Information about http://tier.cs.berkeley.edu/docs/wireless/large_wild.pdf
Rethinking Wireless for the Developing World…
Tags: anmol, berkeley university, connectivity solutions, density environments, density regions, district connectivity, eric brewer, grass roots, initial experiences, intel research, intranet network, local communications, low density, networking research, new york university, research challenges, telecom carrier, university of california berkeley, university of colorado boulder, vision centers,Pages: 7
Language: english
Created: Thu Nov 9 01:36:14 2006
Display cached document
Page 1
Page 2
Page 3
Page 4
Page 5
Page 6
Page 7
Rethinking Wireless for the Developing World
Lakshminarayan Subramanian Sonesh Surana, Rabin Patra, Anmol Sheth
Intel Research, Berkeley and Sergiu Nedevschi, Melissa Ho, Eric Brewer
New York University University of California, Berkeley University of Colorado, Boulder
A BSTRACT benefit because they can be set up by grass-roots organiza-
Many rural regions in developing and developed countries tions as needed, avoiding dependence on a telecom carrier.
with low user densities do not have good connectivity so- This is particularly important for rural areas, which are less
lutions. To date, networking research has largely focused enticing to carriers due to the low density and income of po-
on urban areas of the industrialized world with high user tential consumers.
densities. In this paper, we make the case for research on Intranet usage: Providing network access does not neces-
new appropriate wireless technologies that can provide low- sarily have to be associated with Internet access. In many
cost, rapidly deployable connectivity solutions for low user- developing regions, basic local communications infrastruc-
density regions. To this end, we compare and contrast the ture is absent. A wireless network within a city or a district
connectivity requirements that arise in the two domains and can enable a wide range of applications including telephony,
pinpoint the new research challenges that arise in low user- essential services and health care. For example, we have de-
density environments. We describe our research efforts in ployed an intranet network in South India between hospitals
this space and also share our initial experiences in deploy- and rural vision centers that supports rural telemedicine [8].
ing low-cost Wifi-based Long Distance (WiLD) networks in Despite such a phenomenal growth in the adoption of long-
India, Ghana and the San Francisco Bay area. range wireless networks in developing regions, there have
been very few research efforts that take a concerted view
1 I NTRODUCTION towards analyzing how to build such networks. A primary
Today, the evolution of networks in the developing world is metric that distinguishes urban environments in developed
taking quite an alternate route from the traditional networks countries with a majority of regions in the developing world
we observe in the industrialized world. Many large cities in (with the exception of highly populated cities) is the density
East Africa have currently deployed a large number of tow- of users. We argue that prior work on wireless mesh net-
ers supporting a wide range of different long-range wire- works [4] is best suited for urban environments with high
less technologies such as microwave, long-distance WiFi, user densities. At lower user densities, the type of wireless
WiMax and other commercial wireless broadband solutions. network best suited to provide coverage is significantly dif-
African countries see better opportunity in wireless options ferent from the mesh networking model; such a network
for regions that have low penetration of fiber and other wire- would consist of nodes with directional/sector antennas and
line connectivity solutions; many of these countries have a point-to-point wireless links. Hence, the research challenges
higher cellphone penetration rates than fixed-line penetra- that arise in such an environment also significantly differ
tion [6]. The primary reasons for the boom in the use of long- from those of mesh networks.
range wireless networks within developing countries are: In this paper, we outline the research challenges that arise
in building low-cost, long-range wireless networks for low
Low-cost and decentralized evolution: In developing density regions. Our research has primarily focused on WiFi-
countries, wire-line connectivity solutions are not econom- based networks given that WiFi is much cheaper than other
ically viable in low-user density areas [7]. Satellite links, a wireless technologies and also operates in the unlicensed
common mode of Internet connectivity in much of Africa, spectrum. Some of the early works by Bhagwat et al. [2]
is also very expensive and not widely affordable (typically and Raman et al. [9] in this space focus on the specific as-
US$2,000 per month for 1 Mbps). Establishing wireless pects of tailoring the 802.11 MAC protocol to work in such
distribution networks (microwave, WiMax, WiFi-based or settings; while this is indeed relevant, it represents a small
CDMA450) to extend coverage within a region requires a portion of a much larger puzzle. In this paper, we take an
much lower capital investment. This allows for decentral- end-to-end systems perspective at the overall challenge: how
ized rapid evolution of such networks by local entrepreneurs. does one engineer a large-scale long-distance wireless net-
Among different wireless options today, WiFi-based net- work that can provide predictable coverage and good end-
works are currently much more economically viable than to-end performance in the face of competing traffic (from
WiMax, CDMA450 and microwave. other sources using the same network) and over potentially
Ease of deployment: Wireless networks are relatively easy highly lossy environments (induced by multi-path and exter-
and quick to deploy, particularly in cases where we do not nal interference) and systemic link/node failures? Answering
need new towers. Networks in unlicensed spectrum further this question involves addressing research challenges at var-
Characteristic High User Density Low User Density
Connectivity requirements Full coverage required Islands connected to each other
End Devices Individual, mobile, low power Shared, fixed, high power and LOS
budget and non-LOS
Topology Star-topology Point-to-point with end points within the network
Applications Mainly Internet access Internet as well as peer-to-peer Intranet access
Table 1: Characteristics of Low Density and High Density networks
ious layers of the networking stack. In this paper, we elab- oping countries is an urban phenomenon, with a few excep-
orate on these challenges and describe our initial efforts to- tions. Bangladesh has good rural coverage because it is actu-
wards addressing some of these challenges. We also briefly ally a very high density country, and base stations that cover
describe our deployment experiences in building three such roads and rail lines also cover many villages. China has dic-
WiFi-based long distance networks in India, Ghana and the tated good coverage as policy, despite the economic issues.
Bay Area. Other countries either subsidize rural users through taxation,
much like the US universal access tax, or require some rural
2 L OW VS H IGH U SER D ENSITY R EGIONS coverage as part of spectrum allocation. In its intended de-
In this section, we begin by contrasting low user density (ru- ployment model, with expensive basestations covering many
ral and semi-urban) and high user density environments (ur- users, WiMax also shares the shortcomings of other cellular
ban) and make the case for point-to-point long distance wire- technologies.
less networks using directional antennas in low-density en- Finally, 802.11 mesh networks [4], also assume high user
vironments. We do so by pinpointing why other well-known density. Moreover, mesh networks suffer from two basic
wireless technologies (VSATs, cellular, mesh networks) are problems when scaled to larger areas. First, as the net-
not economically viable in low-density environments. Next, work grows, an increase in the number of APs with omni-
given the distinction between these two environments, we directional antennas leads to increased interference in over-
describe the primary differences in the technical challenges lapping cells. Second, the use of low-gain omni-directional
that arise in point-to-point wireless networks in comparison antennas increases the hop length, and as a result through-
to wireless mesh networks, which have received a lot of at- put decreases. Bicket et al. [3] show that in Roofnet, longer
tention recently. routes (traversing multiple wireless hops) are disproportion-
ately slower mainly due to inter-hop collisions.
2.1 The Case for Point-to-Point Wireless Thus, we argue that for low density of users, approaches
Figure 1 lists some of the fundamental differences between that provide full coverage are not feasible. The alternative
providing wireless connectivity in high user density and low would be to cover only those few places where connectiv-
user density environments. These differences mainly stem ity is required, by employing long-distance point-to-point
from the constraints of providing low cost wireless connec- wireless links. Such links can rely on WiFi, point-to-point
tivity with small per-user cost and minimum or no recurring WiMax, or other technologies that support long-distance
cost. In low density environments people are usually clus- links offering reasonable bandwidths. In choosing such a
tered around small localities (e.g. villages), with large dis- technology, the most important factors are cost and config-
tances among these clusters. Even within villages the user urability. An interesting case are environments that have a
density is low compared to urban areas. In addition, the typ- mix of low and high user density regions. Here, a combined
ically lower incomes lead users to share computer terminals approach where the mesh network is augemented by point-
(e.g. Internet kiosks) to amortize the relatively high cost of to-point links as required can also be considered ([5]).
the devices and network connection. Until now, for practical and cost-related reasons, we have
Satellite networks provide fantastic coverage, but are chosen to examine the possibility of using WiFi-based Long
very expensive. VSAT equipment installation costs over Distance (WiLD) links. WiFi cards are cheap and highly
US$10,000 and the recurring monthly costs are over available, enjoying economies of scale. In our existing WiLD
US$2,000 for an 1 Mbps downlink. In low user-density re- deployments, the cost of a WiLD link is approximately $800
gions, VSAT is affordable only for businesses or wealthy (excludes the cost of tower) with no recurring cost.1 Because
users. they operate in unlicensed spectrum, WiLD links are easy to
Networks with a base-station model such as WiMAX, and deploy and experiment with, and spectrum license costs are
cellular networks like GPRS and CDMA, have an assymet- eliminated. Manufacturers of WiFi chipsets (e.g. Atheros)
ric design philosphy where expensive base stations are amor- often support open-source drivers, allowing us to completely
tized by large number of cheap clients over many users. In subvert the stock 802.11 MAC protocol and tailor the proto-
low-density regions, such base stations simply do not cover col to meet our needs.
enough users to be economically viable. The expectation that An alternative would be to use point-to-point WiMax
cellular solves the connectivity problem for developing re- 1
We are also deploying solar cells in our WiLD deployments
gions is thus somewhat of a myth: cellular success in devel-
links; such links would have a few important advantages Hence in WiLD settings, any external source can act as a
over WiFi: configurable channel spectrum width (and conse- hidden terminal.
quently datarate), better modulation (especially for non-line Therefore, external WiFi interference can be a very im-
of sight scenarios); operation in licensed spectrum would portant source of loss in WiLD environments; this is much
permit higher transmit power, and thus longer distances less so in mesh networks. Figure 1 shows a scatter plot be-
and better signal strengths. However, existing commercial tween the loss rate and the absolute number of external WiFi
WiMax products are only tailored for cellular providers traffic frames received on an urban link over a period of 6
and do not support point-to-point mode of operation. Ex- hours. The figure shows that a subset of the loss rate sam-
isting WiMax hardware is more expensive than WiFi (about ples are strongly correlated with the external traffic.2 This
$10000 for basestations), and the high spectrum license costs result is very different from the measurements reported in
in most countries dissuage grassroot style deployments. Cur- Roofnet [4] where the authors show the correlation between
rently it is also very difficult to obtain licenses for exper- loss rate and external WiFi traffic to be very weak. Although
imental deployment and we are not aware of open-source these measurements are collected in urban links, they also
drivers for WiMax basestations and clients (Wavesat offers a directly apply in low-density networks where one of the end-
mini-PCI based WiMax client development kit [10]). points is in an urban environment.
Consequently we advocate the use of WiLD links as the Multipath characteristics: In Roofnet [1], the authors con-
currently preferred solution; however, research investigating clude that multipath interference was a significant source of
long-distance point-to-point wireless networking should be packet loss. However, in WiLD networks, we observe quite
(for the most part) agnostic to the specific underlying wire- the opposite. This is primarily because the delay spreads in
less technology being used, allowing for other solutions to WiLD environments are an order of magnitude lower than
be used as they become available. We formulate our research that of mesh networks. The two factors contributing to lower
challenges accordingly. delay spreads in WiLD networks are the long distance of
2.2 WiLD vs Mesh networks WiLD links, and the line-of-sight (LOS) deployment of the
nodes. The strong line-of-sight component in WiLD deploy-
We continue by discussing how the characteristics of WiLD ments ensures that the attenuation of the primary signal is
networks differ from those of mesh networks, and thus lead only due to path loss, and most of the secondary paths are
to very different research agendas. We point out three key as- due to reflections from the ground. Furthermore, the long
pects that significantly differ between 802.11 deployments in distance between the endpoints ensures that the primary and
low-density settings (WiLD networks) and high-density set- the secondary reflection travel almost the same distance, and
tings (mesh networks): external WiFi interference, multipath hence reduces the delay spread. In comparison to our WiLD
characteristics and routing protocol characteristics. deployment, the Roofnet deployment has shorter links and
non-LOS deployments, which significantly increases the de-
Ext. Traffic (Kpackets)
3
2.5 lay spread.
2 Routing: From a topology perspective, two distinguishing
1.5 factors between mesh and WiLD networks are that mesh net-
1
works are unplanned while WiLD networks are planned, and
0.5
0
that the quality of links in mesh networks is time-varying and
0 20 40 60 80 100 nodes have several neighbors to potentially forward pack-
Loss Rate (%) ets. Hence, in mesh networks, routing is more opportunistic
Figure 1: Loss rate vs. ext. traffic observed on WiLD link
where nodes forward packets based on the quality of the link
External WiFi Interference: In settings where WiLD links at a given time. Roofnet's routing protocol, Srcr, chooses
co-exist with other external omni-directional WiFi trans- routes with a minimum "estimated transmission time" (ETT)
mitters (access points within the neighborhood), the hid- as a route selection metric [3]. In contrast, WiLD networks
den terminal problem is exacerbated. This is due to two consist of a few dedicated point-to-point links and routing in
features of WiLD links: directional transmissions and links WiLD networks resembles traditional routing protocols.
with long propagation delays. Due to the highly directional
nature of the transmission, a large fraction of interfering 3 E XISTING D EPLOYMENT
sources within range of the receiver act as hidden terminals Currently, we have deployed several WiLD networks in In-
since they cannot sense the directional transmission. How- dia (a 9-link topology ), Ghana (5 links) and the Bay Area
ever, in an omni-directional mesh network with overlapping 2
Based on experiments performed in a wireless channel emula-
transmission regions among neighbors, the fraction of exter-
tor we observed that at a channel separation of 2, the receiver is not
nal interfering sources that act as hidden terminals is much
able to receive the frames from the external interference source.
smaller. Due to long propagation delays, even external inter- However, the signal spillage of the interference source in the pri-
fering sources within the range of a directional transmitter mary channel is sufficient to cause frame corruption. This explains
can interfere by detecting the medium to be busy too late. why a subset of loss rate is not correlated with external WiFi traffic.
in the US (7 links). We use these testbed deployments to un- return exceeds a card-specific maximum timeout, the sender
derstand the different research issues and to implement and will retransmit unnecessarily and waste bandwidth .
evaluate the solutions to those challenges. The WiLD net- · Collisions due to bidirectional traffic: The CSMA/CA
work in India connects several village-based vision centers channel-access mechanism is not suitable for long distance
to the local Aravind Eye Hospital, and supports remote eye links; listening at the transmitter reveals little about the state
care as well as distance learning through interactive video of the receiver, due to the long distance and stale carrier
conferencing. In Ghana, the links are used by the University sense information due to propagation delays.
of Ghana to share Internet access, for distance learning, and · Multi-link Interference: When multiple WiLD links orig-
to exchange electronic library information among its differ- inating from a single node operate on the same or overlap-
ent campuses. Distances of our WiLD links vary from 10 ping channels, the transmission of one link can interfere with
80km, with relays installed where there is not line of sight packet reception on other links, because local side lobes are
due to geographical limitations. of similar strength to the signal received from afar.
We use low power single board computers (SBC) with a TDMA MAC Protocol with sliding window: The above
266 MHz x86-based chip, 128 MB RAM and up to 3 wire- limitations of the stock 802.11 MAC protocol motivate the
less cards for our wireless routers. For radios, we use off-the- need for a TDMA-based MAC protocol that synchronizes
shelf high power 802.11a/b/g Atheros cards with up to 400 the transmissions from the endpoints of a single point-to-
mW of transmit power output. The platform runs a stripped point link. For a node having multiple outgoing point-to-
down version of Linux from a 256 MB CompactFlash card. point links, Raman et al. [9] propose having simultaneous
To form long distance links we use high gain parabolic direc- send and simultaneous receive to eliminate interference. In
tional antennas (24 dBi, 8 degree beam-width). In multihop addition, the stop-and-wait recovery mechanism of 802.11
settings, nodes can use multiple radios with one radio per is unsuitable. We implement a sliding-window based flow-
fixed point-to-point link to each neighbor. control approach with the TDMA slots.
The above choice of hardware enables us to design routers
that are low cost(less than $400), consume less power (5 TDMA Slot Scheduling: Given these constraint of simul-
10W) and are of low weight (1015 kg for a node with two taneous transmit and receive, finding a feasible TDMA slot
antennas). While the small size and weight allows us to use schedule in a multihop network is non-trivial especially if
less expensive guyed-wired towers, the low power consump- we want to achieve optimal throughput across the whole net-
tion means that we can use small solar panels, which re- work. However, it can be shown that for bipartite graphs, we
duce the operating cost and increase reliability when unin- can always find such a slot schedule.
terrupted grid power supply is not available in developing 4.2 Loss Recovery Mechanisms
regions.
Across all of our WiLD networks, the presence of external
4 R ESEARCH C HALLENGES WiFi interference results in very high loss rates on WiLD
links. Furthermore, due to the long distances, the extent of
In this section, we elaborate on the research challenges that
interference could be very different at the two ends, making
arise in engineering large-scale WiLD networks to achieve
WiLD links asymmetric. Also, it is common to have links
predictable end-to-end performance in the face of competing
with loss rates fluctuating between 5 - 80% over short time
traffic from other sources and highly lossy links (induced by
scales.
external interference). We classify the research challenges
Figure 2 shows the loss rate sampled every 1 minute across
into the following categories: (1) MAC layer challenges; (2)
channel 1 and 11 for a 20 km WiLD link. The figure shows
Loss recovery mechanisms; (3) QoS Provisioning; (4) Trou-
that both channel 1 and 11 have long bursts of high loss
bleshooting, reconfigurability and management; (5) Network
rate due to external interference. Even in absence of long
planning and deployment. Associated with each of these
bursts there still exists a residual 58% loss. Given the sit-
challenges, we describe some of our early efforts to address
uation, an important challenge is to device appropriate link
them.
level loss recovery mechanisms that can achieve predictable
4.1 MAC Layer Challenges performance in the face of high loss variations.
The first challenge in running 802.11 on long-distance mul- Retranmissions with Bulk ACKs: The first approach for
tihop links is to adapt the 802.11 MAC protocol [9] to over- loss recovery is where the receiver acknowledges a set of
come its fundamental limitations which can be summarized frames at once using bulk ACKs, in the sliding window set-
as: ting proposed previously. The lost packets are then retrans-
· ACK timeouts: The simple stop-and-wait recovery mecha- mitted accordingly.
nism of the stock 802.11 protocol requires each packet to be Figure 3 shows the comparison of bidirectional TCP
independently acknowledged. This recovery mechanism is throughput achieved at various distances by the stock 802.11
ill-suited for long propagation delays, as it limits utilization MAC protocol (using CSMA) and by our implementation of
and thus bandwidth. Worse, if the time taken for the ACK to the TDMA MAC protocol with bulk ACKs. To emulate long
distances, we use a wireless channel emulator.. We can see
20
WiLD networks experience highly-variable delays due to the
TDMA nature of packet transmissions coupled with loss re-
Loss Rate (%)
15 covery. Hence, providing end-to-end bandwidth and delay
guarantees for flows requires scheduling mechanisms that
10
can take into account the variable link bandwidths and link
5 delays. Traditional QoS mechanisms assume the concept of
Ch. 1
Ch. 11 flow isolation i.e., once a set of resources are allocated to
0
1 21 41 61 81 101
a flow, this flow is unaffected by competing flows. This as-
Time units (1 minute) sumption does not completely hold in WiLD settings since
the introduction of a new flow can potentially affect the re-
Figure 2: Loss variation over time across channels 1 and 11 source allocation of competing flows (either along on links
in the same path or adjacent links along the path).
that as the distance increases, the throughput of CSMA MAC In addition to these differences, WiLD nodes have a
decreases gradually until distance reaches 110 km, which low processing power (266 MHz) and stringent mem-
corresponds with the maximum ACK timeout, and then it ory constraints (128 MB) that may rule out many fancy
drops drastically. However, the TDMA MAC protocol using strict/statistical QoS mechanisms which would require nodes
bulk ACKs provides sustained high throughput even at very to maintain per-flow state and track per-flow usage. We are
long ranges. currently deploying simple QoS mechanisms based on traf-
Adaptive FEC: With such highly variable packet losses fic priority classes similar to Diffserv without supporting
such as shown in figure 2, the retransmissions based ap- any form of strict guarantees. To provide statistical guaran-
proach would give us 0% loss but with highly variable delay tees at a per-hop level, the primary link-layer parameters that
and this is not suitable for audio and video traffic. We there- we can manipulate are: (a) loss-recovery parameters (FEC,
fore propose an adaptive FEC based loss recovery mecha- retransmissions); (b)varying the TDMA slot-size to reduce
nism which limits the delay experienced at each hop while delay. Manipulating these parameters represents a trade-off
guaranteeing a small loss rate. We are currently investigating spectrum between achieved loss-rate, delay characteristics,
appropriate FEC coding mechanisms for our WiLD setting. available bandwidth. As part of future work, we plan to ana-
We observe that the loss variability of the WiLD links are lyze this trade-off spectrum and quantify the achievable QoS
very hard to predict, making the problem of determining the properties in WiLD environments. Another related problem
appropriate FEC recovery mechanism a challenging one. is the optimal TDMA scheduling problem: Given a traffic de-
mand matrix between various sender-receiver pairs, can we
4.3 Quality of Service compute an slot schedule for every link in the network that
Many applications that use WiLD networks require QoS can satisfy all the traffic demands?3
(e.g., video-conferencing sessions in rural telemedicine).
4.4 Troubleshooting, Reconfigurability and Man-
Unlike the case of the Internet architecture, in WiLD net-
agement
works we have the flexibility of modifying routers to im-
plement QoS mechanisms. However, many of the traditional A key aim in WiLD networks is to reduce the operational
QoS mechanisms do not blindly carry over due to pecu- cost of maintaining the network. This is critical due to the
liar constraints imposed by WiLD networks. First, unlike lack of trained manpower in many developing countries, and
traditional wired links, WiLD links cannot be character- long delays involved in accessing the endpoints of a link due
ized by a fixed bandwidth value. In the presence of high to the distances and tower/pole deployments of the wireless
loss variations, the available bandwidth (after recovery) is routers.
time-varying. Also, the need for synchronous packet trans- Our experience with WiLD deployments shows that the
missions and receptions at a node, creates a direct cou- network can malfunction in a number of ways ranging from
pling between the available bandwidth on adjacent links; in complete failure of links (hardware board failure, corrup-
other words, any variation in the slot size along one link, tion of the flash memory cards, lightening strikes), to perfor-
affects the one-way bandwidth on adjacent links. Second, mance degradation over time (from misalignment of anten-
nas, signal attenuation from rain water clogging RF cables,
10
CSMA with 2 MAC retries
interference from external sources).
TDMA with Bulk Acks
Bandwidth (KB)
8 Reconfigurability: One way to deal with complete failure
6 of links or nodes is to design a redundant network topol-
4
ogy, with more than one possible path between the wireless
3
2 This problem assumes that all links are in the same channel.
Given non-overlapping channels, one can imagine a similar prob-
0
0 50 100 150 200 250 lem coupled with the need for an appropriate channel allocation
Distance (km) mechanism.
Figure 3: Comparison of WiLD MAC and stock 802.11 MAC
nodes. To reduce the cost of additional redundant links we electronically steerable antennas can be used for automatic
are exploring the use of low-cost electronically steerable an- alignment. The open research challenge lies in devising effi-
tennas instead. On a link failure, these antennas can dynam- cient algorithms to discover peer nodes and maintain align-
ically realign themselves and reform the topology of the net- ment using continuous adaptation over time.
work to route around failed nodes or links such that network
connectivity is maintained. 5 N ON - TECHNICAL CHALLENGES
Safe Upgrades: A safe upgrade mechanism is also required While deploying wireless networks in developing countries
for changing either the firmware or even the network con- we encountered a variety of non-technical problems. These
figurations on the routers. Any failure during this process deployments present much larger installation, maintenance
could lead to the endpoints being disconnected and out of and servicing costs, due to lack of local technical expertise,
reach. To avoid such failures, we use the built-in hardware equipment availability and logistics. Consequently, there is a
watchdog timer to power cycle the router on a failed kernel need for production-quality solutions, and not just research
change or erroneous configuration change and revert to a de- prototypes. The hardware and software must be robust, user
fault "golden" version. friendly, and simple to install, maintain and manage. Local
partners must be trained as well. Our group has learned these
Monitoring: The challenge in network management is to
lessons the hard way in India and Ghana.
continuously monitor the network with both passive and ac-
Another barrier is local telecommunication regulation,
tive measurements to test for anomalous behavior. Addition-
which is hindered by limited technical staff, "imperfect"
ally, the data aggregated from the distributed end-points in
government, and the presence of local incumbent monop-
the network should be automatically analyzed to pin-point
olies. Some of the problems we encountered are: restrictions
the location of the fault as well as diagnose the root cause of
on using VoIP (favoring local telecom monopolies), licensed
the fault. This information should be provided to the semi-
or even restricted frequency bands that are unlicensed ev-
skilled network administrator in a human readable form with
erywhere else in the world, and unregulated wireless usage
concrete troubleshooting steps to perform.
resulting in significant same-band interference.
Currently, in our existing deployments, we periodically ini-
tiate reverse ssh tunnels from the wireless routers to our 6 C ONCLUSION
server in Berkeley to collect a high level periodic health
We argue the need for concerted research efforts to develop
summary of each router node in the network. An alternate
cost-efficient networking solutions for providing connectiv-
solution is to have a completely orthogonal communication
ity to regions with low user densities. To this end, we ex-
channel like GSM/SMS. They provide a backup path for rare
amined various wireless options and their suitability, and ex-
situations where a remote reboot is required, but are expen-
plored WiLD networks as a promising option. By taking a
sive and assume some form of cellular coverage.
broad view of the problem, we found challenges at essen-
4.5 Planning and Deployment tially every layer of the network and thus a range of areas for
Planning of WiLD networks needs much more careful con- new research.
sideration compared to mesh networks with omnidirectional R EFERENCES
antennas. Since WiLD links traverse long distances, they re- [1] D. Aguayo, J. Bicket, S. Biswas, G. Judd, and R. Morris.
quire line of sight for operation and this usually implies tow- Link-level Measurements from an 802.11b Mesh Network.
ers at each end. As the towers take a substantial part of the to- In ACM SIGCOMM, 2004.
[2] P. Bhagwat, B. Raman, and D. Sanghi. Turning 802.11
tal cost of the network, the challenge is to select the locations
Inside-out. Hotnets-II, 2004.
of sites and the links so that the overall cost of the towers is [3] J. Bicket, D. Aguayo, S. Biswas, and R. Morris. Architecture
minimized (determined by the heights of the towers). Site se- and Evaluation of an Unplanned 802.11b Mesh Network. In
lection is also influenced by the presence of external WiFi in- ACM MOBICOM, 2005.
terference, as well as interference from the nodes which are [4] S. Biswas and R. Morris. Opportunistic Routing in Multi-
Hop Wireless Networks. Hotnets-II, 2003.
part of the WiLD network. WiFi interference from the nodes [5] Dharamsala Wireless-Mesh Community Network.
within the network as well as from the external sources can http://www.tibtec.org.
be minimized be judiciously selecting the transmit power of [6] International Telecommunications Union. World
the nodes. By over-provisioning the signal at the receiver, Telecommunications/ICT Development Report. 2006.
http://www.itu.int/ITU-D/ict/publications/wtdr 06/.
capture effect can be used to eliminate most WiFi interfer- [7] S. M. Mishra, J. Hwang, D. Filippini, T. Du, R. Moazzami,
ence. and L. Subramanian. Economic Analysis of Networking
An additional significant problem in the deployment of Technologies for Rural Developing Regions. Workshop on
WiLD networks is in performing the manual careful align- Internet and Network Economics, 2005.
ment of the directional antennas that is essential for the [8] R. Patra, S. Nedevschi, S. Surana, A. Bakre, E. Brewer,
K. Fall, and L. Subramanian. Achieving High Throughput
working of every long distance link. This is complicated by in Long-distance Wireless Networks. Intel Technical Report,
the fact that factors like wind and wear and tear of towers 2006.
can cause the antennas to misalign over time. In this respect,
[9] B. Raman and K. Chebrolu. Design and Evaluation of a new
MAC Protocol for Long-Distance 802.11 Mesh Networks. In
ACM MOBICOM, 2005.
[10] Wavesat. WiMax 3.5GHz mini-PCI Reference Design Kit.
http://www.wavesat.com/products/mini-pci.html.