated link closely match the theoretical prediction. The emulated throughput is slightly lower consistently - this might by explained by small overheads in the real radio hardware that are not considered

Fig. 3. Comparison of theoretical and measured throughput with increasing distance. With bidirectional UDP traffic.

Fig. 4. Comparison of theoretical and measured delay with increasing distance. With bidirectional UDP traffic.

in the analytical model. VI. E VALUATION The objectives of our evaluation are the following: • Determine optimal WiLD-EDCA timing parameters for any distance. • Compare the throughput and delay achieved with the standard WiFi, WiLD-MAC and WiLD-EDCA protocols. • Examine the throughput/delay tradeoff between using WiLD-MAC and WiLD-EDCA. A. Optimizing 802.11 EDCA parameters We briefly discussed the effects of varying the EDCA timing parameters in Section III. Now we want to find out impact the of these parameters on throughput and delay of point-to-point links of varying distances.

Fig. 5. Throughput as a function of CWmin for a point-topoint link with increasing distance.

Fig. 6. Average delay as a function of CWmin for a pointto-point link with increasing distance.

We first look at the effect of the CWmin . Figures 5 and 6, show the achieved throughput and delay respectively for a point-to-point link over increasing distances as we vary the CWmin parameter. We are using 802.11e EDCA with only one traffic class with bidirectional UDP traffic. The other EDCA parameters are set to T XOP = 0,

CWmax = 1023 and AIF SN = 2. The standard deviation obtained in this case, is less than 0.03 for throughput and 0.05 for delay. We can see that there exists an optimal contention window (CWopt ) that minimizes delay and maximizes effective system throughput. Also note that the effect of this parameter depends on the distance: CWopt drops as the distance grows. This can be explained as follows. To avoid collisions at long distances, we have to change the SlotTime linearly with the distance as explained in the Section III. As a result, as the distance increases, more time is wasted by the CSMA/CA protocol in contention. Thus, the optimum CWmin should also decrease as the distance changes. For the rest of the experiments, we will fix CWmin = CWopt for each distance.

Fig. 7. Throughput with increasing distance point-to-point links at various values of TXOP.

Fig. 8. Average delay with increasing distance point-to-point links at various values of TXOP.

Next, we look at the impact of the TXOP parameter. Figures 7 and 8 show the achieved throughput and delay respectively for a point-to-point link over various distances as we vary the T XOP parameter. We are using 802.11e EDCA with only one traffic class with bidirectional UDP traffic. The other EDCA parameters are set to CWmin = CWopt , CWmax = 1023 and AIF SN = 2. Standard deviation obtained in this case is less than 0.05 for throughput, and 0.1 for delay. We can see that increasing the TXOP time to 20 ms increases, throughput and reduces delay significantly at any distance. However, any increase in TXOP after 20ms does not offer significant gains in throughput. We have also measured the jitter (because of its importance for real-time traffic) and observed it to be low even for high values of TXOP. B. Comparing WiFi, WiLD-MAC, and WiLD-EDCA In Figure 9 we compare WiLD-EDCA, WiFi and WiLD-MAC for point-to-point links using the configuration described in Table 1. Note that WiLD-EDCA is always better than WiFi for all distances in terms of throughput and delay and that the WiLD-MAC behavior is independent of the distance. While both WiLD-EDCA and WiLD-MAC outperform plain WiFi by a large margin and come close to optimal throughput (Figures 9 and 10), for shorter distances (< 40 km) WiLD-EDCA is better than WiLD-MAC in both throughput and delay. At longer distances, WiLD-MAC has better throughput, but it suffers a lot from higher delay (from longer time slots). It is important to note that the decrease of WiLD-EDCA throughput with distance is not because of collisions because we adapt the SlotTime to avoid collisions. Instead, the decrease in throughput with distance is a direct consequence of the simple stop-and-wait protocol used by 802.11 to confirm each frame. At the same time, the difference between WiLD-MAC’s throughput and the maximum throughput of 802.11 even at small distances can be partly explained by the overhead of the current implementation of TDMA in WiLD-MAC, which can be improved by using tighter time synchronization.

Fig. 9. Comparing throughput between WiLD-EDCA, standard WiFi and WiLD-MAC.

Fig. 10. Comparing average delay between WiLD-EDCA, standard WiFi and WiLD-MAC.

It should be noted that WiLD-MAC solution does not support class differentiation while IEEE 802.11e EDCA does. Hence, in this work we are not considering the effect of the AIFSN parameter that would allow us to include traffic differentiation or QoS in our networks [9], [24], which will be considered in future work. VII. C ONCLUSION IEEE 802.11 has been successfully used for the deployment of large area wireless networks in rural areas of developing countries. However, the MAC protocol of the standard was conceived for local area networks and must be changed in order to assure reasonable performance over long distances. This work has compared two different approaches towards maximizing the performance of WiFi on long distance networks - IEEE 802.11e EDCA optimized for long distances (WiLD-EDCA) and the TDMA solution proposed by the TIER group (WiLD-MAC) using an experimental setup using real hardware radios with a wireless channel emulator. Standard WiFi is shown to behave very badly as compared with both solutions. While WiLD-EDCA improves throughput and delay over long distance point-to-point links, it could still potentially suffer from inter-link interference in multihop networks. On the other hand, WiLD-MAC improves performance for both point-to-point and multihop links at longer distances, but it suffers from higher delay. In future work, we will try to use the QoS capabilities of IEEE 802.11e EDCA to perform traffic differentiation, especially for voice and video traffic. Other future work includes the experimental comparison in a real outdoor multi-hop environment. R EFERENCES [1] A., V. Martnez, J. Villarroel, F. Seoane, and Pozo, “Analysis of information and communication needs in rural primary health care in developing countries,” IEEE Transactions on Information Technology in Biomedicine, 2005. [2] Partnership for Higher Education in Africa, “Securing the Linchpin: More Bandwidth at Lower Cost,” 2006. [3] M. B. Yarali A, Rahman S, “WiMAX: The Innovative Broadband Wireless Access Technology,” Journal of Communications, Vol. 3, N. 2, 2008. [4] E. Brewer, M. Demmer, B. Du, M. Ho, M. Kam, S. Nedevschi, J. Pal, R. Patra, S. Surana, and K. Fall, “The case for technology in developing regions,” IEEE Computer, vol. 38, no. 6, pp. 25–38, 2005. [5] “IEEE P802.11, The Working Group for Wireless LANs,” http://grouper.ieee.org/groups/802/11/. [6] J. Simo, A. Martinez, S. Lafuente, P. Osuna, and J. Seoane, “The design of a wireless solar-powered router for rural environments isolated from health facilities,” IEEE Wireless Communications, Vol. 15, N. 3, 2008. [7] L. Subramanian, S. Surana, R. Patra, M. Ho, A. Sheth, and E. Brewer, “Rethinking Wireless for the Developing World,” in Hotnets-V, 2006. [8] Sim´o, R. J, M. J, S. A, and J, “Modeling and optimizing IEEE 802.11 DCF for long-distance links,” IEEE Transactions on Mobile Computing (submitted for revision), 2008.

[9] S. Salmer´on, J. Sim´o, J. Ramos, and A. Mart´ınez, “Long-distances networks for developing countries using IEEE 802.11e EDCA,” Wireless Communications and Mobile Computing (submitted for revision), 2008. [10] R. Patra, S. Nedevschi, S. Surana, A. Sheth, L. Subramanian, and E. Brewer, “WiLDNet: Design and Implementation of High Performance WiFi Based Long Distance Networks,” in NSDI, 2007. [11] B. Raman and K. Chebrolu, “Design and Evaluation of a new MAC Protocol for Long-Distance 802.11 Mesh Networks,” in ACM MOBICOM, August 2005. [12] Wired Magazine, “New WiFi Record: 237 Miles,” http://blog.wired.com/gadgets/2007/06/w wifi record 2.html. [13] P. Bhagwat, B. Raman, and D. Sanghi, “Turning 802.11 Inside-out,” in Hotnets-III, 2004. [14] S. Surana, R. Patra, S. Nedevschi, and E. Brewer, “Deploying a rural wireless telemedicine system: Experiences in sustainability,” Computer, vol. 41, no. 6, pp. 48–56, 2008. [15] “Akshaya E-Literacy Project,” http://www.akshaya.net. [16] “Connecting Rural Communities with WiFi,” http://www.crc.net.nz/. [17] “EHAS Foundation,” http://www.ehas.org. [18] A, A. Rendn, M. Martinez, J. Dulcey, R. Seoane, V. Shoemaker, D. Villarroel, J. Lpez, and Simo, “Rural Telemedicine Infrastructure and Services in the Department of Cauca, Colombia,” Telemedicine Journal and e-Health, 2005. [19] J, P. Simo, D. Osuna, L. Espinoza, R. Camacho, and Quispe, “Application of IEEE 802.11 technology for health isolated rural environments,” in WCIT, 2006. [20] B. Raman and K. Chebrolu, “Revisiting MAC Design for an 802.11-based Mesh Network,” in HotNets-III, 2004. [21] “IEEE Standard for Information Technology Medium Access Method (MAC) Quality of Service Enhancements. November 2005,” http://standards.ieee.org/getieee802/download/802.11e-2005.pdf. [22] Sim´o, J, Figuera, S. C, M. J, and A, “Distance Limits in IEEE 802.11 for Rural Networks in Developing Countries,” in IEEE WRECOM, 2007. [23] Bianchi, T. G, and I, “Remarks on IEEE 802.11 DCF Performance Evaluation,” IEEE Communications Letters, Vol. 9, N. 8, pp. 765-767, 2005. [24] S. Salmer´on, J. Ramiro, A. Mart´ınez, and J. Sim´o, “Traffic engineering in rural wireless networks for developing countries using IEEE 802.11 EDCA,” in IEEE WRECOM, 2007. [25] Atheros, “MadWiFi driver for Atheros Chipsets,” http://sourceforge.net/projects/madwifi/. [26] E. Kohler, R. Morris, B. Chen, J. Jannotti, and F. M. Kaashoek, “The Click Modular Router,” ACM Transactions on Computer Systems, vol. 18, no. 3, pp. 263–297, August 2000. [27] http://sourceforge.net/projects/iperf. [28] Dangerfield, M. Ian, L. David, and D. J., “Understanding 802.11e Voice Behaviour via Testbed Measurements and Modeling,” in WiOpt, 2007. [29] Tinnirello, C. I, and S, “Efficiency analysis of Burst Transmissions with Block ACK in Contention-Based 802.11e WLANs,” in ICC, 2005.