AGRAWAL LAYOUT 8/2/07 3:27 PM Page 79

ACCEPTED FROM OPEN CALL

WIRELESS MESH NETWORKS:

CURRENT CHALLENGES AND
FUTURE DIRECTIONS OF WEB-IN-THE-SKY
NAGESH NANDIRAJU, DEEPTI NANDIRAJU, LAKSHMI SANTHANAM, BING HE, JUNFANG WANG,
AND DHARMA P. AGRAWAL, UNIVERSITY OF CINCINNATI

ABSTRACT Wireless mesh networks (WMNs) is an

upcoming technology that envisages supplement-
IGW Within the short span of a decade, Wi-Fi ing wired infrastructure with a wireless backbone
hotspots have revolutionized Internet service for providing Internet connectivity to mobile
provisioning. With the increasing popularity and nodes (MNs) or users in residential areas and
rising demand for more public Wi-Fi hotspots, offices, and could be called the Web-in-the-sky.
network service providers are facing a daunting WMNs are characterized by self-organizing self-
AP1 task. Wi-Fi hotspots typically require extensive configuring capability, ease, and (quick) rapidity
wired infrastructure to access the backhaul net- of network deployment. Since its inception in
work, which is often expensive and time consum- the early years of this millennium, it has been in
Client ing to provide in such situations. Wireless mesh the limelight of all researchers. Massachusetts
networks (WMNs) offer an easy and economical Institute of Technology’s (MIT’s) Roofnet [1, 2]
alternative for providing broadband wireless and Microsoft’s mesh networking project [3] are
Internet connectivity and could be called the some efforts in this direction.
Client
Web-in-the-sky. In place of an underlying wired Unlike the traditional Wi-Fi networks, with
(a) backbone, a WMN forms a wireless backhaul each access point (AP) connected to the wired
network, thus obviating the need for extensive network, in WMNs only a subset of APs are
cabling. They are based on multihop communi- required to be connected to the wired network.
cation paradigms that dynamically form a con- All the APs that are connected to the wired net-
Wireless Mesh nected network. However, multihop wireless work are called Internet gateways (IGWs) or
Networks offer an communication is severely plagued by many limi-
tations such as low throughput and limited
mesh points (MPs). APs that do not have wired
connections, called mesh routers (MRs), connect
easy and economical capacity. In this article we point out key chal- to the MPs using the multihop communication
lenges that are impeding the rapid progress of paradigm. Similar to the wired network, where
alternative for this upcoming technology. We systematically intermediate routers forward each other’s traffic,
examine each layer of the network and discuss in a WMN the MRs form the backhaul network
providing broadband the feasibility of some state-of-the-art technolo- and forward each other’s traffic in order to
gies/protocols for adequately addressing these establish and maintain their connectivity. Note
wireless internet challenges. We also provide broader and deeper here that MRs and MPs are similar in design,
connectivity and insight to many other issues that are of
paramount importance for the successful deploy-
with the only exception that an MP is directly
connected to a wired network, while an MR is
could be termed as ment and wider acceptance of WMNs. not. Figure 1 shows a sample mesh network in a
typical enterprise such as a university.
the Web-in-the-sky. INTRODUCTION Many companies such as Nortel Networks,
Strix Systems, and MeshDynamics are offering
The past few years have witnessed a tremendous mesh networking solutions for building automa-
growth of wireless LANs (WLANs) mainly due tion, small-scale and large-scale Internet connec-
to their ease of deployment and maintenance. tivity, and so on using customary products. For
WLANs have been deployed in enterprises, uni- instance, Strix systems have deployed 400 Strix
versities, and public wireless hotspots, also MRs in an area of 100 km to realize a city-wide
known as Wi-Fi hotspots (airports, hotels, etc.) WMN in Tempe, Arizona, that would also pro-
These Wi-Fi hotspots typically have one or more vide voice over IP among other functionalities.
access points that are connected to the wired Unfortunately, these companies often face
backbone network. Thus, deploying Wi-Fi tremendous challenges in designing, deploying,
hotspots requires extensive infrastructure and and ensuring optimal performance due to some
careful planning in order to minimize their costs. inherent problems in multihop networks. Multi-

IEEE Wireless Communications • August 2007 1536-1284/07/$20.00 © 2007 IEEE 79

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 80

The concept of mesh networking is very gen-

eral and can be extended to different technolo-
gies. For instance, the IEEE 802.16 WiMax
MN IGW AP
standard defines a mechanism for deploying a
To internet
wide area wireless broadband network using a
MN multihop mesh of powerful base stations. These
AP base stations operate in licensed spectrum and
have very long transmission range (around 70
MN
mi). With the rapid progress in ultra-wideband
AP (UWB) technology, short-range extremely high-
bandwidth WMNs are also envisaged for indoor
MN audio and video streaming applications. Howev-
MANET Gateway
node
AP
er, for the sake of clarity, in this article we focus
AP on WLAN-based mesh networks. Akylidiz et al.
[4] provide an overview of primitive mesh net-
works.
MN
In this article we systematically examine the
AP key challenges at each layer and discuss the fea-
MN MN sibility of some approaches to address these
MN
challenges. Before we proceed to a detailed dis-
Wireless link
cussion of the various issues, we first enlist some
of the unique features and characteristics that
■ Figure 1. A sample enterprise wireless mesh network. distinguish WMNs from existing mobile ad hoc
networks (MANETs) and wireless sensor net-
works (WSNs).
hop wireless communication is beset with several Mesh routers are relatively static — There-
problems such as high interference, increased fore, the route selection should focus on discov-
collisions due to hidden/exposed terminals, and ering high-bandwidth links that could provide
high levels of congestion. In addition, effects of low end-to-end delay and interfere with as few
fading and shadowing lead to very unreliable nodes as possible, and should possibly exploit
link connectivity. Finally, all of them culminate multiple path routes not only for resilience but
in extremely low end-to-end throughput, which is also for load balancing [1].
highly undesirable in the perceived applications Mesh routers have no power constraint — In
of WMNs. contrast to traditional wireless networks
Commercial interest in WMNs is prompting (MANETs and WSNs) where nodes are typically
immediate and increasing attention to integrat- power constrained, MRs have abundant power
ing them with the Internet. IEEE has set up at their disposal. Hence, the MAC/routing pro-
many Task Groups such as 802.11s, for specify- tocols designed for MRs should primarily focus
ing the physical (PHY) and medium access con- on maximizing the available channel bandwidth
trol (MAC) layer standards. Specifically, the rather than on power constraints.
Task Group’s target is to define an extended ser- Mesh routers are equipped with multiple
vice set (ESS) that provides reliable connectivity radios — With the plummeting costs of radios,
and seamless security, and ensures interoperabil- MRs can now be equipped with multiple radios
ity of the devices. It proposes the use of layer 2 within the bounds of permissible form factor and
routing and forwarding. Industry giants such as inter-radio interference. We can thus accomplish
Motorola Inc., Intel, and Nokia are actively par- simultaneous transmission and reception using
ticipating in these meetings. Two main proposals intelligent channel assignment to these radios.
from SEEMesh and the WiMesh Alliance were Different traffic model — Unlike MANETs
considered and successfully merged at the March where the traffic can be from any peer mobile
2006 meeting. The Task Group is actively work- node (MN) to any other MN, traffic in WMNs is
ing toward finalization of the standard and is predominantly between MRs and the IGW. MRs
expected to be approved by the end of 2008. are used primarily as an intermediate hop to for-
Although the envisioned applications of ward others’ traffic.
WMNs seem alluring, considerable work is still Traffic concentration may be higher along
required at all communication layers before certain paths — It may be valid to assume a uni-
widescale deployment of these networks is prac- form traffic distribution in MANETs, but in a
tical. One of the key challenges is improving the WMN, traffic is primarily concentrated along the
capacity and guaranteeing minimum bandwidth. paths directed toward the IGW.
As mentioned earlier, providing broadband Traffic volume and number of users — As
access for a community requires relatively higher mentioned earlier, MANETs have been designed
bandwidth and various quality of service (QoS) essentially for enabling communication within a
provisions. However, current PHY/MAC/routing small group of people. On the other hand, a
protocols cannot completely satisfy these WMN aims to provide high-bandwidth broad-
requirements. These networks are expected to band connections to a large community and thus
be self-configuring, self-healing, and resilient to should be able to accommodate a large number
device failures, and should be highly scalable. of users accessing the Internet. In a WMN, the
Thus, in order to succeed in the end user mar- estimated traffic volume is very high, so scalabili-
ket, a systematic design approach at all layers is ty and load balancing in routing become impor-
incumbent, and use of cross-layer information is tant issues.
inevitable. As there are considerable differences in the

80 IEEE Wireless Communications • August 2007

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 81

architectural design and application scenarios

between WMNs and MANETs/WSNs, we need
to consider these unique features while design-
ing protocols for WMNs.

PHYSICAL LAYER ISSUES

As the WMN is expected to provide broadband Hidden
Transmission
Internet access, it should have higher bandwidth. range terminal
With the fast developments in very large-scale
integration (VLSI) and digital communication Exposed
technologies, the raw data rates that can be sup- terminal
ported at the physical layer are rapidly increas- 3 1 2
ing. Specifically, raw PHY rates have increased RTS
from 1 to 100 Mb/s (IEEE 802.11n [http://www. 4
tgnsync.org/]) and soon may reach up to 1 Gb/s.
However, these rates are only theoretically
achievable, under perfect conditions with abso-
lutely no interference. In real-world deploy-
ments, these idealistic conditions seldom exist.
As the distance increases, the SNR decreases, as
a result of which signals encoded using higher Interference
modulation techniques cannot be decoded at the range
receiver. With the ever increasing wireless 5
devices in the market, interference from other
sources can hardly be avoided. Thus, the theo-
retical data rates specified in the data sheets are ■ Figure 2. The hidden and exposed terminal problems.
not achievable in a practical scenario.
As a large number of nodes may exist in a
WMN, interference also increases substantially. Open Research Issues
Clearly, the use of omnidirectional antennas • The number of noninterfering channels is
severely reduces the achievable throughput due finite and may be insufficient in some scenar-
to interference. One alternative is to utilize mul- ios where the node density is high.
tiple radios that can be operated on noninterfer- • The cost and complexity involved with such
ing channels. Another option is to use directional systems is high.
antennas. Recent developments in multiple input • A careful design of multichannel/directional
multiple output (MIMO) antenna technologies systems needs to be considered so that the
and smart antennas [5] can help in decoding upper layer protocols are able to fully utilize
wireless signals with low SNR, thereby increasing their features.
the achievable bandwidth. Moreover, using • Cross-layer design should be prudently consid-
directional antennas to concentrate signal power ered, as it may lead to complex system design
in particular directions can help in better fre- that may be inflexible for future develop-
quency reuse. Thus, use of such antennas at the ments.
physical layer can help cope with the growing
interference that degrades the performance of MEDIA ACCESS ISSUES
wireless networks.
Although multiple radios can be used to sig- Even though the PHY rates have considerably
nificantly improve the throughput gains, improp- increased, current media access protocols are
er placement may render them ineffective. Liese not able to realize the entire bandwidth provided
et al. [6] observed that if two antennae with a by the PHY layers. The primary reason for this
gain of 5 dBi are used in one access point (AP), subdued performance is due to improper or sub-
a minimum separation of 3 ft is required in optimal media access protocols that have been
order to use noninterfering channels simultane- designed primarily for single-hop networks. For
ously. Moreover, if the antennae are placed too example, the widely accepted MAC protocol
close to the ground, the signal strength degrades IEEE 802.11 DCF, when used in a multihop net-
rather rapidly. work, results in exiguous performance and is
Power control is another interesting aspect unacceptable. Some nodes may remain starved
that should be thoroughly investigated. Since the due to hidden and exposed terminals in a multi-
nodes in a WMN can be placed anywhere, topol- hop environment. Figure 2 illustrates these prob-
ogy control becomes important. Typically, assign- lems. Node 2, which is outside the interference
ing optimal power for controlling the topology range of 3, is unaware of the ongoing transmis-
can reduce interference and in turn help improve sion at node 3, and continues to send requests-
overall network performance. Having access to to-send (RTS) to node 1, causing collision. This
or control of parameters such as transmit power, is called the hidden terminal problem. Node 4 is
modulation, and received signal strength at prevented from transmitting because of the
upper layers will help in optimizing the overall neighboring transmission at node 3. This is
network performance. Cross-layer interaction is called the exposed terminal problem. An ideal
inevitable. Kawadia et al. [7] discuss various pre- MAC protocol for WMNs should provide fair
cautionary measures that should be considered access to all nodes competing for the channel.
while using a cross-layer approach. Prior research in MAC protocols in either

IEEE Wireless Communications • August 2007 81

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 82

44
IGW 1 IGW
IGW .11a AP1
.11a
1 60
AP1

AP1 AP2
AP2 Client .11a
6 40

Clients
Client AP2 AP3
6
Clients
Client Clients
Client Clients
(a) (b) (c)

■ Figure 3. a) Single-radio single-channel mesh network; b) dual-radio single-channel mesh network; c) multiradio
multichannel mesh network.

WSNs or MANETs has been primarily oriented the neighboring MRs at each hop along the
toward energy conservation. However, for MRs, backhaul link to the IGW.
power is no longer a constraint, and the focus of Figure 3c shows the use of a multichannel
MAC protocols should be on achieving higher approach using multiple radios that successfully
throughput rather than energy conservation overcomes all the problems encountered in the
issues. other architectures. Ideally, two radios are
Even with an efficient MAC design, fairness employed for the backhaul link and one for ser-
among flows of different hop lengths may not be vicing the mesh clients. The uplink and down-
achieved [8]. In multihop networks, a packet has link backhaul radios and the service radio are all
to be forwarded by many intermediate nodes. operated at nonoverlapping channels, eliminat-
Whenever an intermediate node receives a pack- ing potential co-channel interference. As each
et for forwarding, it has to perform IP layer mesh router can be equipped with multiple
lookup and Address Resolution Protocol (ARP) radios, fixed channel assignment to these radios
lookup, and then contend for the channel at the is a more viable solution. Efficient and intelli-
MAC layer at each hop. This not only increases gent channel assignment schemes have to be
end-to-end delay, but also dramatically amplifies designed as the number of channels is not infi-
the probability of packet loss. nite or may not be sufficient in a high node den-
A single--radio single-channel mesh architec- sity scenario. Several channel assignment
ture is shown in Fig. 3a. Ramanathan et al. [9] mechanisms [11–14] are being proposed. Elegant
suggests a radical change spanning across rout- assignment schemes can also be derived from
ing, MAC, and PHY layer protocols to speed up graph coloring approaches [15]. Ramachandran
the process of forwarding in these networks. et al. [13] investigate the channel assignment
Assigning orthogonal channels to the MRs with- problem based on an interference-estimation
in the interference range can help alleviate the technique. Adya et al. [11] perform the channel
hidden and exposed terminal problems, and assignment using a measurement-based approach
assist in improving the overall capacity of the that is dependent on channel quality for select-
network. So et al. [10] proposed multichannel ing the appropriate channel. Deafness is another
MAC protocols using a single radio for ad hoc key issue that has yet to be addressed in such an
networks and showed the performance gains. environment.
But considering the traffic characteristics in a Many commercial MRs2 currently in the mar-
WMN, frequent channel switching may be ket use multiple radios with multichannel capa-
required to communicate with neighboring bility for improving channel capacity. As all
nodes. In such scenarios, single-radio multichan- these vendors use their own proprietary MAC
nel MAC may not provide any significant perfor- and routing protocols for their products, interop-
1 In Atheros chipsets mance gains because of the high channel erability cannot be guaranteed. A multiradio
channel switching delay is switching delay.1 The plummeting cost of wire- unification protocol, a virtual MAC, proposed by
close to 10 ms, which is less radios has opened up new avenues for Adya et al. [11] is an interesting approach to
considerably higher when designing MRs using multiple radios. MRs can managing multiple radios. But concrete and
we consider the transmis- now use multiple radios tuned to orthogonal robust native layer 2 protocols can be more
sion rates. channels for simultaneously communicating with effective and efficient than such virtual MAC
their neighbors. Figure 3b shows the use of a solutions. With multiple radios communicating
2 Kiyon Inc. and, dual-radio WMN. One radio provides service to on noninterfering channels, an interesting aspect
MeshDymanics are two of its mesh clients (using channel 6 in 2.4 GHz to focus on is efficient flow scheduling in a
the startup companies 802.11 b/g), and the other radio forms the back- neighborhood so that resource usage can be
actively involved in manu- haul link connecting all the MRs to the IGW maximized.
facturing commercial (using 5 GHz 802.11a). However, it suffers from Another important issue that needs attention
MRs. the disadvantage of sharing the bandwidth with is QoS provisioning. The envisioned scenario of

82 IEEE Wireless Communications • August 2007

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 83

WMNs is expected to support applications like

broadband Internet access, and real-time appli-
which leads to higher packet loss and lower
throughput. The link quality is affected by two
As a WMN is likely
cations such as video streaming and voice con- factors: the distance between the transmitter and to serve large
ferencing. QoS provisioning for such key the receiver, and the interference at the receiver.
applications is an essential requirement and a Choosing a route with minimum hops results in number of users, it
key challenge. Performance of IEEE 802.11e selecting links that are further apart (i.e., the
(which was proposed for QoS provisioning in distance between the end nodes of the link requires high capacity.
WLANs) over multihop networks is yet to be increases). As the distance between the transmit-
investigated. Therefore, existing MAC protocols ter and receiver increases, the SNR correspond- Using multiple
have to be redesigned for efficient operation
over WMNs. Thus, suitability of IEEE 802.11
ingly decreases. But a high SNR is required at
the receiver for decoding packets transmitted at
orthogonal channels
MAC in a WMN is definitely debatable. A sum- high PHY rates.3 Thus, when packets are routed may be inevitable
mary of some open research issues in the MAC by a minimum-hop-centric method, packets
layer are listed below: transmitted at higher rates cannot be decoded, for achieving the
as a result of which packet error rate increases.
Open Research Issues In order to minimize packet error rate, nodes required capacity.
• Intelligent channel assignment should be are forced to compromise for lower data rates,
designed for efficient spectrum utilization which results in lower aggragate throughput in However, using
among multiple radios.
• Innovative techniques should be implemented
the network.
DeCouto et al. [2] take link losses into
multiple channels
to convey the channel assignment of a node to account and present a performance metric based over a single radio
its neighboring nodes. on expected transmission count (ETX) at each
• Scheduling of flows should be done intelligent- node to its next hop. ETX considers the forward can cause considerable
ly for maximizing resource utilization. and reverse drop probability of links and accord-
• Support for different QoS levels in a multi- ingly rates the path. However, it does not take delays, mainly due
radio multi-channel architecture using IEEE the throughput of the links into account; thus,
802.11e should be investigated. this metric may allow lower-throughput links. to channel switching
latency and
ROUTING ISSUES IN MESH NETWORKS MULTIPLE RADIOS, MULTIPLE CHANNELS
FORCE INTERACTION BETWEEN synchronization.
Existing routing protocols for multihop networks
such as AODV and DSR are primarily designed ROUTING AND MAC LAYERS
for low-end MANETs and are inefficient for the
WMN scenario [1]. The chief goal while design- As a WMN is likely to serve a large number of
ing routing protocols for MANETs and WSNs users, it requires high capacity. Using multiple
has been to conserve energy and improve packet orthogonal channels may be inevitable for
delivery ratio. They are designed to search for achieving the required capacity. However, using
stable routes in order to be robust and resilient multiple channels over a single radio can cause
to mobility. In these networks nodes are typically considerable delays, mainly due to channel
equipped with a single radio and share the same switching latency and synchronization. This miti-
channel for communication with each other. As gates the true advantage of using multiple
the channel is shared, choosing a route with a orthogonal channels. Fortunately, as the prices
minimum number of hops is ideal for achieving of wireless adapters keep dropping, it is econom-
better performance in MANETs. This is because ical to equip MRs with multiple radios. Each
a route with a smaller number of hops involves interface can be statically tuned to a different
fewer forwarding nodes, and consequently fewer orthogonal channel, obviating the need for fre-
transmissions and less energy consumption. Fur- quent channel switching. Intelligent channel
thermore, higher packet delivery ratios can be assignment to these radios can greatly improve
guaranteed with fewer hop routes in error-prone the capacity of the network. Alicherry et al. [16]
wireless channels. In contrast, MRs are neither propose heuristics for joint channel assignment,
power constrained nor highly mobile. Thus, the routing, and link scheduling for multiradio
focus of routing protocols for WMNs should be WMNs. They focus on achieving optimal chan-
on achieving higher throughput rather than sav- nel assignment with an interference-free link
ing energy or improving resiliency for mobility. schedule and routing by satisfying certain fair-
Another key challenge in WMNs lies in identify- ness, link congestion, and link scheduling con-
ing the best possible routes to the gateways from straints among others. However, with the
and to mesh clients. There are several other introduction of multichannel multiradio architec-
design considerations that require significant ture, the number of possible routing alternatives
attention, which we discuss further in this sub- increases, and route selection becomes more
section. complex. The routing algorithms should not only
enable selection of high-throughput links with
MULTIRATE LINKS low end-to-end delay, but also ensure minimal
Ironically, minimum hop routing metric has very interference between neighboring nodes. Inter-
poor performance in a stationary multihop net- ference is a critical issue in wireless networks.
work [2]. Thus, when selecting paths in WMNs, Estimating link interference is nontrivial and 3 Packets transmitted at
instead of a hop-centric design, we need to con- typically requires O(n4) comparisons for check- high PHY rates are encod-
sider throughput and an interference-centric ing the interference between every pair of links ed using complex modula-
approach. The primary reason for such poor per- in an n-node network. Padhye et al. [17] propose tion techniques and thus
formance in a static environment is the selection an empirical methodology for estimating the require high SNR at the
of suboptimal links (that are longer in range) same that minimizes the comparisons to O(n2). receiver to decode them.

IEEE Wireless Communications • August 2007 83

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 84

Even with this reduction, the time required for tree-based protocol, AODV-ST, that modifies
Capacity is a key estimating remains very high, necessitating fur- the popular AODV protocol by including expect-
ther improvement in the design of effective tech- ed transmission time (ETT) as the routing met-
issue in WMNs. niques. ric. In their work the mesh relays/routers
There are several IMPACT OF THE TRAFFIC MODEL
construct a spanning tree corresponding to each
gateway in the network. Each MR maintains a
elements that decide MANETs have been predominantly designed to primary gateway (with the least end-to-end
facilitate sharing of network resources and data delay) and routes traffic through this gateway.
the capacity of among a set of dynamic users. Typically, the traf- Load balancing is achieved by prudently routing
fic in a MANET is between any two nodes in the the traffic flows through the least loaded gate-
WMN, such as network. However, in WMNs the traffic is ori- way. In order to estimate the least loaded gate-
ented either toward (upstream) or away from way, periodic round-trip time (RTT) probing is
network architecture, (downstream) the IGW. Also unlike MANETs, performed. If the least loaded gateway found is
network topology, the expected traffic volume in WMNs is very
high. The use of traditional routing protocols
not the default gateway, new traffic flows are
routed through that node, and existing flows
traffic pattern, designed for MANETS in WMNs may lead to continue to use the earlier path. However, they
subdued performance due to the presence of do not consider routing in multichannel architec-
network node hotspots near an IGW. ture, and inter- and intraflow interferences
It is imperative for any routing protocol remain a big challenge.
density, number designed for WMNs to include efficient load bal- In summary, while designing routing algo-
ancing capabilities. It is also important to incor- rithms and metrics for WMNs, the following
of channels, porate efficient search techniques to select the challenges should be addressed.
transmission power best among the multiple paths and employ effi-
cient traffic splitting among these routes [1]. Open Research Issues
level, etc. Multipath routing increases the reliability and • Routing metrics should ensure that low-band-
robustness of the network. Existing literature on width links are seldom selected when links of
multipath routing in MANETs/WSNs primarily higher bandwidth are possible.
focuses on finding edge disjoint or node disjoint • Routing metrics should incorporate link stabil-
routes for energy efficiency. Although use of ity so that frequent route fluctuations can be
node disjoint routes may be feasible in a WMN, avoided.
the use of maximal disjoint routes can yield bet- • Metrics should choose routes that have good
ter distribution of traffic load. channel diversity to minimize intra- and inter-
Another important issue is link stability. Due flow interference.
to varied reasons such as interference, fading, • A routing algorithm should perform load bal-
and shadowing, links are highly unstable. As a ancing and ensure that a router does not
result, frequent route flaps occur, leading to an become a bottleneck node.
extremely unstable and unpredictable network. • Any load balancing algorithm that is designed
Thus, metrics designed for WMNs should be should have quick convergence and not be
able to capture the link stability in the network. oscillatory. At the same time, it should allow
Draves et al. [3] propose a new routing met- good usage of network resources.
ric, Weighted Cumulative Expected Transmis- • Loop-free routing should be guaranteed.
sion Time (WCETT), for multiradio • Routing algorithms should provide QoS guar-
multichannel networks. They improve ETX by anteed paths or at least some support for QoS
taking into consideration throughputs of the provisioning.
links and channel diversity. The proposed metric • Efficient route recovery and maintenance
chooses routes having high throughput links should be provisioned.
(consequently lower end-to-end delay). Further- • Elegant techniques to handle handoffs and
more, they also consider intraflow interference minimize handoff latencies should be
(packets of the same flow contend with each designed.
other at different hops) and alleviate this by
cleverly selecting routes that involve channel CAPACITY ISSUES IN WMNS
diversity. Although WCETT is shown to work
well in a multiradio multichannel network, it has Capacity is a key issue in WMNs. In the past
limitations due to interflow intereference that decade, several research efforts have been made
make it unsuitable for large networks. to study the capacity of wireless ad hoc networks
Raniwala et al. [14] propose Hyacinth, a mul- that can be adapted to the area of stationary
tichannel WMN in which each mesh node is multihop networks. These studies mainly focused
equipped with multiple 802.11 network interface on the issue of scalability of the throughput with
cards (NICs). They explore distributed channel the size of the network (i.e., the number of
assignment and routing by first creating a span- nodes n in the network). There are several ele-
ning tree rooted at the gateway. The available ments that decide the capacity of a WMN, such
NICs are divided among the parent node (UP- as network architecture, network topology, traf-
NIC) and the children nodes (DOWN-NIC). fic pattern, network node density, number of
The DOWN-NICs of the parent and the UP- channels, and transmission power level.
NICs of the children are tuned to the same One of the seminal works was done by Gupta
channel for communication, and the channel to and Kumar [19], who analyzed the capacity of
be used is assigned by the parent. However, stationary wireless networks from the viewpoint
load-sensitive path metrics are not suitable or of information theory. Their research is based
optimal for WMNs. on the principle that the interference between
Ramachandran et al. [18] propose a spanning- neighboring nodes utilizing the same channel to

84 IEEE Wireless Communications • August 2007

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 85

transmit data is the reason for the constriction in

capacity. When several nodes transmit simulta-
[20], their results showed that the delay scales as
Θ(n 1/2 /v(n)), where v(n) is the velocity of the
Though there exist
neously, a receiver can successfully receive the mobile nodes. In [22] they also designed a several research
data sent by the desired sender only if such scheme that achieves the optimal order of the
interference from local neighborhood is suffi- delay for any given throughput value. To achieve solutions for the
ciently small. In their model, they consider n the optimal throughput-delay trade-off, the
homogeneous nodes (with the same capability of scheme varies several elements — the number of capacity problem for
transmitting data at W b/s and the same radio hops, the transmission range, and the degree of
range) that are randomly deployed. For each node mobility — to achieve the optimal trade- the ad hoc networks,
node, a destination node is randomly chosen.
Based on the analysis on both the protocol and
off. Both the Gupta-Kumar and Grossglauser-
Tse models are included in this scheme as two
considering the
physical models, the following analytical lower extreme cases. differences between
and upper bounds of network capacity are Due to the difference between WMNs and
derived: When the nodes are randomly deployed MANETs, analytical results of MANETs may WMNs and ad hoc
with random communication pattern, the not be directly used in WMNs. Jun et al. [23]
throughput per source-destination pair is considered the major difference between WMNs networks, many
and ad hoc networks, which is the traffic pattern.
 W  Typically, traffic in WMNs is gateway oriented open research issues
Θ   , (i.e., either to or from a gateway), while in ad still exist in adopting
 n log n  hoc networks the traffic flows between any arbi-
trary pair of nodes. Based on this special proper- the solutions for
while the throughput achieved is ty of WMNs, they study the capacity from the
chain topology and propose that for WMNs the WMNs.
W  throughput of each node decreases as O(1/n).
Θ 
 n The reason for the poor performance of WMNs
compared to pure MANETs is the fact that gate-
when node placement and communication pat- ways are the hot spot of network traffic and
tern is optimal. From the analytical results, it could be a potential bottleneck of the whole net-
follows that the throughput capacity per node work’s capacity.
reduces significantly when the node density Another limitation of the existing research is
increases. An important implication derived that the focus has been on the theoretical analy-
from [19] is regarded as a guideline to improve sis for the asymptotic case. Exact capacity analy-
the capacity of ad hoc networks: A node should sis is another major contribution of [23]. For a
only communicate with nearby nodes. Another given topology and number of active nodes, they
important conclusion is that dividing the channel provide exact upper bounds on the throughput
into subchannels does not change the above of any node. Jain et al. [24] showed more
capacity bounds. research progress in obtaining the exact capacity
By allowing the nodes to move, Grossglauser of a multihop wireless network. Previous work
and Tse [20] showed that the per-user through- focused on asymptotic capacity bounds always
put can be increased dramatically. In their worked under the assumptions of homogeneity
research, if node motion is independent across of nodes, and randomness of network topology
nodes and has a uniform stationary distribution, and/or workload. However, they presented a
a constant throughput scaling Θ(1) per source- model and a methodology for computing upper
destination pair is feasible. This is obtained by and lower bounds of the optimal throughput for
exploiting the multi-user diversity benefits of the given network topology and workload. To
having additional “routes” between a source and achieve such optimal throughput, the nodes
a destination. should have the ability to control and schedule
Assuming the same physical model as in [19] packet transmissions according to requirements.
but with a different traffic pattern (a relay traffic Some optimal routes are found instead of the
pattern), the network capacity is studied in [21]. traditional shortest path routes. A conflict graph
In this model there is only one active source-des- is then used to model wireless interference
tination pair, while all other nodes assist this between neighboring nodes.
transmission. Under two additional assumptions, Recently, one approach has been discovered
they derive upper and lower bounds for the for increasing the network capacity of WMNs by
capacity using the max-flow min-cut theorem: employing multiple channels. Most existing
When the number of nodes in the network goes research on wireless network capacity typically
to infinity, the asymptotic capacity is O(log n) b/s. considered wireless networks with a single chan-
As discussed above in [19] and related publi- nel. Kyasanur et al. [25] derived the lower and
cations, the throughput optimal schemes have upper bounds on the capacity of a static multi-
been well studied, while the trade-off between channel wireless network. They study a wireless
the delay and throughput for such schemes has network model that has c channels and m ≤ c
not received much attention. Based on the mod- interfaces per node. They analyze the impact of
els studied in [19, 20], Gamal et al. [22] analyzed the number of channels and interfaces per node
the delay and determined the throughput trade- on network capacity, and find that in an arbi-
off in both fixed and mobile ad hoc networks. trary network, if the number of interferences per
For the Gupta-Kumar fixed network model, they node is smaller than the number of channels,
showed that the optimal throughput-delay trade- there may be a loss in network capacity. An
off is given by D(n) = Θ(nT(n)), where T(n) and interesting result in a random network with up
D(n) are the throughput and delay, respectively. to O(log n) channels is that for each node, even
For the Grossglauser-Tse mobile network model with a single interface, there is no capacity

IEEE Wireless Communications • August 2007 85

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 86

A selfish node degradation. They also studied the relationship

between interface switching delay and network
Thus, when designing routing protocols for these
WMNs, an efficient methodology should be in
always acts alone capacity. Their results show that in a random place to carefully identify and punish such selfish
network with up to O(log n) channels, if each MRs in the network. Detecting active and pas-
and does not collude node is equipped with some extra interfaces, sive selfishness is one of the key challenges in
interface switching delay will have no impact on ensuring secure operation of the network. Classi-
with others. In network capacity. fying an MR as selfish requires careful monitor-
Although there are several research solutions ing of the network. As a wireless channel
contrast, a malicious to the capacity problem for ad hoc networks, experiences interference, multipath fading, hid-
misbehaving node, considering the differences between WMNs and
ad hoc networks outlined earlier, many open
den terminal problems, and stray packet drops
might occur that should not be misclassified as
with the intent to research issues still exist in adopting the solu- selfishness.
tions for WMNs. Some the challenges are out- Incentive-based approaches that include cred-
disrupt network lined below. it or reputation-based schemes proposed for
MANETs are not suitable for static networks
activity by creating Open Research Issues such as WMNs. In credit-based approaches [26],
• In order to increase the capacity and reliability usually nodes earn credit by forwarding others’
routing disruptions of the network, new gateways can be added. A packets. However, a node in the periphery of the
and possible network strategy to decide on the optimal placement
of gateways should be designed.
network is handicapped as it would not be able
to earn credits by forwarding others’ packets. In
partitions, is more • Cross-layer consideration may be used to a MANET, as the topology changes are relative-
improve network capacity. ly frequent, any disadvantage due to improper
likely to collude • Optimal route selection strategies should be location is usually mitigated. But in a WMN, the
considered to increase the throughput of the MRs are relatively static. In addition to this, the
with other network. traffic in a WMN is primarily between the IGW
and MRs, unlike in MANETs where it could be
misbehaving nodes. SELFISH BEHAVIOR AND COOPERATION between any two MNs. And for the envisioned
mesh applications like broadband Internet
The self-configuring and distributed nature of access, IGW is primarily the source of traffic
WMNs and the ease with which an MR or MP flow as mesh clients download and upload data.
can be added makes it pertinent to ensure the Hence, credit-based schemes that restrict the
secure operation of the network. It also paves traffic from an MP/MR by the amount of credits
the way for rogue nodes called free-riders that it possesses are not applicable in WMNs. A node
merely enjoy the network resources without con- could also build a false façade by initially for-
tributing the needed functionality to the net- warding packets dutifully and then dropping
work. Furthermore, as the network can consist packets after becoming rich, which is counterin-
of nodes from different domains with conflicting tuitive to a credit-based detection scheme. In
interests, the nodes might not act cooperatively, credit-based schemes, a node lacking sufficient
and there is a high possibility of such free-riders. credit would either buffer or blindly drop pack-
Naive assumptions that all nodes are cooperative ets, making this approach highly unsuitable for
in forwarding each other’s packets impose seri- real-time applications like VoIP and videocon-
ous threats to the secure operation of a network. ferencing that need guaranteed QoS. Hence, we
There is an important distinction between see that a credit-based scheme is nothing but a
selfish and malicious nodes that should be noted. potpourri of loopholes for WMNs.
A selfish node would try its best to avoid detec- Next, we consider the compatibility of reputa-
tion, and these stealthy counter-attacks against tion schemes for WMNs. Although, they are not
detection have to be taken into account when as problematic as credit-based approaches, they
designing a detection scheme. A selfish node, in have some indigenous problems such as building
general, is less motivated to get involved in col- mutual trust indices between neighboring nodes.
lusion as it is greedy by principle in using the Transformation of a monitored network activity
channel for itself. Acting so would be rather into reputation is also a challenging task. In gen-
counterintuitive to its intentions. Hence, a selfish eral, this knowledge is gained by eavesdropping
node always acts alone and does not collude on the neighboring node’s transaction in promis-
with others. On the other hand, a malicious mis- cuous mode to check if it is forwarding packets
behaving node, with intent to disrupt network [27]. The strength of this method lies in the
activity by creating routing disruptions and possi- assumption that an omnidirectional transmission
ble network partitions, is more likely to collude as in 802.11 networks is used, which might not
with other misbehaving nodes. be feasible in multiradio and multichannel-capa-
A selfish MP that provides access to other ble WMNs. This also requires dedicated memory
MRs might try to greedily consume the available at each node. In a reputation scheme it is very
bandwidth by favoring its own traffic and dis- important to discount old ratings and have a fad-
cretely dropping others’. This leads to a situation ing factor associated with all ratings, as a node
where the performance of certain flows can be in with a good rating can become selfish at a later
jeopardy. We term this selfish behavior active time. Hence, we need to focus on the problem
selfishness. Another possibility is that some MRs of building a distributed monitoring system that
may refrain from advertising routes through weeds out in a timely manner the selfish MRs in
them, thus proactively becoming selfish to maxi- the network.
mize their bandwidth and economize their ener- In [28] we suggested the use of special local-
gy, consequently creating a network partition. ized detection agents called sink nodes for iden-
We call such misbehavior passive selfishness. tifying selfish nodes in the network, and discuss

86 IEEE Wireless Communications • August 2007

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 87

the appropriate action to be taken against such

selfish nodes. The sink nodes are delegated the
internal. An external intruder can disrupt the
routing by partitioning the network. An inside
An external intruder
duty of policing their local neighborhood. All attacker in the form of compromised nodes is a can disrupt the
nodes in the network submit to their nearest much more pernicious attack as it could go
sink node a summary of their transactions with undetected. An intruder can render the net- routing by partition-
their neighbors. These traffic reports consist of work dysfunctional in various ways such as
comprehensive information about the traffic route poisoning in the form of generating rout- ing the network.
exchanges of an MR with neighboring MRs. ing loops and misrouting of data, failing to for-
These reports are further analyzed individually ward traffic, executing a denial of service (DoS) An inside attacker
by the sink nodes in a distributed fashion, occa-
sionally consulting a central manager to break
attack, manipulating the content of payload, or
a man-in-the-middle attack via masquerading.
in the form of
ties. The sink nodes then advertise the identity Hence, the fundamental security primitives of compromised nodes
of the stealthy cheaters so that other MRs authentication, integrity, and confidentiality are
reroute their data along alternate paths and very much essential for the correct functioning is a much more
drop any traffic originating from selfish nodes. of WMNs. Authentication refers to the verifica-
Thus, the packet delivery ratio of a good MR tion of identities of the communicating entities, pernicious attack
improves over time, and that of a selfish MR is integrity refers to the validity of the original
kept in check. Bahl et al. [29] propose special message to guarantee that it has not been tam- as it could go
agents called air monitors to manage enterprise
wireless networks. They propose the use of desk-
pered with by an intermediate node, and confi-
dentiality refers to the establishment of a secure
undetected.
tops rather than relying solely on APs to moni- channel to transmit cipher text that would be
tor the wireless network. Thus, they create a garblde to an intruder who is eavesdropping.
low-cost management infrastructure with air Authentication, authorization, and accounting
monitors to monitor the wireless network. Such (AAA) services are generally provided in any
air monitors can be deployed at reliable nodes in wireless network by a centralized RADIUS
a WMN to sense the traffic in their neighbor- server. The key challenge involved in imple-
hood and evaluate the integrity of the network. menting authentication using centralized servers
There are several other issues that need to be like RADIUS lies in its scalability. As the
addressed to ensure fairness in packet forward- RADIUS server is typically located at either
ing among nodes. A group of bad nodes might the IGW or an MP in WMNs, when the num-
falsely implicate an innocent node by colluding. ber of mesh clients increases, the authentica-
This stealthy slow poisoning by liars could inflict tion overhead becomes excessively high. Hence,
undue actions against well behaving nodes affect- a scalable security solution has to be developed.
ing their throughput. A good node entrapped by Another main challenge is key management,
a misbehaving node is helpless in rerouting its which requires centralized issuing from an
traffic through alternate benign routes and total- authority. But the absence of a central trusted
ly under the mercy of its colluders. As a result, authority and the hierarchical nature of MRs
the performance of an entire network may be create additional overhead in key distribution.
affected. There are several challenges to be met A scalable distributed scheme is required to
in this regard, which are summarized below. distribute keys.
There are two dimensions to security in
Open Research Issues WMNs, which consist of MPs and mesh clients.
• A reliable identity association is required so As the interconnected MPs form the backbone
that a quarantined selfish node does not reap- of the network, the highest level of security is
pear with a new identity. required here. Hence, all ongoing traffic should
• An effective mechanism is needed to prevent be encrypted using secure standards like 128- or
source address spoofing, as a bad node could 256-bit AES encryption, and all MPs should be
spoof the address of a good node and impli- authenticated in the network. An equal amount
cate it. of attention should also be paid to mesh clients
• A node on joining the network could act good so as to prevent intruders in the network. This
in order to earn a good reputation for itself can be implemented by using authentication
and could then commence its selfish activity servers like RADIUS and 802.1x, and encrypting
under this facade. all the ongoing client traffic using standards like
• Even a quarantined selfish node should be 802.11i/EAS/TKIP. There are various secure
given a chance to resocialize in the network routing protocols in MANETs [30] that can be
after sometime. However, if resocializing is tailored to address route attacks in WMNs, but
allowed, transient liars that alternate between this requires careful consideration.
good and bad behavior pose severe problem. The 802.11 MAC protocol or another propri-
Such vacillating nodes should be quarantined etary MAC protocol currently being used in
forever [26]. WMNs is susceptible to several security flaws in
the backoff procedure and Network Allocation
SECURITY ISSUES Vector (NAV, used for carrier sensing). An
attacker can insert large NAV values, thereby
The thrust of research in WMNs is primary selfishly monopolizing the channel and resulting
focused on routing, and security is very much in in a DoS attack on an innocent node, select a
its infancy. Just like any multihop wireless net- small backoff value to regain the channel, incor-
work, WMNs are also plagued with several secu- rectly place a de-authentication request to the
rity issues. AP for a neighboring node by spoofing their
The openness of the WMN network makes address, or congest the network by a DoS attack
it vulnerable to intruders, both external and [31]. Hence, we also need a secure MAC proto-

IEEE Wireless Communications • August 2007 87

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 88

There are two col that would fix all these potential loopholes
that could be exploited by an intruder.
[7] V. Kawadia and P. R. Kumar, “A cautionary Perspective on
Cross-Layer Design,” IEEE Wireless Commun., Feb. 2005.
[8] N. Nandiraju et al., “A Novel Queue Management
dimensions to As an additional check, continuous monitor- Mechanism for Improving Performance of Multihop
ing of the network by an intrusion detection sys- Flows in IEEE 802.11s based Mesh Networks,” Proc.
security in WMNs tem that helps detect misbehavior is required. IPCCC, 2006.
[9] R. Ramanathan, “Challenges: A Radically New Architec-
Qui et al. [32] propose a novel troubleshooting
which consists of technique to trace the exact root source of prob-
ture for Next Generation Mobile Ad-Hoc Networks,”
Proc. ACM MOBICOM, 2005.
lems plaguing the multihop wireless network like
MPs and mesh packet dropping, link congestion, and MAC mis-
[10] J. So and N. Vaidya, “Multi-Channel MAC for Ad Hoc
Networks: Handling Multi-Channel Hidden Terminals
Using A Single Transceiver,” Proc. MobiHOC 2004.
clients. As the behavior. Physical and link layer parameters
such as received signal strength, packet transmis-
[11] A. Adya et al., “A Multi-Radio Unification Protocol for
IEEE 802.11 Wireless Networks,” Proc. IEEE Conf.
interconnected MPs sion, and retransmission counts are collected as Broadband Networks, 2004.
[12] P. Kyasanur and N. H. Vaidya, “Routing and Interface
trace data and reused to recreate the network
form the backbone scenario in a simulator like Qualnet for post- Assignment in Multi-Channel Multi-Interface Wireless
Networks,” IEEE WCNC, New Orleans, LA, Mar. 2005.
analysis. The simulator is then used as a fault
of the network, diagnosis tool for identifying deviant behavior.
[13] K. Ramachandran et al., “Interference-Aware Channel
Assignment in Multi-Radio Wireless Mesh Networks,”
In order to maintain secure operation of IEEE INFOCOM, 2006.
the highest level of WMNs, there are several open research avenues [14] A. Raniwala and T. Chiueh, “Architecture and Algo-
rithms for an IEEE 802.11-Based Multi-Channel Wireless
security is required to be explored that are summarized below. Mesh Network,” Proc. IEEE INFOCOM, 2005.
[15] T. R. Jensen and B. Toft, Graph Coloring Problems,
here. Hence, all the Open Research Issues Wiley Interscience, 1995.
[16] M. Alicherry, R. Bhatia, and L. Li, “Joint Channel
• Authentication of MRs is an important issue
ongoing traffic to be addressed.
Assignment and Routing for Throughput Optimization
in Multi-radio Wireless Mesh Networks,” Proc. MOBI-
• Scalable key management technique needs to COM, 2005.
should be encrypted be developed. [17] J. Padhye et al., “Estimation of Link Interference in
Static Multi-hop Wireless Networks,” Proc. IMC 2005.
• A secure multipath routing protocol is
using secure required.
[18] K. Ramachandran et al., “On the Design and Imple-
mentation of Infrastructure Mesh Networks,” Proc. IEEE
standards. • A multilayered security protocol that address-
es all the above mentioned issues is desired to
Wksp. Wireless Mesh Networks, 2005.
[19] P. Gupta and P. R. Kumar, “The Capacity of Wireless
Networks,” Proc. IEEE Trans. Info. Theory, vol. 46, no.
provide the highest level of protection. 2, 2000.
• A distributed intrusion detection system is [20] M. Grossglauser and D. Tse, “Mobility Can Increase
required for continuous monitoring of the net- the Capacity of Ad Hoc Wireless Networks,” Proc. IEEE
work. INFOCOM, 2001.
[21] M. Gastpar and M. Vetterli, “On the Capacity of Wireless
Networks: The Relay Case,” Proc. IEEE INFOCOM, 2004.
CONCLUSION [22] A. E. Gamal et al., “Throughput-Delay Trade-off in
Wireless Networks,” Proc. IEEE INFOCOM, 2004.
WMNs introduce a new paradigm of true wire- [23] J. Jun and M. L. Sichitiu, “The Nominal Capacity of Wire-
less Mesh Networks,” IEEE Wireless Commun., 2003.
less Internet access, providing the maximum [24] K. Jain et al., “Impact of Interference on Multihop Wireless
degree of flexibility at reduced cost to users. The Network Performance,” Proc. ACM MOBICOM, 2003.
scalability, self-configuring, and self-healing abil- [24] I. F. Akyildiz and X. Wang, “A Survey on Wireless Mesh
ities of WMNs makes it a versatile technology Networks,” Proc. IEEE Radio Commun., Sept. 2005.
[25] P. Kyasanur and N. Vaidya, “Capacity of Multi-Channel
expected to surpass other wireless technologies. Wireless Networks: Impact of Number of Channels and
As wireless mesh networks gain momentum in Interfaces,” Proc. MOBICOM, 2005.
an endeavor to complement the wired backbone [26] Y. Yoo, S. Ahn, and D. P. Agrawal, “A Credit-Payment
network, many issues are hindering its smooth Scheme for Packet Forwarding Fairness in Mobile Ad
Hoc Networks,” Proc. IEEE ICC, 2005.
progress. In this article we highlight some open [27] S. Marti et al., “Mitigating Router Misbehavior in
research challenges at different layers, examine Mobile Ad-Hoc Networks,” Proc. MOBICOM, 2000.
the feasibility of some state-of-the-art protocols, [28] L. Santhanam et al., “Distributed Self-policing Archi-
and discuss various issues. tecture for Packet Forwarding Fairness in Wireless Mesh
Networks,” to Appear in 11th IFIP Int’l. Conf. Pers.
Thus, the incumbent research proposals at Wireless Commun., Albacete, Spain, Sept. 20–22, 2006.
each layer should strongly advocate a secure and [29] P. Bahl et al., “DAIR: A Framework for Troubleshooting
scalable design seeking to further optimize the Enterprise Wireless Networks Using Desktop Infra-
performance of WMNs. Although they have structure,” Proc. ACM HotNets-IV 2005, College Park,
MD, 2005.
developed in leaps in recent years, there is still [30] Y. Hu, A. Perrig, and D. Johnson, “Ariadne: A Secure
ample research work for radical progress of on Demand Routing Protocol for Ad Hoc Networks,”
WMNs and true realization of Web-in-the-sky. Proc. ACM MOBICOM, 2002, pp. 12–23.
[31] V. Gupta, S. Krishnamurthy, and M. Faloutsos, “Denial
REFERENCES of Service Attacks at the MAC Layer in Wireless Ad Hoc
Networks,” Proc. IEEE MILCOM, 2002.
[1] C. Cordeiro and D. P. Agrawal, Ad Hoc & Sensor Net-
[32] L. Qui et al., “Troubleshooting Multihop Wireless Net-
works, Theory and Applications, World Scientific,
works,” Proc. SIGMETRICS (extended abstract), June 2005.
Spring 2006.
[2] D. D. Couto et al., “A High-Throughput Path Metric for
Multi-Hop Wireless Routing,” Proc. ACM MOBICOM, ADDITIONAL READING
2003. [1] S. Zhong, Y. Yang, and J. Chen, “Sprite: A Simple,
[3] R. Draves, J. Padhye, and B. Zill, “Routing in Multi- Cheat-Proof, Credit-Based System for Mobile Ad Hoc
Radio, Multi-Hop Wireless Mesh Networks,” Proc. Networks,” Proc. IEEE INFOCOM, 2003.
MOBICOM, 2004.
[5] V. Jain et al., “A Cross Layer MAC with Explicit Synchro-
nization through Intelligent Feedback for Multiple BIOGRAPHIES
Beam Antennas,” Proc. IEEE GLOBECOM, 2005. DHARMA P. AGRAWAL [M’74, F’87] (dpa@ececs.uc.edu) is the
[6] S. Liese, D. Wu, and P. Mohapatra, “Experimental Char- Ohio Board of Regents Distinguished Professor of Comput-
acterization of an 802.11b Wireless Mesh Network,” UC er Science and Engineering and founding director of the
Davis Comp. Sci. Dept. tech. rep. Center for Distributed and Mobile Computing in the

88 IEEE Wireless Communications • August 2007

AGRAWAL LAYOUT 8/2/07 3:27 PM Page 89

Department of Electrical and Communications Engineering DEEPTI NANDIRAJU (nandirds@ececs.uc.edu) has been a doc-
and Computer Science (ECECS), University of Cincinnati, toral student in the Center for Distributed and Mobile Thus, the incumbent
Onio. He has been a faculty member at North Carolina Computing at the University of Cincinnati since 2004. She
State University, Raleigh, (1982–1998) and Wayne State received her B.S. degree in 2001 and M.S. degree in 2003 research proposals at
University, Detroit, Michigan (1977–1982). His current with a gold medal in computer science from Assam Univer-
research interests are energy-efficient routing and informa-
tion retrieval in sensor and mesh networks, QoS in inte-
sity, Silchar, India. Her current research interests are in the
area of fairness issues and routing protocols for wireless ad
each layer should
grated wireless networks, use of smart multibeam
directional antennas for enhanced QoS, and various
hoc and mesh networks.
strongly advocate a
aspects of sensor networks including environmental moni- L AKSHMI S ANTHANAM (santhal@ececs.uc.edu) received her
toring and secured communication in ad hoc and sensor B.E. degree in computer science and engineering from the secure and scalable
networks. His co-authored textbook, Introduction to Wire- University of Madras, India, in 2003. She is currently a doc-
less and Mobile Systems (Thomson) has been adopted toral student working as a research assistant in the Center design seeking to
throughout the world and revolutionized the way the for Distributed and Mobile Computing at the University of
course is taught. His second co-authored textbook, Ad Hoc
and Sensor Networks, has just been published. He has
Cincinnati. Her research interests include detection of self-
ish behavior, traceback of DoS attacks, combating DoS
further optimize the
served as an editor of IEEE Computer magazine, IEEE Trans-
actions on Computers, and the International Journal of
attacks, intrusion detection in multihop networks, and
other security concerns in wireless ad hoc and mesh net-
performance of
High Speed Computing. He is an editor for the Journal of works.
Parallel and Distributed Systems, International Journal on WMNs. There is still
Distributed Sensor Networks, International Journal of Ad B ING H E [S ’01](heb@ececs.uc.edu) is a Ph.D. student in
Hoc and Ubiquitous Computing, International Journal of computer engineering at the Center for Distributed and ample research work
Ad Hoc & Sensor Wireless Networks, and Journal of Infor- Mobile Computing in the Department of ECECS, University
mation Assurance and Security. He has been Program Chair
and General Chair for numerous international conferences
of Cincinnati. His current research interests include archi-
tecture and capacity of wireless mesh networks, and
for radical progress
and meetings. He has received numerous certificates and
awards from the IEEE Computer Society. He was awarded a
resource allocation in 802.16 wireless MANs. He received
his B.S. degree in communication engineering and M.S.
of WMNs and a true
Third Millennium Medal by the IEEE for his outstanding degree in signal and information processing from Northern
contributions. He has also delivered keynote speeches at Jiaotong University of China. He has been an engineer at realization of
five international conferences. He also has four patents and Honeywell Technology Solutions Laboratory in China.
16 patent disclosures in wireless networking. He has been Web-in-the-sky.
selected as a Fulbright Senior Specialist for a duration of J UNFANG W ANG (wangjf@ececs.uc.edu) received her B.S.
five years. He is a Fellow of the ACM, AAAS, and WIF. degree in computer science from Northeastern University (,
China, and her M.S. degree in computer science from Nan-
NAGESH NANDIRAJU [S] (nandirns@ececs.uc.edu) is currently jing University of Aeronautics & Astronautics, China. She
pursuing his doctoral studies in the Department of ECECS, was previously a senior software engineer and team man-
University of Cincinnati. He is working as a research assis- ager at Zhongxing Telecom Company, China, working on
tant in the Center for Distributed and Mobile Computing the PCS and has a patent pending on it. Currently she is
at the University of Cincinnati. He received his B.E in com- pursuing her Ph.D. degree in the Center for Distributed and
puter science and engineering first class with distinction Mobile Computing in the Department of ECECS, University
from the University of Pune in 2001. His research interests of Cincinnati. Her current research interests include posi-
are in the broad area of wireless ad hoc and infrastruc- tioning techniques, channel assignment, and cross-layer
tured networks. His work includes performance evaluation routing in wireless mesh networks, and interference control
and design of efficient MAC and routing protocols for mul- in WLAN and wireless mesh networks.
tihop wireless ad hoc, mesh, and sensor networks. He has
prior work experience at the National Institute of Technolo-
gy, Silchar, India.

IEEE Wireless Communications • August 2007 89