Você está na página 1de 5

Interference-Aware Load Balanced Routing in

Wireless Mesh Networks


Jing Feng

Ran Xia

Huaibei Zhou

Computer School
Wuhan University
Wuhan, China
genefeng@hotmail.com

FiberHome Telecommunications
Technologies Co., Ltd.
Wuhan, China
rxia@fiberhome.com.cn

International School of Software


Wuhan University
Wuhan, China
zhouhuaibei@vip.sina.com

AbstractWireless mesh networks (WMNs) will play


an important role in the next generation wireless
communication systems because it can support
broadband services with ubiquitous coverage by low
transmission power. However, the question will be
caused by using the shortest path routing protocol
such as DSR and AODV in such multi-hop networks.
Load unbalance of nodes will sharply degrade the
traffic quality in WMNs. In this paper, we put
forward a traffic load model considering the
interference among wireless nodes. By using the
probability of channel busy as the load of a node, we
propose a load balanced routing scheme based on
AODV. Simulation results show that InterferenceAware Load Balanced Routing (IALBR) achieves
significant traffic quality enhancement over basic
IEEE 802.11 mesh networks.
Keywords-interference-aware;
load
routing; wireless mesh network; multi-hop
I.

balanced

INTRODUCTION

Recently, multi-hop wireless mesh network (WMN) has


attracted increasing attention and deployment as a new
approach to provide last-mile broadband Internet access [1].
WMNs provide a mechanism for information exchange without
the need for pre-existing infrastructure or administrative
support. Nodes act as repeaters to transmit data from nearby
nodes to peers that are beyond the radio range of the sender.
This can result in a network that has the following advantages:
(a) very inexpensive network infrastructure due to the
proliferation of IEEE 802.11 and IEEE 802.16 based devices,
(b) easiness of deploying and reconfiguring the network, (c)
spatial reuse for increased capacity, (d) coverage enhancement
[2].
It is likely that WMNs will become an increasingly popular
mechanism to interconnect wireless devices, such as laptop
computers and PDAs. The mesh topology allows for fully selfconfiguring networks where each node can relay message on

behalf of others, thus increasing both the range and available


bandwidth with the number of nodes active within the system,
unlike the point-to-point structure of legacy wireless networks
[3]. However, a key problem arising in WMNs is that some
nodes which have heavy load will become the bottlenecks of
the traffic flows by using the shortest path routing protocols
such as DSR [4] and AODV [5].
Routing with load balancing in multi-hop wireless networks
has been exploited in various approaches [6], [7], [8], and [9].
In [6] Sung-Ju Lee et al. proposed Dynamic Load-Aware
Routing (DLAR) protocol that considers intermediate node
routing loads as the primary route selection metric. The
protocol also monitors the congestion status of active routes
and reconstructs the path when nodes of the route have their
interface queue overloaded. In [7] Xiaojing Tao et al. proposed
traffic balancing to improve dramatically performance for
wireless Ad Hoc networks with right placement of the access
points. In [8] Jorma Virtamo et al. study the load balancing
problem in a dense wireless multi-hop network, where a typical
path consists of large number of hops and present a general
framework for analyzing the traffic load resulting from a given
set of paths and traffic demands. They formulate the load
balancing problem as a minmax problem and give two lower
bounds for the achievable minimal maximum traffic load. And
in [9] Jie Gao et al. study routing algorithms on wireless
networks that use only short paths, for minimizing latency, and
achieve good load balance, for balancing the energy use. Most
of the early load balanced algorithms aim to distribute the
routing load of one node to multiple nodes in order to balance
the load without considering the interference from the other
nodes which do not directly communicate with this node.
The remainder of this paper is organized as follows. In
Section , we introduce IEEE 802.11 background. Then the
interference among the nodes of WMNs is analyzed. In Section
, an interference-aware load model is given and the load
balanced routing scheme is proposed. Simulation methodology
and performance evaluation of our proposal are detailed in
Section . Section concludes the paper by summarizing
results.

1-4244-1312-5/07/$25.00 2007 IEEE

1730

II.

IEEE 802.11 BACKGROUND

In this paper, we take the IEEE 802.11a WLAN standard as


an example to evaluate the load of a node in the WMNs. The
developed analytical framework for mesh network can be
applied straightforwardly to various wireless system, such as
IEEE 802.11/15/16 networks as well as free space optics
system. This paper assumes use of the IEEE 802.11 Distributed
Coordination Function, the access method used in WMNs
mode. In order to reduce collisions caused by hidden terminals
in the network, IEEE 802.11 uses a four-way
RTS/CTS/Data/ACK exchange. In brief, a node that wishes to
send a data packet first sends an RTS (request to send) packet
to the destination. If the destination believes the network is
idle, it responds with a CTS (clear to send) packet. The sender
then transmits the data packet, and waits for an ACK
(acknowledgment) from the receiver. If a node overhears an
RTS or CTS, it knows the medium will be busy for some time,
and avoids initiating new transmissions or sending any CTS
packets. Figure 1 shows the frame transmission scenarios
according to IEEE 802.11 CSMA MAC protocol with
RTS/CTS.

A wireless nodes reception is impacted by other nodes


which do not directly communicate with it. Consider the
network shown in Figure 2, where node d is in the process of
sending a data packet to e . The solid-line circle denotes the
node B s interference range, and the dotted-line circle denotes
the node d s interference range. Before sending a data packet,
node A first sends a RTS packet to B to assure the link
between A and B is available. When B receives a RTS from
A , it checks the NAV value. Because of the interference with
node d s sending packets to e , the NAV value of the node B
is not zero. According to the IEEE 802.11 CSMA MAC
protocol with RTS/CTS, node B will not send a CTS packet
back to A . From the above analysis we can see that though
node B is not busy, it cannot receive any data packet from
other nodes because of the interference of node d .

Figure 2. Node d's sending packets to e will corrupt B's reception from A

Figure 1. Timing frame transmissions according IEEE 802.11 CSMA

802.11 RTS and CTS packets include the amount of time


the medium will be busy for the remainder of the exchange.
Each node uses these times to update its network allocation
vector (NAV). The NAV value indicates the amount of time
remaining before the network will become available. Upon
successful receipt of an RTS frame not addressed to itself, a
node updates its NAV to the maximum of the time carried in
the RTS frame and its current NAV value. Upon receiving an
RTS addressed to itself, a node returns a CTS frame only if its
NAV value is zero, otherwise no CTS is sent. Hence, a sender
will see no CTS if its RTS packet has collided with another
transmission at the receiver, or if the receivers NAV indicates
that the network is not available. A node times out and re-sends
the RTS if it receives no CTS. 802.11 double its backoff
window each time a timeout occurs; it resets the backoff to a
minimum value after a packet is transmitted successfully or is
dropped after reaching maximum retry limit.
III.

LOAD BALANCED ROUTING SCHEME

A. Interference-Aware Load Model


In wired networks, the traffic load carried by each node
includes the forwarded traffic from other nodes and the
demanded traffic from itself. Dissimilarly, according to the
CSMA MAC protocol with RTS/CTS, wireless nodes with
omnidirectional antennas will increase the extra load from
other nodes interference whether they communicate directly
with each other or not.

At the mean time, a wireless nodes sending packets is


impacted by other nodes within the interference range of it.
Lets take the network shown in Figure 3 as an example, where
node B wants to send a data packet to node C . The solid-line
circle denotes node B s interference range, and the dotted-line
circle denotes node C s interference range. Firstly, node B
sends a RTS packet to C . Though node C receives the RTS
packet from B , impacted with the d s sending packets to e ,
C will not send back a CTS packet to B . Draw a conclusion,
B s sending packets is impacted not only by the load of itself
but also by the load of its next hop node.

Figure 3. Node d's sending packets to e will corrupt node B's sending to C

The traffic load carried by each mesh node includes both


the forwarded traffic from other nodes and the demanded
traffic from itself as well. So the node i s load can be given by
the formula (1).

Li = Lirelay + Lisend

(1)

We consider the relay load as two parts: reception load and


sending load. Then we have formula (2) which can be
simplified as formula (3) furthermore.

1-4244-1312-5/07/$25.00 2007 IEEE

1731

Li = Lirelay _ recv + Lirelay _ send + Lisend

(2)

Li = Lirecv + Lisend

(3)

In this paper, we consider the interference from the other


nodes as one part of the nodes traffic load. According to IEEE
802.11 CSMA MAC protocol, the interference from the other
nodes will impact the nodes channel idle. In order to scale the
traffic load, we use the probability of the channel busy as node
i s load. Then, we get
i
Lirecv = (1 Precv
_ idle )

Lirecv _ idle (t ) = e

r + f ) t

rR ( i )

Where R (i ) is the set of nodes which will send data packets to


node i , and IF (i ) is the set of nodes which is within i s
interference range but not sending data packets to i .
We have considered M/D/1 queuing model because
packets arriving according to Poisson processes with rate ,
and these packets are serviced in a deterministic manner as of
constant size at a rate of . In M/D/1 queuing model, packets
are queued up in the buffer and are serviced in the order of
their arrivals [10]. So we can have

(4)

Lisend _ idle (t ) = e
and

i
send

i
send _ idle

= (1 P

j
recv _ idle

s / i )t

(10)

sS ( i )

(5)

i
where j is the next hop of the node i . Precv
is the idle
_ idle
i
is the
probability of node i s reception channel, and Psend
_ idle
idle probability of node i s sending channel.
i
i
and Psend
.
Now we will solve the parameter of Precv
_ idle
_ idle
Assume the packet arrival rate of node i follows the Poisson
distribution and the average arrival rate of a node is . Let
Pni (t ) denotes the probability of n packets arriving at node i
during interval time t . We have

Pni (t ) =

(9)

f IF ( i )

( t ) t
e
n!
n

(6)

Therefore, the probability of no packet arriving at node i


during the interval time t is
i
i
t
Precv
_ idle (t ) = P0 (t ) = e

(7)

In the IEEE 802.11 CSMA/CA wireless network, nodes in the


same interference range will contend and share the bandwidth.
For simplicity, we join the Poisson processes of multiple
sources in an aggregate Poisson process. Let IF (i ) be the
number of nodes which will interfere with node i . From the
point of view of node i , the total packet arrival rate of other
nodes in node i 's interference range is

= i

(8)

iIF

i
Therefore, the idle probability of node i is pidle
(t ) = e t , and
we have

where S (i) is the set of traffics which will be relayed by node


i.
Finally, we can have the load of source nodes and
destination nodes separately:

Lsrc (t ) = (1 e

s
sS ( src )

/ src ) t

Ldst (t ) = (1 e

r
rR ( dst )

r
rR ( src )

f
f IF ( dst )

f
f IF ( src )

)t

(11)

)t

(12)

And we can also get the load of an intermediate node i of the


traffic as follows:
( r + f ) t
Li (t ) = (1 e rR ( i ) f IF ( i ) )
( s / i )t
( r + f ) t
+ (1 e sS ( i )
e rR ( j ) f IF ( j ) )

(13)

Where j is the next hop of the node i . If node i is the


destination of one traffic, and there is no next hop of i , we can
have

Ldst (t ) = (1 e

r
rR ( dst )

f
f IF ( dst )

)t

(14)

From equation (11-13) we can see that node i s load is caused


by not only traffics transmitted on itself, but the interference
from others within the interference range of node i .
B. Routing Algorithm
From the equation (13), we can see that one nodes (except
the destination node) load value is dependent on its next-hop
nodes reception load value. And each node can get the values

1-4244-1312-5/07/$25.00 2007 IEEE

1732

r + f and s / i from MAC layer. Therefore, a cross-

rR ( i )

f IF ( i )

sS ( i )

layer should be designed for Interference-Aware Load


Balanced Routing (IALBR). Above two values and the sum of
nodes' load values that the RREQ have passed will be stored in
RREQ (Route Request) packets. And to a node, the load value
sum of traffic will be stored in route table. Following will detail
the process of finding a lowest load router from S to D in the
network as shown in the Figure 4.

are compared. If the information in RREQ is the latest, the


sequence number and the sum of load value about this
traffic in route table are updated. Then a RREP is sent back
to the source via the selected route.
If two sequence values are equal, the sum of load values
in the packet and it stored in the destination are compared.
If the load value in the packet is smaller, the value stored in
the destination is updated and RREP is sent back to the
source via the selected route, otherwise, RREQ is dropped.
Not similar to AODV [5], the intermediate nodes in IALBR
will not give a RREP packet as a response back to the source
because the total load value of the traffic is needed to be
collected. Therefore, IALBR can find a lowest load route
between sources and destinations instead of the shortest path
route.
IV.

Figure 4. Interference-aware load balanced routing

Source node

A route is required but no information to the destination


is known, the source node S floods the RREQ packet to
discover a route. Load information is attached in RREQ
packets.
Once source receives the RREP (Route Reply) packet,
information in the route table will be updated and then data
packets will be sent to the next-hop according the new route
table.

Our simulations all run in NS-2. We evaluate the IALBR


schemes by comparing the performance with AODV [5] and
DLAR [6]. In our simulations, each node has a radio
propagation range of 250 meters and the channel capacity is 2
Mb/s. Each run has executed for 300 seconds of a simulation
interval.
A. Simulation Scenario Shown in Figure 4
In this simulation scenario, the size of data payload is 1024
bytes and the packet rate is 10 packets/sec. There are three
constant flows carried on this model, as shown in Table 1.

Intermediate node

When intermediate node c receives a RREQ packet, it


looks up whether <source, destination> pair is already
existed in its route table. If the pair cannot be found, a route
entry of the <source, destination> pair and its load value are
built and the previous hop to that entry is recorded. This
previous node's information is needed later to replay the
RREP packet back to the source of the route.
Otherwise, if the <source, destination> pair already
existed in the route table, load values in the route entry and
in the RREQ packet are compared. If the value in the
packet is smaller, load value in the route entry is updated,
otherwise, RREQ is dropped.
Before rebroadcast RREQ, two values of node

i , r + f and s / i replace the older value in


rR ( i )

f IF ( i )

sS ( i )

RREQ separately, and the load value sum of the nodes that
RREQ has passed is also recalculated.

SIMULATIONS AND RESULTS

Destination node

After receiving a RREQ packet, node D looks up


whether <source, destination> pair already existed in route
entries. If the pair cannot be found, a route entry of the
<source, destination> pair and its load value are built and
the previous hop to that entry is also recorded. Otherwise
the sequence number in route table and it in RREQ packet

TABLE I.

No.

Src.

DESCRIPTION OF THREE CONSTANT FLOWS

Dest.

Start Time

Stop Time

1 second

150th second

D
e

100th second

200th second

151st second

300th second

st

From the simulation results shown in Figure 5, we can see


that IALBR and AODV almost have the same performance
value on flow , since there is no interference from other
nodes when S begins to find route to D . Therefore, IALBR
and AODV have the same traffic S c D . Though
interference is caused by flow at 100th second, the route will
not change unless a link break occurred, then a RREQ packet is
sent out to find a new route. On flow , IALBR chooses the
traffic S a b D according to the traffic load, while
AODV chooses the same traffic as flow , because of its low
sensitivity with traffic load. In this case, IALBR achieves lower
delay than AODV and DLAR which two have the same
performance because they both dont consider interference
between nodes within interference ranges.

1-4244-1312-5/07/$25.00 2007 IEEE

1733

Packet rate, described in our load model, is the main factor


of interfering one nodes sending and reception data packets.
The simulation results shown in Figure 6 indicate the
relationship between packet rate and delay. From this figure,
we can see that high packet rate can significantly increase
traffic delay.
Further more, interferences among nodes have little impact
on throughput of traffics in simulations.

Figure 8. Average end-to-end delay with 100 nodes

V.

Figure 5. Average end-to-end delay of flow and flow

Figure 6. Average end-to-end delay with different packet rate

B. Random Simulation Scenarios


In this instance, our simulations adopt two types of WMN,
a sparse network and a dense one. In the sparse network, we
create a random wireless scenario of 50 nodes move with the
maximal speed of 5 meter/sec within an 800 meters square area
using CMUs setdest module, and create 10 ftp data flows
using cbrgen program. In the dense network, 100 mobile nodes
move within a 500 meters square area. Simulation results
shown in Figure 7 and Figure 8 indicate that IALBR is much
more suit for WMN than AODV because of its low end-to-end
delay.

Figure 7. Average end-to-end delay with 50 nodes

CONCLUSION

In this paper, we inspect the interference among nodes and


present an Interference-Aware Load Balanced Routing (IALBR)
protocol that uses the routing load of the intermediate nodes as
the main route selection criteria. Consider the interference from
the other nodes, a nodes probability of channel busy is taken
into account for the value of nodes load. As the Route Layer
cannot know the load value of one node, a cross-layer between
Route Layer and Mac Layer is designed to implement the
IALBR. The route table that each node maintains in IALBR is
similar to that in AODV [5] routing protocol, with some
changes in the route entry. Load values are stored both in route
table of each node and in the RREQ packets in IALBR, so that,
the traffic load can be calculated by the destination node. And
the lowest load traffic can be selected by the destination for
having the good quality traffic to transmit data packets.
Simulation results indicate that IALBR is quite suit for such
a multi-hop wireless networks. And significant traffic quality
enhancement can be achieved over basic IEEE 802.11 wireless
mesh networks.
REFERENCES
[1]

R. Karrer, A. Sabharwal, and E. Knightly, Enabling large-scale wireless


broadband: The Case for Taps, In HotNets, 2003.
[2] B. Braunstein, T. Trimble, R. Mishra, S. B. Manoj, and R. Rao, On the
traffic behavior of distributed wireless mesh networks, in Proc. of
WoWMoM 2006.
[3] David Waiting and Neco Ventura, A metric to increase throughput in
untrustworthy multi-hop wireless mesh networks, in Proc. of IEEE
WCNC 2006.
[4] J. Broch, D. B. Johnson, and D. A. Maltz, The dynamic source routing
protocol for mobile ad-hoc networks, Internet draft, draft-ietf-manetdsr-07.txt, Feb. 2002.
[5] C. E. Perkins and E. M. Royer, Ad hoc on-demand distance vector
(AODV) routing, Internet draft, draft-ietf-manet-aodv-09.txt, Jan. 2002.
[6] Sung-Ju Lee and Mario Gerla, Dynamic load-aware routing in ad hoc
networks, in Proc. of IEEE ICC 2001.
[7] Xiaojing Tao, Thomas Kunz, and David Falconer, Traffic balancing in
wireless mesh networks, in Proc. of IEEE WIRELESSCOM 2006.
[8] Esa Hyyti and Jorma Virtamo, On load balancing in a dense wireless
multihop network, in Proc. of IEEE NGI 2006.
[9] Jie Gao and Li Zhang, Load-balanced short-path routing in wireless
networks, IEEE Transactions on Parallel and Distributed Systems, vol.
17, no. 4, April 2006.
[10] Donald Gross and Garl M. Harris, Fundamentals of queuing theory, 3rd
ed., John Wiley & Sons, 1998.

1-4244-1312-5/07/$25.00 2007 IEEE

1734

Você também pode gostar