Congestion Free Routes for Wireless Mesh Networks

Congestion-free Routes for Wireless Mesh Networks
Nemesio A. Macabale Jr.*†, Roel M. Ocampo†, and Cedric Angelo M. Festin†
*
Central Luzon State University, Philippines
†
University of the Philippines, Philippines
E-mail:{namacabale, roel, cmfestin}@up.edu.ph

Abstract— Recently proposed wireless mesh routing metrics neighbors. Consequently, we found out that the information provided
based on awareness of congestion, load or interference typically by the MAC layer of a wireless node would be sufficient to achieve
employ queue occupancy of a node's wireless interface to our goals.
estimate traffic load. Queue occupancy, however, does not The rest of the paper is organized as follows: Section II elaborates
directly reflect the impact of channel contention from neighbor further on the motivation for this work, and discusses similar work
nodes. We propose an alternative called the channel load-aware found in the literature. Section III presents an analysis that leads to
(CLAW) routing metric that takes into consideration not only CLAW's design and implementation, while Section IV discusses the
the traffic load within the node itself, but also the degree of results of the preliminary evaluation. Finally, Section V concludes by
interference and contention within the channel. CLAW uses enumerating the contribution of this work.
local information from a node's MAC layer to estimate channel
busyness and contention levels. It does not require complex II. RELATED WORK
computations, nor the exchange of link-level statistics with
neighbors. Our preliminary results show that CLAW can In a multi-hop WMN, routing is more critical than in wired
identify congested regions within the network and thus enable networks, because the wireless medium is shared and is highly
the determination of routes around these congested areas. We dynamic [11]. Different packet flows may interfere with each other
present the results of simulations we conducted to evaluate the even when they do not necessarily traverse the same path. Along a
use of CLAW in mesh-wide routing. path, neighboring nodes that share a channel compete for its use
forming a collision domain (see Figure 1). As more flows traverse
Keywords - wireless mesh networks, routing, routing metric, nearby paths and nodes, they compete for access to the shared
congestion awareness. channel, eventually congesting the path and lowering throughput
significantly.
I. INTRODUCTION There have been several efforts to address this issue through the
use of load-aware, interference-aware, and/or congestion-aware
Wireless mesh networks (WMN) have attracted significant routing metric either singly [12-23] or in combination with multiple
attention in recent years for flexible and rapid deployment of metrics [24-32]. Load-aware routing algorithms such as DLAR [16]
wireless services in a wide variety of applications. These and ALARM [30] measure load based on the number of packets
applications include broadband home networking and automation buffered in the interface queue. However, a single node's internal
[1], [2], community mesh networking [3-5], in transportation load as gauged from the state of its buffers cannot reliably estimate
systems [6], public safety and disaster scenarios [7], [8], and in the level of congestion within a collision domain, because the queues
medical applications [9]. of other nodes within that domain could be empty or lightly loaded.
Mesh networks are composed of wireless nodes that participate In this case, the heavily- and lightly-loaded nodes do not jointly paint
either as routers or clients of the network. The mesh routers are a consistent picture of the channel. In other words, while interface
generally static or minimally mobile and serve either as dedicated queue occupancy accurately measures load on nodes, it does not
forwarding nodes, access points for clients like desktop PCs, laptops necessarily estimate the load on a region in a network.
and mobile devices, or both. Collectively, mesh routers form the To measure loaded regions, many proposals either obtain the sum
backbone of the wireless network, enabling traffic to be transported [13], [18], [20], [21], [28], [29], [31] or the average of queue length
and ensuring reachability between participating nodes. [14], [15], [19], [25] of nodes within a collision domain. This
However, despite advances in the field, there are still many approach requires the data to be collected or exchanged among
interesting research challenges in optimally routing traffic within a neighbors, and thus generates additional overhead in terms of
wireless mesh network. Due to the shared nature of the wireless bandwidth and route convergence time.
channel, routing based on metrics traditionally used in wired Other proposals measure channel load based on radio-frequency
networks such as hop counts do not take into account interference (RF) channel interference [18], [23] and delay [33], [34]. However,
and contention within the channel shared among neighboring mesh in most wireless environments there are other potential sources of
nodes. As a result, routing algorithms that use such "congestion- interference and delay aside the load in the channel, such as physical
agnostic" metrics may tend to direct multiple traffic flows naively layer impairments and bad channel conditions [11]. Hence, there
along known best paths, eventually congesting wireless channels should be a way to both measure and differentiate channel and node
along the path and causing significant drops in network throughput. load. The interference awareness and load-balancing metric in [26],
In contrast, a routing algorithm that is able to veer the traffic flow [27] requires probe packets and neighbor-wide gathering of link-state
towards calmer regions of the network would be less likely to suffer statistics, which likewise generate overhead in the bandwidth and
from such a scenario. time needed to calculate the metric.
To address this issue, we propose a routing metric called the Some proposal that truly measure congestion, interference, and
channel-load aware routing metric (CLAW) designed to take into load include LWR [12] and C2WB [17]. LWR however combines
account congestion, interference and load-imbalance issues found in multiple metrics to achieve its goal, requiring more calculations than
wireless mesh networks. Our design goal is to come up with a simple CLAW, which relies on a single metric. In addition, LWR collects
yet accurate congestion / interference / load-aware routing metric information from neighbors. Similarly, C2WB requires probing
that can be incorporated into a more general concept of capacity packets, neighbor information, and a complex computation. In
awareness [10]. To accomplish this task, in CLAW's design, we addition, it requires a change in the MAC layer protocol,
avoided the need to advertise and collect link-level statistics between

We proposed CLAW to address the issues mentioned, through the Ch _ load=T sensedEnergyT blockedForAccess (1)
use of node-local information, and by requiring only simple
computations. In our investigation, we found that the MAC layer has
all the information needed to accurately estimate channel load, where :
T sensedEnergy is that fraction of time that a node is transmittinga packet to
interference, and node load. CLAW can be used by routing protocols
as an alternative to existing congestion awareness mechanisms either thechannel , is receiving a packet from the channel , issensing
in single channel or multi-channel environments. transmissionenergy beit collision , interference , or noise
in thechannel
III. DESIGN AND IMPLEMENTATION
T blockedForAccess isthat fraction of time that anode
Our analysis begins by looking at a node j's collision domain. It is is backing - off or deferring
comprised of all nodes within j's carrier sensing range that operate on
the same channel. Transmissions of these nodes may interfere with
transmissions from j. This is illustrated in Fig. 1, with the 01 02 03 04 05
simplifying assumption that the carrier sensing range is circular. The Node i's collision
nodes in this diagram are furthermore assumed to operate using the domain
IEEE 802.11b wireless standard. 06 i 08 09 10
Because of the shared nature of the channel, the load on a node

k
affects all the neighbor nodes that can sense its transmission. That is,
an idle node will respond to a new traffic flow request like a busy or 11 12 j 14 15
loaded node if a neighbor within its carrier sensing range is in fact
busy or loaded. Hence, identifying busy regions, rather than busy
16 17 18 19 20
nodes, is a more effective approach in avoiding congestion,
preventing interference, and distributing traffic loads. The routing Node j's collission
domain
protocol may then assign a lower cost to the next-hop node that has 21 22 23 24 25
the least busy collision domain. This is the basic intuition behind,
and our motivation for, the development and use of the CLAW Figure 1: A Wireless Mesh Network with 25 nodes
metric.
A. Channel Load busy _ count
Ch _ load= (2)
From the point of view of a node, the channel is in use, i.e. busy, scan _ count
when the node is either transmitting or receiving a packet from the
channel, or if it senses any transmission energy that hinders CLAW j t =1−×CLAW j t −1×Ch _ load j (3)
successful transmission such as those resulting from collisions,
interference, or other forms of noise. In addition, the channel may where :
CLAW j t The value of CLAW at timet
likewise be considered busy when the node is blocked from
accessing the channel, such as due to the back-off and defer periods α isa tunable parameter :0≤ α ≤ 1,here 0.5is used
in the distributed coordination function (DCF) in the IEEE 802.11 Ch _ load j isthe current observed channel load at node j
standard [35]. If all these events can be classified into one of two CLAW j t−1 isthe previous CLAW
fractional components of time, called TsensedEnergy and t refers tothe current measuring period
TblockedForAccess, then channel load is the total fraction of time that a
node is busy due to any of these contributing events. Equation (1) CLAW P t = ∑ CLAW j t 
expresses this definition of channel load. (4)
j∈P
We derived this definition from the result of a simple experiment
with three IEEE 802.11b nodes placed within a single collision where :
domain. In the experiment, a node Node0 sent packets to another CLAW P t  isthe equivalent path metric based on CLAW
node Node1 until channel saturation, while a third node Node2
1.2
silently observed. Although the physical layer of all three nodes
sensed the channel with the same degree of actual utilization (i.e.
amount of time packets occupied the channel), the sender Node0 was 1
loaded/busier (see Fig. 2) than the the receiver Node1 and the
Estimated
observer Node2, all the way through saturation, because of the Channel Load
blocking time (back-off and defer periods) in the DCF functionality 0.8
percent of time

Estimated
of IEEE 802.11b[35]. At saturation, although the sender viewed Packet In the Air
Mac Load of
channel load to be 100% the receiver and observer only viewed the 0.6 Node 0
channel as around 78% loaded. It is interesting to note that the 78% Mac Load of
load approximated the ratio of time the packets propagating in the air Node 1
occupied the channel. This is comparable to the throughput 0.4 Mac Load of
Node 2
saturation encountered at around 80% channel busyness by others
[36]. Generally, saturation throughputs have not been achieved at
100% busyness [36], [37] as may be intuitively expected from such a 0.2
metric, because the back-off and defer periods in the IEEE 802.11
MAC protocol were not taken into account. In contrast, by taking 0
these into account, the CLAW metric is able to account for the
0 1 2 5 8
missing ~20% busyness. Thus, not only can CLAW effectively
input traffic (Mbps)
identify busy regions, in addition, it can discriminate between loaded
and non-loaded nodes within such busy regions. Figure 2: Channel Load Measurement

It is also worth noting that we do not make any assumption about V. CONCLUSIONS
the operating channel of a collision domain. Our analysis only
require that i and j's collision domain operate on the same channel. If We propose the channel-load aware (CLAW) routing metric to
some collision domains operate over different channels the analysis address issues in congestion, interference and load-imbalance
will follow the same process. In addition, the analysis (TsensedEnergy problem in wireless mesh networks. CLAW does not require
and TblockedForAccess) will still be valid had a different mac layer complex computations, nor any exchange or collection of neighbor-
technology been used. Thus, CLAW is suitable for both single- and wide link-level statistics. Its simplicity allows it to be easily
multi-radio and multi-channel mesh networks. integrated, if necessary, with other capacity-aware routing metrics
with minimal overhead. Analysis also shows it is suitable to both
B. Implementation single-and multi-channel or multi-radio mesh networks.
To estimate the channel load, we simply monitor how the MAC Initial simulation results demonstrated its ability to effectively
layer views the channel. The MAC layer senses the busyness of the estimate channel busyness and enable flows to avoid congested
channel through carrier sensing (provided by the physical layer) and regions.
virtual carrier sensing through its NAV (network allocation vector) Although it shows promise, our initial comparison with hop-count
[35]. Within a defined observation period the MAC layer is queried routing merely demonstrates CLAW's basic ability to support
whether it senses the channel to be busy, backing-off, or deferring. congestion-free routing. A more comprehensive performance
The number of times where the MAC layer reports any of these three comparison with similar congestion-aware metrics is therefore in
conditions (busy_count), divided by the number of times the MAC order. Ultimately, the usefulness of this metric can only be fully
layer is queried (scan_count) becomes the estimated channel load as realized through actual, working implementations, rather than
defined in Eq. (2). It is interesting to note that the channel load through theoretical simulations. We will hopefully address all of
computed using Eq. (2) consistently matched the estimated channel these in our future work.
load (for Node0) and actual fractional packet-in-the-air time (for
Node1 and Node2) as observed and presented in Fig. (2). ACKNOWLEDGMENT
To account for sudden changes in traffic and the dynamic
This work has been supported by the Engineering Research and
behavior of the wireless channel, we employ a moving average for
Development for Technology (ERDT) Consortium, Department of
the channel load using a tunable parameter α. We initially used
α=0.5, although further experimentation and study may suggest other Science and Technology – Science Education Institute (DOST-SEI),
Republic of the Philippines.
values. The CLAW metric is thus defined in Eq (3) as the moving
average of the estimated channel load. Equation (4) is the equivalent 20 21 22 23 24
path metric based on CLAW.

IV. SIMULATION AND DISCUSSION 16 17 18 19
15
We performed preliminary qualitative and quantitative
experiments to evaluate the performance of our proposed routing
metric using ns-2 [38] with the OLSR extension as used in [39]. We 10 11 12 13 14
wanted to quickly test whether our metric would in fact avoid busy
regions of the network, and whether it would achieve better
throughput compared to hop count-based routing. 05 06 07 08 09
The first set of simulations were designed to show whether the
CLAW metric could steer flows away from loaded regions of the
network. The set-up shown in Fig. 3, similar to that in [36] involved
00 01 02 03 04
25 nodes uniformly distributed in a grid of 800 x 800 square meters.
Data-rates between nodes is set to 11 Mbps. For the main traffic Figure 3: Node 00 originates an FTP flow towards Node 24.
flow, FTP bulk traffic over TCP was used, with the packet size set of CBR traffic between Node 11 and Node 12 form an
1040 bytes (NS2 default size [38]). Constant bit rate (CBR) is used interference flow. While hop-count based routing would
for the interference flow. To simplify the simulation both result in the straight-line path 00-06-12-18-24, CLAW
transmission range and sensing range were set to 250m, while the routes the flow through 00-01-02-03-09-14-19-24,
distance between nodes was set to 176 m. At the start of the avoiding busy regions
simulation, the interference flow between nodes 11 and 12 was
initiated, creating the busy region indicated by the two circular areas 600 CLAW
in Fig. 3. With a traditional hop count metric, packets traversed the HopC
path 00-06-12-18-24. With CLAW, packets followed the path 00- 500
01-02-03-09-14-19-24, effectively avoiding the busy region in the
400
Throughput (kbps)

network.
In the second set of simulations, the interfering traffic was varied 300
from 0, 0.5 Mbps, 1 Mbps, 1.5 Mbps, …, 5 Mbps in order to observe
network behavior and performance with varying degrees of 200
busyness. Fig. 4 compares the throughput attained by the main flow
with hop count and CLAW routing metrics. Each data point in the 100
graph represents the average from 10 simulation runs. The dramatic
decrease in the throughput of the network that used hop count 0
routing, especially around 2-2.5 Mbps interference traffic, was due to 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
packet drops within the busy region. In contrast, CLAW was able to Interfering Traffic (Mbps)
avoid the busy region, resulting in significantly better end-to-end
throughput even with high levels of busyness within the network. Figure 4: Throughput comparison between CLAW and Hop-
count metrics

REFERENCES [21] V. Saigal, A. K. Nayak, S. K. Pradhan, and R. Mall, “Load balanced
routing in mobile ad hoc networks,” Computer Communications, vol.
[1] C. Reinisch, W. Kastner, G. Neugschwandtner, and W. Granzer, 27, no. 3, pp. 295-305, Feb. 2004.
“Wireless Technologies in Home and Building Automation,” in 2007 [22] H. Hassanein and A. Zhou, “Routing with load balancing in wireless Ad
5th IEEE International Conference on Industrial Informatics, 2007, vol. hoc networks,” in Proceedings of the 4th ACM international workshop
1, pp. 93-98. on Modeling, analysis and simulation of wireless and mobile systems,
[2] E. Callaway et al., “Home networking with IEEE 802.15. 4: a Rome, Italy, 2001, pp. 89-96.
developing standard for low-rate wireless personal area networks,” [23] G. Karbaschi and A. Fladenmuller, “A link-quality and congestion-
Communications Magazine, IEEE, vol. 40, no. 8, pp. 70–77, 2002. aware cross layer metric for multi-hop wireless routing,” in In Proc. of
[3] Carol Ellison, “Municipal Broadband: A Potential twenty-first century IEEE International Conference on Mobile Adhoc and Sensor Systems
utility,” New York University Journal of Legislation and Public Policy, Conference, Washington, DC, 2005.
2008. [24] H. Aiache, V. Conan, L. Lebrun, and S. Rousseau, “A load dependent
[4] F. Bar and N. Park, “Municipal Wi-Fi networks: The goals, practices, metric for balancing Internet traffic in Wireless Mesh Networks,” in 5th
and policy implications of the US case,” Communications & Strategies, IEEE International Conference on Mobile Ad Hoc and Sensor Systems,
vol. 61, no. 1, pp. 107–124, 2006. 2008. MASS 2008, 2008, pp. 629–634.
[5] A. Powell and L. R. Shade, “Going Wi-Fi in Canada: Municipal, and [25] D. M. Shila and T. Anjali, “Load aware traffic engineering for mesh
Community Initiatives,” Government Information Quarterly, vol. 23, no. networks,” Comput. Commun., vol. 31, no. 7, pp. 1460-1469, 2008.
3-4, pp. 381–403, 2006. [26] Y. Yang, J. Wang, and R. Kravets, Interference-aware Load Balancing
[6] V. Naumov and T. Gross, “Connectivity-Aware Routing (CAR) in for Multihop Wireless Networks. Illinois: University of Illinois at Urban-
Vehicular Ad-hoc Networks,” in in Proc. 26th IEEE International Champaign, 2005.
Conference on Computer Communications, 2007. [27] A. P. Subramanian, M. M. Buddhikot, and S. Miller, “Interference
[7] H. Aiache et al., “Widens: Advanced Wireless Ad-Hoc Networks for aware routing in multi-radio wireless mesh networks,” in 2nd IEEE
Public Safety,” in in Proc 14th Mobile & Wireless Communication Workshop on Wireless Mesh Networks, 2006. WiMesh 2006, 2006, pp.
Summit, Dresden, Germany, 2005. 55-63.
[8] K. Kanchanasut, A. Tunpan, M. Awal, T. Wongsaardsakul, D. Das, and [28] L. Ma and M. K. Denko, “A Routing Metric for Load-Balancing in
Y. Tsuchimoto, Building A Long-distance Multimedia Wireless Mesh Wireless Mesh Networks,” in Advanced Information Networking and
Network for Collaborative Disaster Emergency Responses Operation Applications Workshops, 2007, AINAW ’07. 21st International
in Disaster-affected Areas. Thailand: nternet Education and Research Conference on, 2007, vol. 2, pp. 409-414.
Laboratory,, Asian Institute of Technology, 2007. [29] Jiahui Zhu, Fuqiang Liu, Xinhong Wang, and Lu Rong, “A congestion
[9] T. Trimble, R. Mishra, B. S. Manoj, R. Rao, L. Lenert, and B. and interference aware routing metric for Wireless Mesh networks,” in
Braunstein, “Feasibility of using distributed Wireless Mesh Networks Communication Technology, 2008. ICCT 2008. 11th IEEE International
for medical emergency response.,” AMIA Annual Symposium Conference on, 2008, pp. 62-65.
proceedings AMIA Symposium AMIA Symposium, vol. 2006, pp. 86-90. [30] A. A. Pirzada, R. Wishart, M. Portmann, and Jadwiga Indulska,
[10] N. J. Macabale, R. Ocampo, and C. A. Festin, “Attainable Capacity “ALARM: An Adaptive Load-Aware Routing Metric for Hybrid
Aware Routing Metric for Wireless Mesh Networks,” in In Proc. of the Wireless Mesh Networks,” in in Proc. 32nd Australasian Computer
Second International Conference on Adaptive and Self-adaptive Systmes Science Conference, Wellington, New Zealand, 2009.
and Applications 2010, Lisbon, Portugal, 2010. [31] Qiming Tian, “A New Interference-Delay Aware Routing Metric for
[11] T. Rappaport, Wireless Communications: Principles and Practice (2nd Multi-Interface Wireless Mesh Networks,” in Wireless Communications
Edition). Prentice Hall PTR, 2002. Networking and Mobile Computing (WiCOM), 2010 6th International
[12] Y. Yi, T. J. Kwon, and M. Gerla, “A load aWare routing (LWR) based Conference on, 2010, pp. 1-5.
on local information,” in 2001 12th IEEE International Symposium on [32] S. Waharte, B. Ishibashi, R. Boutaba, and D. Meddour, “Interference-
Personal, Indoor and Mobile Radio Communications, 2001, vol. 2, p. G- Aware Routing Metric for Improved Load Balancing in Wireless Mesh
65-G-69 vol.2. Networks,” in Communications, 2008. ICC ’08. IEEE International
[13] A.-N. LE, D.-W. KUM, Y.-Z. CHO, and Y.-C.-K. TOH, “Routing with Conference on, 2008, pp. 2979-2983.
Load-Balancing in Multi-Radio Wireless Mesh Networks,” IEICE [33] J.-H. Song, V. Wong, and V. C. M. Leung, “Load-aware on-demand
TRANS. COMMUN., vol. 92, no. 3, pp. 700-708, Mar. 2009. routing (LAOR) protocol for mobile ad hoc networks,” in Vehicular
[14] D.-I. Choi, J.-W. Jung, K. Y. Kwon, D. Montgomery, and H.-K. Kahng, Technology Conference, 2003. VTC 2003-Spring. The 57th IEEE
“Design and Simulation Result of a Weighted Load Aware Routing Semiannual, 2003, vol. 3, pp. 1753- 1757 vol.3.
(WLAR) Protocol in Mobile Ad Hoc Network,” in Information [34] O. Al-Jarrah and M. S. Al-Hadrusi, “Load Balanced Ad Hoc Routing
Networking, 2005, pp. 178-187. Protocol,” presented at the in Proc. of the First Mobile Computing and
[15] J.-W. Jung, D. Choi, K. Kwon, I. Chong, K. Lim, and H.-K. Kahng, “A Wireless Communication Interantional Conference, Amman, 2006, pp.
Correlated Load Aware Routing Protocol in Mobile Ad Hoc Networks,” 20-26.
in Universal Multiservice Networks, vol. 3262, M. M. Freire, P. [35] IEEE, “802.11: Wireless LAN Medium Access Control (MAC) and
Chemouil, P. Lorenz, and A. Gravey, Eds. Berlin, Heidelberg: Springer Physical Layer (PHY) Speciﬁcations- Revision of the 802.11-1999
Berlin Heidelberg, 2004, pp. 227-236. standard,” 2007.
[16] S.-J. Lee and M. Gerla, “Dynamic load-aware routing in ad hoc [36] L. T. Nguyen, R. Beuran, and Y. Shinoda, “A load-aware routing metric
networks,” in Communications, 2001. ICC 2001. IEEE International for wireless mesh networks,” in Computers and Communications, 2008.
Conference on, 2001, vol. 10, pp. 3206-3210 vol.10. ISCC 2008. IEEE Symposium on, 2008, pp. 429-435.
[17] L. T. Nguyen, R. Beuran, and Y. Shinoda, “An interference and load [37] E. Pelletta and H. Velayos, “Performance Measurements of the
aware routing metric for Wireless Mesh Networks,” International Saturation Throughput in IEEE 802.11 Access Points,” in Third
Journal of Ad Hoc and Ubiquitous Computing, vol. 7, pp. 25–37, Dec. International Symposium on Modeling and Optimization in Mobile, Ad
2011. Hoc, and Wireless Networks (WiOpt’05), Riva del Garda, Trentino,
[18] X. Gao, X. Zhang, D. Shi, F. Zou, and Wenbo Zhu, “Contention and Italy, pp. 129-138.
Queue-Aware Routing Protocol for Mobile Ad Hoc Networks,” in [38] “The Network Simulator 2 - NS2.” [Online]. Available:
Wireless Communications, Networking and Mobile Computing, 2007. http://www.isi.edu/nsnam/ns/. [Accessed: 21-Jul-2009].
WiCom 2007. International Conference on, 2007, pp. 1628-1631. [39] W. Cordeiro, E. Aguiar, W. Moreira, A. Abelem, and M. Stanton,
[19] Y. Yuan, Huimin Chen, and Min Jia, “An adaptive load-balancing “Providing Quality of Service for Mesh Networks Using Link Delay
approach for ad hoc networks,” in Proceedings. 2005 International Measurements,” in Proceedings of 16th International Conference on
Conference on Wireless Communications, Networking and Mobile Computer Communications and Networks, 2007. ICCCN 2007, 2007,
Computing, 2005, vol. 2, pp. 743 - 746. pp. 991-996.
[20] K. Wu and J. Harms, “Load-sensitive routing for mobile ad hoc
networks,” in Proceedings, Tenth International Conference on
Comuputer Communications and Networks, 2001, pp. 540 - 546.

Congestion Free Routes for Wireless Mesh Networks

More Related Content

What's hot (19)

Viewers also liked (6)

Similar to Congestion Free Routes for Wireless Mesh Networks (20)

Congestion Free Routes for Wireless Mesh Networks