Research Paper 1

Journal of Interconnection Networks W K 3 World Scientific
Vol. 8, No. 4 (2007) 407-426 V S ^ www.woridscientific.com

© World Scientific Publishing Company
E N H A N C E D R O U T I N G METRIC FOR LOAD-BALANCING IN

WIRELESS MESH N E T W O R K S
LIANG MA
Department of Computing and Information Science,
by UNIVERSITY OF AUCKLAND LIBRARY - SERIALS UNIT on 02/25/15. For personal use only.
University of Guelph, 50 Stone Road East,

Guelph, NIG 2Wl,Canada
lma02@uoguelph. ca
J. Inter. Net. 2007.08:407-426. Downloaded from www.worldscientific.com
MIESO K. DENKO
Department of Computing and Information Science,
University of Guelph, Guelph, NIG 2W1,Canada
lma02@uoguelph. ca
Received 22 June 2007

Revised 18 September 2007
Wireless mesh networks (WMNs) have been drawing significant attention in recent years
due to their flexibility in providing extensive wireless backbone. WMNs typically consist
of mesh routers and mesh clients with each node operating not only as a host but also as a
router. Due to the traffic patterns in WMNs, load-balancing becomes an important issue
and may degrade the performance of the entire network. This paper proposes a routing
metric known as Weighted Cumulative Expected Transmission Time with Load-Balancing
(WCETT-LB) for wireless mesh networks. WCETTT-LB enhances the basic Weighted Cu-
mulative Expected Transmission Time (WCETT) by incorporating load-balancing into the
routing metric. Unlike existing schemes, WCETT-LB implements load-balancing at mesh
routers. WCETT-LB provides a congestion-aware routing and traffic splitting mechanism
to achieve global load-balancing in the network. By conducting an extensive simulation
experiments, the result shows that WCETT-LB outperforms the existing routing metrics
in load-balancing in terms of achieving high packet delivery ratio, low average end-to-end
delay and low average congestion level in wireless mesh networks. The qualitative and
quantitative analysis also show the significance of the proposed scheme.
Keywords: Wireless mesh network; load-balancing; routing metrics.
1. Introduction
Wireless mesh networks (WMNs) typically consist of mesh routers and mesh clients
with each node having the capability of operating not only as a host but also as a
router. Based on the functionality of the nodes, WMNs can be classified into three
categories: Infrastructure backbone, client backbone and hybrid [6,7]. Mesh routers
are used to form a multi-hop and multi-path wireless relay backbone capable of
communicating with gateways and clients. Mesh clients can form a self-organized
ad hoc networks which can access services by relaying requests to wireless backbone
407
408 L. Ma & M. K. Denko
network. The hybrid mesh network architecture is a combination of infrastructure

and client meshing and is expected to be the best choice in next generation WMNs.
Wireless mesh networks have been drawing significant attention in recent years
due to their flexibility in providing extensive wireless backbone[8] to the netork. The
potential applications include wireless broadband services, community networking,
instant surveillance systems, high speed metropolitan area networks, and back-haul
service for large-scale wireless sensor networks.
Some of the technical challenges in WMNs are load-balancing, optimal routing,
fairness, network auto-configuration and mobility management. Our focus in this

paper is routing and load-balancing. Existing solutions in mobile ad hoc and sensor
networks cannot be directly applied to WMNs due to the difference in traffic pat-
terns, mobility scenarios, gateway functionalities and bandwidth requirements. Since
most users in WMN are primarily interested in accessing the Internet or other com-
mercial servers, the traffic in WMNs is routed either toward the Internet gateways
(IGWs) or from the IGWs to clients . Thus, if multiple edge mesh routers choose the
best throughput path toward a gateway, the traffic loads on certain paths and mesh
routers increases tremendously thereby significantly decreasing the overall perfor-
mance of the network. The routing algorithm therefore needs to determine routes
between each traffic access point in a way that balances the load on the entire mesh
network. Effective load-balancing mechanisms can help in avoiding congestion and
can increase the efficiency of network resource utilization. Therefore, the motiva-
tion of this paper is to provide a routing scheme which overcomes the problem of
load-balancing.
In order to achieve load-balancing in WMNs, suitable routing protocols should
be designed. Existing schemes for load-balancing in wireless mesh networks con-
sider load-balancing at the Internet gateways [10,7,9,2]. This paper proposes load-
balancing at mesh routers and also introduces a traffic splitting algorithm to divide
the traffic among mesh routers. The main contributions of this paper are: (1) We pro-
pose a routing metric that provides load-balancing at mesh router and; (2) We intro-
duce a dynamic traffic splitting algorithm to balance load distribution among mesh
routers. Our proposed scheme also provides a mechanism for handling intra-flow
and inter-flow interference in the network. From simulation results using Network
Simulator 2, the proposed routing metric and scheme achieved better performance
in load-balancing than previous routing metrics.
The rest of this paper is organized as follows: Section 2 discusses existing related
work. Section 3 presents the proposed routing metric and traffic splitting algorithm.
The architecture of the wireless mesh network is also presented. Section 4 presents
the analysis and discussion of the simulation results. Both qualitative and quantita-
tive results are presented. Finally, Section 5 presents conclusions and some ongoing
experimental work related to the proposed scheme with the indication of future
research direction.
Enhanced Routing Metric for Load-Balancing in Wireless Mesh Networks 409
2. Related Work
Due to the many interacting parameters such as network load, link transmission
rate, intra-flow and inter-flow interferences and link dynamics, routing in wire-
less mesh networks is a challenging problem. Load-balancing in WMNs can be
achieved through path-based load-balancing, gateway-based load-balancing or mesh
router-based load-balancing [12,10,7]. In path-based load-balancing, the traffic is
distributed across multiple paths. In gateway-based load-balancing, the load is bal-
anced either at all Internet gateways or at some selected gateways. Load-balancing
can also be carried out at the mesh routers over the wireless backbone.
In this section we will briefly discuss recently proposed routing metrics and multi-
path routing schemes for WMNs. Routing metrics are critical for determining the
performance of the networks. A good metric should contain sufficient information

about the link or the routing path. Each node in the network chooses the best path
in terms of all the properties contained in routing metric. Recently proposed routing
metrics for mesh networks include hop count, Expected Transmission count (ETX),
Expected Transmission Time (ETT), Weighted Cumulative ETT (WCETT) and
Metric of Interference and Channel-switching (MIC) [3,11,14,15].
The hop count metric reflects only the effects of path lengths in the network.
The routing metric ETX captures both the packet loss ratios and path length by
counting number of MAC layer transmissions for successfully delivering a packet
through a wireless link. ETT considers the differences in link transmission rate
for different links in wireless mesh networks. It enhances ETX by integrating the
data transmission rate of each particular link into the routing metric. WCETT is a
routing metric which considers both the difference in link bandwidth and diversity
of channel assignment.
WCETT enables nodes in the network to set up the best path of channel diversity
and link bandwidth to IGWs. MIC [15] improves WCETT by catching both intra-
flow interference within a path and inter-flow interference between adjacent paths. It
considers both the channels used in the current link and in the previous link within a
path. If the current link is using the same channel as its previous link, then MIC will
assign a larger parameter value to the current link in order to capture the intra-flow
interference within each path. In order to capture inter-flow interference, MIC also
catches the set of neighbors that the transmission on each current link interferes
with. Hence, packets can be routed through routers with less traffic concentration.
However, these routing metrics do not consider load-balancing.
Multi-path routing protocols can be used to provide load-balancing in wireless
networks [4,10]. In order to achieve load-balancing, each node needs to maintain
multiple paths from itself to the IGWs. If the current best path is heavily loaded or
congested, the node can switch to the second best path efficiently. In [10], authors
proposed a multi-path routing protocol for balancing traffic over multiple paths and
improving the overall network performance. The proposed multi-path protocol es-
tablishes paths at each node to IGWs by selecting maximal disjoint paths. However,
"^ ^ " Wired Link

Wireless Link Internet )
| | Internet Gateway
'' ^ \
(/) Wireless Mesh Router
S<
A Mesh Client „-'"" i - " " ^ " i"""- Wireless Backbone
GO Gl G2 " ^
*
/'" ^
);;"""
Xp:;'"
"M" ;@;"" / ., A......4 )

\5
'% 5' „,--''
A A A
A' A '"A".. 4 A
A' -A A A s,
Fig. 1. A three-layer wireless mesh network architecture
the scheme does not consider load-balancing at mesh routers.
Table 1. Notations used in the paper.

Notation Description
a Congestion level threshold for mesh routers.
5 Load-balancing threshold for path switching in a mesh network.
Ni Set of children nodes using node i as their next hop in paths.
G% Internet Gateway i in the network.
WCETT — LB\urrent WCETT-LB of current path that node i chose to IGW.
WCETT-LBlest WCETT-LB of current best path in terms of traffic load and
throughput from node i to IGW.
3. The Proposed Load-balancing Scheme

In this paper, we propose a routing metric and a traffic splitting algorithm to provide
load-balancing in WMNs. The proposed routing metric known as Weighted Cumu-
lative Expected Transmission Time with Load-balancing (WCETT-LB) is based on
the WCETT routing metric. WCETT-LB introduces load-balancing feature at the
mesh routers and supports global load-aware routing. The integration of a load-
balancing metric to WCETT and the global congestion-aware routing scheme can
provide performance improvement in the entire network.
Our proposed load-balancing scheme is based on the architecture shown in Figure
1. In this architecture, the upper layer is the wired Internet with the IGWs. The
middle layer consists of wireless mesh routers connected to both the IGWs and
entities in the third layer (mesh clients). Mesh routers connect to IGWs with wireless
links. The mesh routers automatically form a wireless backbone network to provide
Internet connectivity for all mesh clients. In the client layer, mesh clients can request
Internet services by relaying their requests to the wireless backbone. Mesh clients
can be connected to wireless backbone through a single hop or multi-hop routing
scheme. The notations in Table 1 are defined to simplify the presentation of the
protocol.
3.1. WCETT-Load Balancing Metric

In the proposed approach, we integrate load-balancing component into WCETT

metric. The load-balancing component consists of two parts: congestion level and
traffic concentration level at each node in a particular path. The congestion level
at each node is evaluated by considering the average queue length at each nodes in
a particular path. If the average queue length is higher than a threshold, then the
path is heavily loaded. So, we can monitor the loads on all paths by evaluating their
nodes congestion level as follows.
£ f1 (3-D
nodeiEp
In Formula 3.1, QLi is the average queue length at a node in a particular path
and bi is the transmission rate at a node. We take the fraction of the average queue
length over the transmission rate to get the actual time needed for transmission. We
consider the traffic concentration of each node by using Ni. So, if more child nodes
choose i as their next-hop to transmit packets, the traffic at the node increases.
Therefore, such a node has a higher probability of becoming congested in the net-
work. We normalize the value so that we can integrate the component to WCETT.
This can be achieved as follows.
Y^ min(£TT)iVj (3.2)
nodeiEp
The quantity mm(J5TT) is the smallest ETT in the network. The higher value
of ram(ETT)Ni indicates more traffic concentration at node iVj. We add the values
of all nodes to capture the traffic concentration level in a path. Hence, this model
tends to make source node choose the more independent paths in the network. Now,
we formulate our routing metric for load-balancing as follow:
WCETT - LB{p) = WCETTip) + Up) (3.3)
where
i
L(p)= Y, ^ + min(ETT)Ni
nodeiEp
Thus, in addition to all the features of WCETT, WCETT-LB considers the

traffic concentration and congestion level at all nodes in path p. If a particular
path is heavily loaded, WCETT-LB can captures this situation. It can be further
combined with our global congestion-aware routing to balance the load at all mesh
routers in the network.
3.2. Global Congestion-Aware Routing

Previous work on traffic splitting algorithm only balance the load between paths
based on the neighbors' congestion level in the network. So, since the load-balancing
is done locally at each mesh router, packets may still be routed to highly congested
nodes located further in the path. We propose a global congestion-aware routing

scheme to make sure that each node can be aware of the congestion level along all
paths to IGWs. Nodes can choose the best paths to balance the traffic based on the
received information.
3.2.1. Congestion-Aware Routing

We introduce a congestion level threshold a to determines whether a particular
node in the network is congested or not. A mesh router is said to be congested if
its congestion level %-*• > a. We use threshold a to reduce periodic broadcasting
of congestion information to the neighbors since if a node is not congested, com-
munication with neighbors periodically results in wasting network bandwidth. The
proposed algorithm let the node determine its congestion level by itself reducing
communication overhead. Basically, a node i will compute its own ^-* periodically.
If ^M < a, then we assume that the load is balanced at node i. But if its congestion
level is greater than the a threshold, then the node will update this information
by re-computing WCETT-LB. We can achieve this by letting each node remember
the old WCETT-LB from itself to IGWs during the path establishment phase. After
re-computed WCETT-LB, this congested node will multicast the updated WCETT-
LB routing metric to all the nodes in iVj. So, all the neighbors which are currently
using node i as their next hop can be aware of congestion in the path by receiving
this information. All the nodes in iVj will further multicast this congestion informa-
tion in the same fashion until this congestion information is received by edge mesh
routers.
Algorithm 1 shows the congestion-aware process at each mesh router. Each node
determines its own congestion level by computing ^-^ periodically.
As shown in algorithm 1, each mesh router has two states: congested and load-
balanced. If ^-^ > cr at node i, then this mesh router is in congested state. If the
average queue length of this mesh router is back to normal load level, then the mesh
router will notify its recovery to all its previous child nodes.
Algorithm 1
At each mesh router i:
Compute ^-^ periodically;
I f ( £ k > a at node i)
then re-compute WCETT-LB;
multicast WCETT-LB to Nf,
Else
Load is balanced at mesh router i;
3.2.2. Load-Balancing
Each time an edge mesh router receives updated WCETT-LB from its neighbor,
it knows that congestion occurs in the path towards IGW. The global congestion
awareness routing allows the mesh router to balance the load along all its paths.
The mesh router can either choose to stay on current path or switch to other paths
that have better performance. Algorithm 2 shows the path switching process at
each mesh router and it is used to determine the network load. Each node compares
WCETTLBiurrent with WCETTLBlbest after it receives updated WCETT-LB. If
(WCETTLBiurrent - WCETTLBlbest) > 8, then the mesh router will switch from
the current path to the new best path. Otherwise, the mesh router can chose to stay
on the current path.
Algorithm 2
At each edge mesh router i:
If i received updated WCETT-LB
and (WCETTLBiurrent-WCETTLBlst)>5
then switching is made
Else
Load is balanced in all the paths at mesh router i;
There are two main advantages of the control parameter 8. It prevents mesh
routers from frequently switching paths. If the mesh router switches paths each
time when it receives WCETT-LB, then the mesh router will keep switching between
paths when the load changes. This will decrease network performance dramatically.
Furthermore, 8 prevents all the traffic away from the congested nodes. If all edge
mesh routers choose to switch to other paths then, there may be no traffic through
the congested node. This is not a desirable load-balancing approach. Hence, by using
the parameter 8, some of the edge mesh routers can switch to other path if the new
path is better than the current path.
Wired Link
Wireless Link
& ' -
&.. •®
-¥-
©.
(!)
•-•©-
Fig. 2. An example wireless mesh network.
We now illustrate the operation of our protocol using an example. Consider

a wireless mesh network architecture in Figure 2. The network consists of three
gateways GO, Gl and G2 and nine mesh routers 1-9. Mesh router 1 and mesh router
3 are connected to GO. Mesh router 3 is also connected to Gl. Mesh router 3 and
mesh router 4 are connected to G2. Instead of using other routing metric, we use
our WCETT-LB to set up paths. So, mesh router 8 will choose mesh router 4 as
its next-hop, since mesh router 2 and mesh router 3 may have more concentrated
incoming traffic and they have higher probability of congestion.
After path establishment phase, mesh router 6 has two paths to IGWs, 6-2-3 and
6-5-1 where and 6-2-3 is the default path. Mesh router 9 has two paths to IGWs,
9-8-3 and 9-8-4 where 9-8-3 is the default path. Now, if we assume that mesh router
3 is congested. Then, mesh router 3 will re-compute WCETT-LB and multicast the
routing information to mesh router 2 and mesh router 8. Mesh router 2 and mesh
router 8 will further pass this information to mesh router 6 and mesh router 9.
Mesh router 6 and mesh router 9 will decide to stay on the current path or switch
to another path by comparing the congested path with the second best path. In this
case, mesh router 6 may choose to stay on the current path and mesh router 9 may
choose to switch to the path 9-8-4 if the difference is greater than the threshold.
If mesh router 5 and mesh router 2 are also congested in future, then mesh router
6 can choose the best path based on global congestion awareness. It will select the
least congested path, since path 6-2-3 could be highly congested. Therefore, load-
balancing can be achieved in the network.
4. The analysis and discussion of simulation results

Since our load-balancing routing metric WCETT-LB integrates load-balancing com-
ponent into traditional WCETT, we preserve all the features and properties of
WCETT. The load-balancing component enhances the performance of WCETT by
considering traffic congestion. Our global congestion-aware routing also enhances
several features of congestion-aware routing protocol proposed in [10]. In our scheme

each node receives the congestion information along all paths to IGWs. Introducing
the load-balancing threshold 5 in our algorithm can prevent mesh routers from fre-
quently switching paths back and forth (i.e. ping-pong effect). Another advantage is
that our algorithm allows each node to determine its congestion level instead of pe-
riodically sending the average queue length to its neighbors. This can greatly reduce
traffic overhead in network and improve the network performance.
Table 2. Comparison of existing routing metrics.

Routing Metrics LB Inter-FTI Intra-FTI
ETT No No No
WCETT No No Yes
MIC No Yes Yes
WCETT-LB Yes Yes Yes
We compared our load-balancing routing metric with other three existing metric
through a qualitative measures as shown in Table 2. Four routing metrics and three
parameters namely, load-balancing (LB), inter-flow interference (Inter-FTI), intra-
flow traffic (Intra-FTI) were used. Among all these routing metrics, only WCETT-
LB considers load-balancing in the network. The rest of the metrics do not consider
load in each path which is an important parameter for network performance. It can
be noted that only MIC and WCETT-LB metrics capture the inter-flow interferences
between paths in the network. Although WCETT-LB does not directly consider the
inter-flow interference, it can be easily adapted to catch the interference between
nodes. Furthermore, for intra-now interferences WCETT, MIC and WCETT-LB
metrics perform well. They consider the interference within each path. The other
possible parameter is isotonic property of each routing metric. Only ETT is isotonic,
but it performs poor with respect to other parameters. MIC is not directly isotonic,
but it has been shown to be isotonic by introducing virtual nodes. Our load-balancing
metric is isotonic, so it can be integrated with other isotonic routing metric. Overall
results show that our WCETT-LB is the only metric that addresses all the metrics
load-balancing, inter-flow interference and intra-flow interferences. We have also

performed preliminary quantitative experiments. To this end, the performance of
our proposed routing metric was evaluated using NS2 [13] simulation tool.
4.1. Performance Metrics

In our simulations, four performance metrics are used to evaluate the performance
of our proposed routing metric. They are delivery ratio, end-to-end delay, average
congestion level and standard deviation of traffic volumes at mesh routers.
(1) Packet delivery ratio (PDR): We compute the packet delivery ratio of rout-
ing metric by dividing total number of packets successfully delivered over total
number of packets originally sent in the network.

(2) End-to-end delay: We define the end-to-end delay to be the time taken for a
packet to be transmitted from source to destination in a network.
(3) Average congestion level: Since the idea of performing load-balancing in WMNs
is to reduce or eliminate congestion in the network, we want to see the average
congestion level of each mesh router in our simulation runs. We define the av-
erage congestion level to be the average queue length which is the number of
packets waiting in the queues at each mesh router.
(4) Standard deviation of traffic volumes: This is a new performance metric that is
used to evaluate the performance of routing metric. We compute the standard
deviation of traffic volumes at mesh routers in WMNs. Basically, we observe the
traffic volume at each mesh router and we consider these values as a population.
Then, we perform statistical analysis to compute the standard deviation of this
population. We want to see the variance of traffic volumes at mesh routers. If
the computed standard deviation is close enough to zero, then we can conclude
that the load is truly balanced at each mesh router.
4.2. Simulation Parameters

We also change simulation parameters to test the performance of routing metrics
under different conditions. Three simulation parameters were used:
(1) Traffic load: This performance metric is evaluated under different level of traffic
load in the network. Constant Bit Rate (CBR) is constantly increased from 5
packets per second to 30 packets per second.
(2) Number of traffic flows: We want to test the performance of routing metrics by
changing the number of traffic flows in the network. From 2 to 18 nodes will
be randomly selected to transmit packets to IGWs in the network during the
entire simulation time.
(3) Simulation time: The performance of routing metrics will also be evaluated by
different period of simulation runs.
Enhanced Routing Metric for Load-Balancing in Wireless Mesh Networks A17
4.3. Simulation results and discussion

In this section, we present the experimental results that compare WCETT-LB with
hop count and WCETT. The performance of WCETT-LB routing metric is evalu-
ated using NS2 as network simulator. Simulation experiments are divided into three
series. Each performance metric is tested under different simulation parameters such
as traffic load, number of traffic flows and simulation time.
4.3.1. Simulation setup

Table 3. Network parameters for simula-

tion.
Network simulator NS2

Simulation area 1500mX1500m
Transmission range 250m
Carrier sensing range 550m
Wireless nodes 80
Traffic flows 10
Packet size 512 bytes
Transmission type CBR
In this paper, NS2 which is a discrete event network simulator is used to evaluate
the performance of our proposed scheme. There are certain environment variables
that need to be set before simulation start. These variables are simulation area,
wireless transmission range of each node, carrier sensing range, number of wireless
nodes in simulation area, number of generated traffic flows when simulation begin,
size of each packet and transmission type. Table 3 shows the values of network
parameters for our simulation environment. The simulation environment consisted
of three IGWs, 80 nodes with 10 traffic flow sources. The simulation was conducted
in an area of 1500 X 1500 square units. Due to the nature of wireless mesh network
traffic, all traffic flows were to or from the IGWs. The CBR traffic source were used.
We set the CBR packet size to 512 bytes. The transmission range was 250m with
550m carrier sensing range.
4.3.2. Simulation under different traffic load

In the first series of simulation runs, we tested each performance metric under dif-
ferent traffic load level. The traffic load was increased from 5 packets per second up
to 30 packets per second. The performance of WCETT-LB, WCETT and hop count
are compared and evaluated under these conditions.
First, we run our simulation and tested the packet delivery ratio of hop count,
WCETT and WCETT-LB under the same CBR rate. As the CBR increases, the
traffic load in the network increased dramatically. As shown in Figure 3, the packet
delivery ratio of hop count drops a lot. And, the delivery ratio of WCETT is better
1
0.6 -I 1 1 1 1
5 10 15 20 25 30
Load (pkts/second)
Fig. 3. Packet delivery ratio under traffic load.
Fig. 4. Average end-to-end delay under traffic load.
than hop count, but it still dropped significantly. The reason is that congestion
happened among the paths in the network when using hop count and WCETT
as routing metric. Again, from Figure 3, WCETT-LB outperforms WCETT and
hop count in terms of achieving higher delivery ratio. The PDR of WCETT-LB is
relatively stable during the entire simulation time.
Average end-to-end delay is the second performance metric that was tested in
our simulation. Since WCETT-LB is designed to reduce congestion in wireless mesh
network, we expect that WCETT-LB can achieve lower end-to-end delay than the
Fig. 5. Average congestion level under traffic load.
others. From Figure 4, WCETT-LB had the lowest average end-to-end delay. These
three routing metrics achieved relatively the same end-to-end delay before load
reached 10 packets/sec. As load increases, end-to-end delay for hop count increased
dramatically. Although WCETT also had low end-to-end delay, it does not con-
sider load-balancing. The hop count metric performed the worst among these three
routing metrics.
In the next experiment, we investigated average congestion level using simula-
tion. For each level of CBR, average congestion level in terms of number of pack-
ets waiting in queues is recorded and compared between hop count, WCETT and
WCETT-LB. As shown in Figure 5, the performance of these three routing met-
ric achieves relatively the same congestion level when CBR is less than 15. But,
hop count and WCETT had much higher congestion level when CBR is greater
than 15. The congestion level of WCETT-LB is lower and much stable than the
other routing metrics, because WCETT-LB considers the congestion level and traf-
fic concentration level in the network. WCETT-LB establishes paths in an effort to
select non-overlapping and most independent paths. If congestion happened in the
network, then WCETT-LB switches to other paths to avoid further congestion in
the path. Therefore, congestion can be eliminated or reduced in our load-balancing
scheme.
Standard deviation of traffic volume at each mesh router is a new performance
metric that we use to evaluate the performance of WCETT-LB routing metric. For
each level of CBR, we first record the traffic volume at each mesh router. Then we
can apply the statistical analysis to compute the standard deviation of these traffic
volumes. If the computed standard deviation is close enough to zero, then we can
conclude the load is balanced at each mesh router. From Figure 6, WCETT-LB
-V\CETT-LB
-WCETT
-Hop count
100
& so
10 15 20 25 30
Load (pkts/second)
Fig. 6. Standard deviation under traffic load.
achieved much lower standard deviation of traffic volume than the other routing
metrics. Since WCETT-LB tends to distribute network traffic among all the mesh
routers across the entire WMN, therefore the difference in traffic volume at each
mesh routers is small.
4.3.3. Simulation under different traffic flows

In this series of simulation runs, we tested each performance metric under different
traffic flows. The traffic flows increased from 2 to 18. The performance of WCETT-
LB, WCETT and hop count is compared and evaluated.
We distributed the traffic flows randomly across the network. As traffic flow
increases in the network, the traffic volume also increases. Therefore, nodes have
higher probability to be congested during simulation. Since traffic flows are randomly
distributed, certain area of network may eventually become a hot spot or bottleneck.
This situation really affects the performance of the entire network. From Figure 7,
the packet delivery ratios of these three routing metrics are computed under different
traffic flows. WCETT-LB adapts to the situation and achieved the best PDR than
the other two routing metric.
In the second simulation run of this series, average end-to-end delay of each
routing metric is considered. When the network traffic level is low, the end-to-end
delay of WCETT-LB is higher than hop count and WCETT. As shown in Figure 8,
average end-to-end delay of WCETT-LB is higher than the others when the number
of traffic flows is equal to 2 and 4. Since WCETT-LB may not select the best paths in
terms of best link transmission rate and bandwidth, so when the network is not busy,
its end-to-end delay is slightly higher than the others. But WCETT-LB achieved
much lower delay as the traffic flows increase in the network.
2 4 6 8 10 12 14 16 18
Number of traffic flows
Fig. 7. Packet delivery ratio under traffic flows.
Fig. 8. Average end-to-end delay under traffic flows.
In the third simulation runs, as shown in Figure 9, the performance of WCETT

and hop count is not acceptable. Hop count had an average congestion level of 27
packets. The average congestion of WCETT reached 25 packets. WCETT-LB had
the lowest average congestion level. Since WCETT-LB balances load at each mesh
router and distributes network traffic across the entire network, congestion level is
reduced due to the load-balancing scheme. Therefore, WCETT-LB outperforms hop
count and WCETT in this performance metric.
As the traffic flow increases in the network, mesh routers using WCETT-LB
26
-WCETT-LB
16 -WCETT
-Hop count
6 8 10 12 14 16 18
Fig. 9. Average congestion level under traffic flows.
350
-WCETT-LB
-WCETT
-Hop count
55
-1 1 1 J— —\ 1-
2 4 6 8 10 12 14 16 1E
Fig. 10. Standard deviation under traffic flows.
as routing metric maintained a low and stable level standard deviation of traffic
volumes. From Figure 10, the result is much better for WCETT-LB compared to
hop count and WCETT. Standard deviation of traffic volume was just over 100 for
WCETT-LB. The result for hop count was over 300 and 250 for WCETT. From
the result, WCETT-LB performs the best among these three routing metrics as
we expected. WCETT-LB really balanced the load between each mesh router and
distributed the traffic very well.
Fig. 11. Packet delivery ratio under simulation time.
4.3.4. Simulation under different time

In the third series of simulation runs, the performance of WCETT-LB is evaluated
using different simulation time. The simulation period varied from 30 seconds to 210
seconds in length. Simulation time is an important simulation parameter. We tested
WCETT-LB with WCETT and hop count using different simulation time in this
series of simulation. As previous experiments, each performance metric is evaluated
and compared in our discussion.
Under different simulation time, WCETT-LB achieved highest packet delivery
ratio in each category. As shown in Figure 11, due to the generated traffic volume, all
routing metrics experienced congestion in the network. But, WCETT-LB adapted
congestion situation very well. The PDR for WCETT-LB was still over 80 percent.
WCETT has a delivery ratio of 71 percent. Hop count performed the worst among
these three routing metrics.
From previous simulation result, since WCETT-LB achieved the best delivery
ratio, the average end-to-end delay of WCETT-LB is expected to be lower than
the other routing metrics. From Figure 12, average of end-to-end delay is investi-
gated under different simulation time. The average delay time of these three routing
metrics differ slightly. But, WCETT-LB still had the lowest end-to-end delay time.
In the last simulation run of this series, average congestion level at each mesh
router is considered as performance metric. With the load-balancing scheme in
WCETT-LB and traffic splitting scheme during routing, WCETT-LB achieved the
lowest average congestion level through each simulation time. As shown in Figure
13, WCETT-LB can reduce the average congestion level in a noticeable amount.
-•--WCETT-LB
-A-WCETT
-•— Hop count
90 120 150 210

Simulation time
Fig. 12. Average end-to-end delay under simulation time
50 *
40 ys
-WCETT-LB
-WCETT
«••
30 -Hop count
__*"~"
<g 20
^ "C*"
^W^
I K—"""""
10
30 60 90 120 150 180 210
Simulation time
Fig. 13. Average congestion level under simulation time.
5. Conclusions and Future Work

In this paper, a new routing metric called WCETT-LB is proposed to provide load-
balancing ability over multiple paths in wireless mesh networks. In order to cooperate
with WCETT-LB, a traffic splitting algorithm is also proposed to divide the traffic
among mesh routers. All the features of the proposed scheme were described and
compared with existing similar schemes qualitatively and qualitatively. The quan-
titative evaluation was carried out in a simulation settings using NS2 simulation
tool.
As future work, the proposed scheme will be extended for load-balancing in the
presence of inter-domain mobility and in multi-radio and multi-channel network
environment.
Acknowledgments
This research is supported in part by the Natural Science and Engineering Research
Council of Canada (NSERC). T h e authors would like to t h a n k NSERC for the
financial support.
References
1. A. Raniwala, T. Chiueh, "Architecture and Algorithms for an IEEE 802.11-based Multi-
channel Wireless Mesh Network," In Proceedings of IEEE Infocom, 2005, Vol 3, pp. 2223-
2234.
2. B. Xie, Y. Yu, A. Kumar, and D.P Agrawal, "Load-balancing and Inter-domain Mobility
for Wireless Mesh Networks," IEEE GLOBECOM, 2006, pp. 1-6.
3. D. Couto, D. Aguayo, J. Bicket, and R. Morris, "A High-Throughput Path Metric for
Multi-Hop Wireless Routing," ACM MobiCom, 2005, Vol 11:4, pp. 419-434.
4. E. Belding-Royer, P. Sambasivam, and A. Murthy, "Dynamically adaptive multipath
routing based on aodv," In Proc. Mediterranean Ad Hoc Networking Workshop, Bodrum,
Turkey, June, 2004.
5. Guido R. Hiertz, Sebastian Max, Dee Denteneer, Stefan Mangold, Erik Wei, and Lars
Berlemann, "Mesh technology enabling ubiquitous wireless networks: invited paper,"
Proceedings of the 2nd annual international workshop on Wireless internet, 2006, Vol
220, Article No.7.
6. I.F. Akyildiz, X. Wang, and W. Wang, "Wireless Mesh Networks: A Survey," Computer
Networks Journal (Elsevier), June, 2005.
7. K. Ramachandran, M. Buddhikot, G. Chandranmenon, S. Miller, E. Belding-Royer, and
K. Almeroth, "On the Design and Implementation of Infrastructure Mesh Networks," In
Proceedings of IEEE Workshop on Wireless Mesh Networks (WiMesh), 2005.
8. Kyu-Han Kim, Kang G. Shin, "On accurate measurement of link quality in multi-hop
wireless mesh networks," Proceedings of the 12th annual international conference on
Mobile computing and networking, 2006, pp. 38-49.
9. Mohit Virendra, Qi Duan, Shambhu Upadhyaya, and Vishal Anand, "A New paradigm
for Load-Balancing in Wireless Mesh Networks," CSE Dept. State University of New
York, Buffalo, 2006, pp. 133-142.
10. Nagesh S. Nandiraju, Deepti N., D.P. Agrawal, "Multipath Routing in Wireless Mesh
Networks," IEEE International Workshop on Heterogeneous Multi-Hop Wireless and Mo-
bile Networks, 2006, pp. 741-746.
11. R. Draves, J. Padhye, and B. Zill, "Routing in Multi-Radio, Multi-Hop Wireless Mesh
Networks," Proceedings of the 10th annual international conference on Mobile computing
and networking, 2004, pp. 114-128.
12. S. Kandula, D. Katabi, B. Davie, and A. Charny, "Walking the tightrope: Responsive
yet stable traffic engineering," In Proceedings of ACM SIGCOMM, Philadelphia, PA,
August, 2005, pp. 253-264.
13. S. McCanne, S. Floyd, "Network Simulator," http://www.mash.cs.berkeley.edu/ns/.

14. Y. Yang, J. Wang, and R. Kravets, "Designing Routing Metrics for Mesh Networks,"
IEEE Workshop on Wireless Mesh Networks, 2005.
15. Y. Yang, J. Wang, and R. Kravets, "Interference-aware Load Balancing for Multihop
Wireless Networks," Technical Report UIUCDCS-R-2005-2526, Dept. of Computer Sci-
ence, Univ. of Illinois at Urbana-Champaign, 2005.
16. Yun Hu, Shoubao Yang, Dapeng Wang, and Lei Zhang, "SMETT: A New Routing Met-
ric for Multi-Radio and Multi-Channel Wireless Mesh Network," International conference
on WiCOM, 2006, Vol 22:24, pp. 1-4.

Research Paper 1

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Research Paper 1

Uploaded by

Copyright:

Available Formats

Journal of Interconnection Networks W K 3 World Scientific

Vol. 8, No. 4 (2007) 407-426 V S ^ www.woridscientific.com

E N H A N C E D R O U T I N G METRIC FOR LOAD-BALANCING IN

University of Guelph, 50 Stone Road East,

Received 22 June 2007

Keywords: Wireless mesh network; load-balancing; routing metrics.

network. The hybrid mesh network architecture is a combination of infrastructure

fairness, network auto-configuration and mobility management. Our focus in this

performance of the networks. A good metric should contain sufficient information

"^ ^ " Wired Link

"M" ;@;"" / ., A......4 )

'% 5' „,--''

Fig. 1. A three-layer wireless mesh network architecture

the scheme does not consider load-balancing at mesh routers.

Table 1. Notations used in the paper.

3. The Proposed Load-balancing Scheme

3.1. WCETT-Load Balancing Metric

In the proposed approach, we integrate load-balancing component into WCETT

Thus, in addition to all the features of WCETT, WCETT-LB considers the

3.2. Global Congestion-Aware Routing

nodes located further in the path. We propose a global congestion-aware routing

3.2.1. Congestion-Aware Routing

path is better than the current path.

Fig. 2. An example wireless mesh network.

We now illustrate the operation of our protocol using an example. Consider

4. The analysis and discussion of simulation results

several features of congestion-aware routing protocol proposed in [10]. In our scheme

Table 2. Comparison of existing routing metrics.

load-balancing, inter-flow interference and intra-flow interferences. We have also

4.1. Performance Metrics

congestion level and standard deviation of traffic volumes at mesh routers.

number of packets originally sent in the network.

4.2. Simulation Parameters

4.3. Simulation results and discussion

4.3.1. Simulation setup

Table 3. Network parameters for simula-

Network simulator NS2

4.3.2. Simulation under different traffic load

Fig. 3. Packet delivery ratio under traffic load.

Fig. 4. Average end-to-end delay under traffic load.

Fig. 5. Average congestion level under traffic load.

Fig. 6. Standard deviation under traffic load.

4.3.3. Simulation under different traffic flows

Fig. 7. Packet delivery ratio under traffic flows.

Fig. 8. Average end-to-end delay under traffic flows.

In the third simulation runs, as shown in Figure 9, the performance of WCETT

Fig. 9. Average congestion level under traffic flows.

Fig. 10. Standard deviation under traffic flows.

Fig. 11. Packet delivery ratio under simulation time.

4.3.4. Simulation under different time

90 120 150 210

Fig. 12. Average end-to-end delay under simulation time

Fig. 13. Average congestion level under simulation time.

5. Conclusions and Future Work

13. S. McCanne, S. Floyd, "Network Simulator," http://www.mash.cs.berkeley.edu/ns/.

You might also like