Content uploaded by Waldir Moreira
Author content
All content in this area was uploaded by Waldir Moreira
Content may be subject to copyright.
Using Multiple Metrics with the Optimized Link State Routing
Protocol for Wireless Mesh Networks
Waldir A. Moreira Jr.1,4, Elisangela Aguiar2,4, Antˆ
onio Abel´
em3,4, Michael Stanton5
1Programa de P´
os-Graduac¸˜
ao em Ciˆ
encia da Computac¸˜
ao (PPGCC)
2Programa de P´
os-Graduac¸˜
ao em Engenharia El´
etrica (PPGEE)
3Faculdade de Computac¸˜
ao
4Grupo de Redes de Computadores e Comunicac¸˜
ao Multim´
ıdia (GERCOM)
Universidade Federal do Par´
a, Rua Augusto Corrˆ
ea 01, 66075-110, Bel´
em, PA, Brazil
5Instituto de Computac¸˜
ao
Universidade Federal Fluminense, Rua Passo da P´
atria 156, bloco E, sala 350,
24210-240, Niter´
oi, RJ, Brazil
{mraranha, eaguiar, abelem}@ufpa.br, michael@ic.uff.br
Abstract. Wireless mesh networks (WMNs) can be used in many different ap-
plications. However, they lack standards and, as a consequence, a number of
issues must still be addressed to ensure the proper functioning of these networks.
Amongst these issues, routing is this paper’s main concern. Thus, we propose
the use of multiple metrics with the proactive Optimized Link State Routing
(OLSR) protocol, in order to provide quality of service routing. Even though
it has already been proved that routing with multiple metrics is an NP-complete
problem, we show how the techniques of Analytic Hierarchy Process (AHP) and
Pruning may be combined to perform multiple-metric routing, offering the best
available routes based on the considered metrics. A study on the performance of
the metrics considered for the proposal is also carried out in the NS simulator.
1. Introduction
Over the years wireless mesh networks (WMNs) have shown their usefulness in differ-
ent scenarios, especially those where extending already existing networking services is
desired but cabling is not a feasible alternative.
As a result, these networks provide a more versatile and inexpensive solution if
compared to wired, and even some other wireless, technologies. They became quite pop-
ular mainly due to their extended coverage, robustness, self-configuration, easy mainte-
nance and low cost features [Lee et al. 2006].
Some examples of the great utility of WMNs include: to extend the coverage area
of enterprises and universities; to reach areas where cabling is somehow difficult due to
cost and/or physical obstacles; to provide communications in emergency situations such
as earthquakes, fire fighting, and other catastrophes; to provide public Internet access; to
operate intelligent transportation systems; and to help in military and rescue operations
[Bruno et al. 2005].
However, even though WMNs are considered to be very useful, they still lack stan-
dards, and this has resulted in the emergence of many different solutions, proprietary or
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
903
not, that are not interoperable among themselves. In fact, the IEEE 802.11 working group
is investigating proposals which will specify the WMN functionalities, to be incorporated
in the IEEE 802.11s standard.
Besides the standardization problem, the problem of WMN routing is also of great
interest, and must be carefully addressed to guarantee the proper functioning of WMNs.
Due to these WMN characteristics (dynamic topology, lack of resources as band-
width, security, and scalability), WMN routing protocols must display the features of
self-management, self-configuration, and self-healing [Bruno et al. 2005].
Many different routing protocols have already been proposed. However, these
protocols are unable to answer all the needs of this kind of network because each of them
was developed to deal with a specific application [Kowalik and Davis 2006].
A number of different approaches have been considered in order to develop WMN
routing protocols, including the use of heuristics, a single metric, a single compound
metric, a single mixed metric, a composite metric, multiple metrics, and multidimensional
metrics [Costa et al. 2002] [Badis et al. 2003] [Aslam et al. 2004] [Alkahtani et al. 2006]
[Faccin et al. 2006]. Multiple metrics are of interest to this paper.
This paper’s main goal is to use multiple metrics with the proactive Optimized
Link State Routing (OLSR) protocol, to guarantee the selection of routes which are com-
posed of good quality links. However, it is worth pointing out that the problem of working
with multiple additive/multiplicative metrics in any combination is non-trivial, and turns
out to be NP-complete [Wang and Crowcroft 1996] [Badis et al. 2003].
A solution for getting around the NP-complete problem when integrating multiple
metrics may be achieved by combining the two techniques known as Analytic Hierarchy
Process (AHP) and Pruning. This becomes our secondary goal which is applying these
two simple techniques normally used in wired scenarios to a WMN context.
The idea is to use the AHP multi-criteria technique, a methodology of decision
analysis developed by [Saaty 1980] to aid in the decision of choosing the best route be-
tween a given source and destination. [Alkahtani et al. 2006] proposed a solution which
made use of this technique for wired networks with good results. Yet, this solution suffers
from great complexity due to the number of matrix computations required to determine
the best route. [Alkahtani et al. 2006] suggest changes to the size of the matrices being
calculated, which would reduce the number of steps required to determine the best route
as well as to reduce this complexity.
There is no need to say that the complexity is even greater in a WMN context,
since all nodes might be able to ”sense” all, or most of, the other ones present in the
network. As a result, the size of the matrices increases, and that is what we would like to
avoid.
So, we introduce a second technique, called Pruning, to be applied before the first
AHP step in order to improve AHP’s computations. This technique simply gets rid of
the paths that are not feasible, that is, those which have qualities exceeding the desired
threshold. Both techniques will be detailed in Section 4.
The remainder of this paper is organized as follows. Section 2 presents related
904
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
work. Section 3 gives a general overview on WMNs, routing protocols, and link quality
metrics. Section 4 presents the AHP and Pruning techniques. A case study using simu-
lation is presented in Section 5. Finally, Section 6 presents the conclusions and current
work.
2. Related Work
This section presents previously published work on matters related to this paper. The
following publications provide relevant information on WMNs, defining these networks,
showing their applicability, and pointing out important issues currently being investigated.
[Akyildiz et al. 2005] present a detailed study on advances and open research is-
sues related to all protocol layers in WMNs highlighting system architectures, applica-
tions, testbeds, commercial practices and related standards activities. Case studies, tech-
nical issues and solutions for developing WMNs, as well as an overview on standardiza-
tion of mesh technology, are presented by [Faccin et al. 2006]. [Held 2005] explores some
WMN applications and protocol operations, analyzing problems affecting these networks
as well as suggesting solutions for each of them.
An overview on mesh technology is provided by [Bruno et al. 2005] using exam-
ples of proprietary and commercial solutions, concepts on which WMNs must be based,
and challenges faced by WMN design. Another overview on mesh technology, pointing
out some standards which are applicable to the concept of multihop techniques in different
wireless networking technologies, is presented in [Lee et al. 2006].
[Nandiraju et al. 2007] discuss challenges slowing down the development of
WMNs, showing how each layer of the network could be improved to address such
challenges. Finally, [Zhang et al. 2007] point out problems and challenges in designing
WMNs, considering a number of important issues, and detailing techniques to improve
WMN performance.
Regarding routing, [Clausen and Jacquet 2003] provide important information on
the protocol considered in this paper, OLSR. Reasons for the existence of so many differ-
ent WMN routing protocols are presented in [Kowalik and Davis 2006].
A number of different approaches have been proposed for improving WMN rout-
ing, which may simply combine different metrics with OLSR, or create new routing pro-
tocols. [Costa et al. 2002] evaluate the use of a single mixed metric and heuristics, com-
pared with the multiple individual metric approach, to speed up routing computations,
with the former presenting outstanding results. [Badis et al. 2003] propose a single metric
solution to achieve QoS routing. The use of a composite metric for routing improvement
is explored in [Aslam et al. 2004]. A new protocol is proposed by [Alkahtani et al. 2006]
using a multiple metric approach, while [Faccin et al. 2006] discuss how multidimen-
sional metrics can be used for QoS routing.
The metrics we study in this paper for integration are Expected Trans-
mission Count, Minimum Loss, and Minimum Delay, and these are described
in [DeCouto et al. 2003], [Passos et al. 2006], and [Cordeiro et al. 2007], respectively.
However, the use of multiple metrics is non-trivial, and [Wang and Crowcroft 1996] prove
it turns out to be NP-complete problem. [Saaty 1980] presents one of the techniques con-
sidered in this paper to solve the NP-completeness problem, and [Alkahtani et al. 2006]
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
905
present a routing protocol based on this technique known as Analytic Hierarchy Process
(AHP).
A second technique, Pruning, can be used to improve routing computations, when
more than one additive/multiplicative metric is considered, and [Costa et al. 2002] de-
scribe its use with excellent outcomes.
3. Wireless Mesh Networks
WMNs have self-organizing, self-configuring, and self-healing features with easy de-
ployment/maintenance at a very reasonable cost providing high scalability and reliable
services as well as improving capacity, connectivity, and resiliency of the already exist-
ing network. Due to these characteristics, WMNs have been recognized as a promising
technology, which will play an important role in future generations of wireless networks.
These networks are an extension of wireless ad hoc networks [Bruno et al. 2005].
However, protocols and architectures developed for ad hoc networks have a weak per-
formance when applied to WMNs. This is explained by the differences between these
two kinds of networks, regarding their applicability, deployment goals and the resource
limitations to which they are subjected [Held 2005].
3.1. Routing metrics in WMNs
WMNs are a combination of mobile and fixed nodes which communicate through wireless
links forming a multi-hop network.
In most cases, WMN nodes are fixed and not battery operated. Thus WMN routing
protocols must focus on reliability and performance improvement rather than dealing with
mobility or minimizing energy consumption.
Giving the characteristics of the WMN scenario under consideration (fixed
nodes and small number of nodes), proactive protocols are more suitable for it
[Zhang et al. 2007]. Amongst the proactive protocols, Optimized Link State Routing
(OLSR) protocol [Clausen and Jacquet 2003] has been widely used in mesh solutions
[Passos et al. 2006].
However, the original OLSR is not quite suitable for WMNs since it does not take
into account the link quality while computing routing tables. Instead, it considers the min-
imum hop count to determine the best path to reach a given destination. This metric, hop
count, has been shown not to be at all useful in multi-hop networks [DeCouto et al. 2003].
As a result, researchers started to propose metrics based on what constitutes a good qual-
ity link. That is, these metrics reflect the quality of each link, and the routing protocol
considers this information when computing its routing tables.
Many proposals have been made to improve WMN routing. The use of heuristics,
a single metric, a single compound metric, a single mixed metric, a composite metric and
multiple metrics are some of the solutions which have been proposed to create new routing
protocols, or simply to be combined with OLSR in order to improve its performance
[Costa et al. 2002] [Badis et al. 2003] [Aslam et al. 2004] [Alkahtani et al. 2006].
This paper focuses mainly on the combination of two metrics with OLSR, and
evaluates the performance of the ones presented as follows.
906
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
3.1.1. Expected transmission count (ETX) [DeCouto et al. 2003]
The goal here is to choose routes with the minimum expected number of transmissions
(including retransmissions) a packet will need to be delivered and have its receipt ac-
knowledged. Consequently, the selected routes have high throughput. The main advan-
tage is that probe packets are broadcast, which results in reduced probing overhead. The
main disadvantage resides in the fact that the probe packets are small and are sent at the
lowest data rate possible, and tend not to suffer the same loss rate as larger data packets
sent at higher data rates.
3.1.2. Minimum loss (ML) [Passos et al. 2006]
This is based on ETX, with the aim of selecting the path with the minimum loss proba-
bility. It uses the probability of successful transmissions, and not the inverse probability,
as in ETX. Another difference is related to a route composed of two or more links. The
route probability is given by the product of the link probabilities instead of the sum of
their inverse probabilities. It has the advantage of eliminating high loss rate routes, and
the disadvantage that some low quality links may still be taken into account in choosing
a given route, since the metric considers only the total probability product.
3.1.3. Minimum Delay (MD) [Cordeiro et al. 2007]
The routing table computation is based on the total minimum transmission delay. The
transmission delay measurements come from a variant of a link capacity estimation tech-
nique, known as Adhoc Probe. The use of the Adhoc Probe technique is a great advan-
tage, because it takes into account differences in clock synchronization, thus providing
a more reliable measurement. A disadvantage is that this metric considers routes which
have nodes sharing a collision domain with many others, and this tends to degrade the
communication on such routes.
It is worth pointing out that, for this proposal, we only considered metrics that
are frequently discussed on the available related researches such as ETX. Since we have
proposed another metric, MD, we decided to simulate it along with ETX and ML in order
to determine the two metrics to be considered for the proposal.
4. Techniques for Combining Multiple Metrics with OLSR
It has been proved that the selection of routes based on the combination of additive and/or
multiplicative metrics is NP-complete [Wang and Crowcroft 1996] [Badis et al. 2003].
However, there is a technique which can be used to get around the NP-
completeness of the use of multiple metrics, known as Analytic Hierarchy Process. Thus,
a secondary goal of this proposal is to combine another technique, called Pruning, to
AHP in order to reduce its complexity still offering routes based on the best link qualities
available at the moment.
This section is intended to present what these techniques are and how they work.
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
907
4.1. Analytic Hierarchy Process (AHP) [Saaty 1980]
AHP is widely known in the field of decision making, when different qualitative and/or
quantitative criteria must be applied. A number of applications already make use of this
methodology, in fields such as telecommunications and the provision of health services.
By making small changes to the methodology, [Alkahtani et al. 2006] proposed a
routing protocol which takes different metrics into account, when deciding the best route
to a given destination. The goal was to provide support for multimedia applications which
are characterized by multiple Quality of Service (QoS) requirements.
To illustrate the approach proposed by [Alkahtani et al. 2006], we apply it to the
network in Figure 1 shown below. It is worth pointing out that this network is a small
WMN where every node may have a link to every other one unless they are too far apart.
Every link has two metrics: ETX and MD, for this example. Finally, we assume we wish
to set up a connection between nodes 1 and 4.
Figure 1. Example of a small network
Step 1: First, we need to find all possible routes between nodes 1 and 4. Ta-
ble 1 shows each possible path along with its respective overall values of ETX and
MD. Further information for these metrics can be found in [DeCouto et al. 2003] and
[Cordeiro et al. 2007], respectively.
Table 1. Possible paths
Paths A B C D
Links 124 134 1234 1324
ETX 2.19 2.25 3.23 3.45
MD 1.01 0.20 0.51 1.40
Step 2: The computation of the path-path pair-wise comparison matrix (ppcm)
for each metric is carried out to determine how well each path is scored for each metric
compared to the other paths. [Alkahtani et al. 2006] define three criteria: min - metrics
to be minimized (i.e. delay), max - metrics to be maximized (i.e. bandwidth), and bin -
binary nature metrics (i.e. link security - 1: secure, 0: insecure). Since ETX and MD are
quantitative metrics which need to be minimized, the matrix calculation is based on the
following equations. Here iand jare paths, and mjis the overall value of the metric for
pathj:
•ppcm(i, i)=1, when comparing the same path;
•ppcm(j, i) = 1/ppcm(i, j), for reciprocal paths; and
•ppcm(i, j) = mj/mi, for min criterion.
908
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
As an example, the ppcm for ETX is given by:
ppcm(ET X ) =
1 1.027 1.475 1.575
0.973 1 1.436 1.533
0.678 0.697 1 1.068
0.635 0.652 0.936 1
Step 3: The normalized path-path pair-wise comparison matrix (nppcm) is calcu-
lated, based on the following equation with iand j= 1, .., P, where Pis the number of
paths (four in this example):
nppcm(i, j) = ppcm(i, j )/Σppcm column(j)
As a result:
nppcm(ET X ) =
0.3043 0.3043 0.3043 0.3043
0.2962 0.2962 0.2962 0.2962
0.2063 0.2063 0.2063 0.2063
0.1932 0.1932 0.1932 0.1932
Step 4: The average normalized path-path pair-wise comparison matrix (anppcm)
is calculated, based on the following equation:
anppcm(i) = Σnppcm row(i)/P
The anppcm matrix is [nxP] where nis the number of metrics and Pis the number
of paths. The anppcm for ETX is:
anppcm(ET X ) = h0.3043 0.2962 0.2063 0.1932 i
Steps 2 - 4 must be carried out for every metric. And as result, the complete
anppcm is showed below. It is worth mentioning that each line of this matrix refer to the
metrics ETX and MD, respectively.
anppcm ="0.3043 0.2962 0.2063 0.1932
0.1143 0.5770 0.2263 0.0824 #
Step 5: The average normalized priority pair-wise comparison matrix (anprpcm)
is calculated to determine the relative importance of each metric compared with the other
metrics. For the original AHP, an absolute number is given to each metric, based on the
decision maker’s feelings. Then Steps 2 - 4 are performed to find the anprpcm. An-
other modification proposed by [Alkahtani et al. 2006] is that these metrics are assigned
weights directly in the range [0, 1] where the sum of all weights is equal to one, as pre-
sented in the matrix below.
anprpcm =h0.5 0.5i
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
909
Step 6: Select the required priority of metrics, if more than one priority is set. In
this example, the priority set is 0.5 for both ETX and MD, since priority is not considered
for this paper.
Step 7: Calculation of the total score for each path, Table 2, through the equation:
P athjscore =
n
X
i=1
(anprpcm[i]×anppcm[i, j]), j = 1, ..., P
Table 2. Total scores for each path
Paths A B C D
Total Score 0.2093 0.4366 0.2163 0.1378
Step 8: Select the path with the maximum total score to be used in the communi-
cation between nodes 1 and 4. As it can be seen, path B is the one that has the best link
quality overall.
To reduce the number of matrix computations, it would be enough to make ppcm
and nppcm be P x 1 matrices instead of P x P, since the columns of the P x P nppcm have
the same values. Thus, Step 5 would be eliminated.
This technique is a very interesting approach to get around the NP-completeness
problem of using multiple metrics for routing.
However, if the WMN is quite large, there may be a large number of possible
paths between a source and destination, due to the fact that each node may ”hear” a major
part of, if not all, the surrounding nodes. This situation will result in a large number of
calculations in order to obtain just ppcm and nppcm. To combat this, we adopt another
technique, Pruning.
4.2. Pruning
This technique has been applied to many different applications in order to improve their
performance [Costa et al. 2002].
It consists of eliminating links which have quality values greater than or equal to
a given threshold. To determine this threshold, we need to know both the network and
the functioning of the metrics used in this scenario. For this proposal, we calculate the
threshold through the median of ETX values of the links, discarding links with values
above this threshold since ETX is to be minimized.
In the previous example, path D in Table 1 would not even be determined since
one of its links has the highest values for the ETX and MD metrics. By doing this, the
number of possible paths calculated between the source and destination can be reduced,
improving even more the performance of the AHP technique.
5. Case Study
In this section we present the scenario used in carrying out simulations, and how the
routing metrics were chosen.
910
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
5.1. Scenario
The simulations attempted to reproduce the behavior of the metrics to be used in the
routing protocol for a WMN backbone at the main campus of the Federal University of
Par´
a, in Bel´
em, Par´
a state, Brazil. This campus is located by the Guam´
a river within a
large wooded area, and and contains many buildings separated by parking areas. Figure 2
shows the WMN scenario under consideration.
Figure 2. Federal University of Par ´
a Campus
Since this scenario happens to be in a tropical region, deploying an outdoor wire-
less network is rather challenging due to sometimes heavy rain and the number of trees
present in the area.
5.2. Metrics used
To decide which metrics were to be used for routing, simulations were run to determine
the metrics’ performance in the aforementioned scenario. These simulations were carried
out on Network Simulator 2.31 [NS 2007] using different seeds for the random number
generator. A confidence interval of 95% was considered for the calculations according to
[Jain 1991]. Each simulation was run for 50 seconds and repeated 10 times.
The two metrics with the best overall performance would be considered for the
routing protocol. To show that the real scenario and equipments were closely represented
in the simulation, some variables were chosen, as shown in Table 3.
It is worth pointing out that the values of Path Loss Exponent and Shadowing
Deviation could have been obtained directly from the NS manual [NS 2007]. However,
the values used here were obtained from field measurements carried out at each of the
ten points as shown in Figure 2. Using a notebook computer running the NetStumbler1
1http://www.netstumbler.com/
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
911
Table 3. Simulation Parameters
Parameters Values
IEEE Standard 802.11b
Propagation Model Shadowing
Antennas Omnidirectional, 18dB gain
Router’s Carrier Sense Threshold -76dBm [IEEE 1999]
Router’s Receiver Sensitivity -80dBm [IEEE 1999]
Router’s Transmit Power 17dBm (WRT54G)
Frequency 2.422GHz (Channel 3)
Path Loss Exponent 1.59
Shadowing Deviation 5.4dB
software with a GPS device attached to it, measurements took place at each point, as
illustrated in Figure 3 for the Capacit point and discussed below.
Figure 3. Paths taken for data gathering
At each point, data was collected in eight different ways starting close to the point
and moving away from it until a signal level of -85 dB was reached. The directions of
measurement took into account the position of the point in regards to other neighboring
points. Once the measurements had been carried out, a final value was attributed to each
variable by calculating the average based on the data collected at each point.
In the simulator, a total of six Voice over Internet Protocol (VoIP) calls and
three background Paretto traffic flows were simulated. These calls involved the fol-
lowing points: Capacit/Graduac¸˜
ao Profissional, Reitoria/Capacit, Reitoria/CT, DI/CT,
Secom/Laborat´
orio, and DI/Secom, and the points with background traffic were:
DI/Laborat´
orios, Graduac¸˜
ao B´
asico/CT, e Secom/Graduac¸ ˜
ao Profissional. For each of
the metrics, jitter, delay, blocking probability and throughput were calculated, and are
shown in Figure 4. Since a VoIP call is bidirectional, each call is represented by two
flows. The same indicators were also calculated for the original OLSR, which uses hop
count as a metric. These points were selected so that the communication between them
happened through three hops at most and nodes really competed for the wireless medium.
For jitter, OLSR-ML had the highest variation amongst the metrics, as can be
seen in Figure 4a. OLSR-ETX and OLSR-MD had the best performance in regards to
912
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
throughput as shown in Figure 4b.
In Figure 4c, OLSR-ML had the best performance amongst the metrics, but its
values still are not considered appropriate for VoIP calls. This high-delay behavior is
explained by the number of different flows attempting to use wireless medium in the sim-
ulations. It is important to mention that, during the calls, the protocol was still computing
its routing tables during the beginning of the VoIP calls. This was done on purpose, in
order to determine the performance of each metric in the worst-case scenario. As for the
blocking probability, Figure 4d, OLSR-ML also had the highest values.
(a) Jitter per VoIP flow (b) Throughput per VoIP flow
(c) Delay per VoIP flow (d) Block Probability per VoIP flow
Figure 4. Results obtained from the simulations
It is important to remember that the main goal here is to choose the metrics with
the best performances in the scenario under consideration to be used in the proposal.
Original OLSR was simulated only for comparison purposes. Even having performed
better for some flows than the other metrics, it was not considered for use in WMN rout-
ing, since it is widely known that it does not take into account the quality of a given
link. Instead, it uses the shortest path approach, that is, the minimum number of hops
between a given source and destination. And that is not useful in a mesh network context
[Faccin et al. 2006].
From the results, the metrics ETX and MD had the best overall performance for
the considered scenario, and thus they were chosen for use with OLSR to improve its
routing table computations.
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
913
5.3. A Practical Example
For the following practical example, consider the scenario presented in Figure 2. As the
mesh network simulated high-gain antennas, every node is able to ”sense” all the other
nodes despite the distance between them. It is important to point out that this is not a
problem since every routing decision is made considering the link qualities, ETX and
MD.
Table 4 shows the routing table of each node in the network with the metrics’
values for every link towards all other nodes at a given time.
Table 4. Nodes and their respective routing tables
Firstly, Pruning is applied to get rid of links of unsatisfactory quality, and the
metric considered was ETX. To do this, the median is calculated and links with ETX
values equal to or greater than the median are pruned since this metric is to be minimized.
We could also consider MD for pruning threshold calculation since we are not prioritizing
any of these metrics as in [Saaty 1980]. In Table 4, the median is in bold digits and the
pruned links’ values are in italics.
As the first step of the AHP solution requires all possible routes to be found,
Pruning becomes very important. Dijkstra’s algorithm [Cormen et al. 2001] can be run to
find all these paths. Without Pruning, a total of 48.929 possible paths were found, whilst
with this technique the number of paths was reduced to 256. Thus, the goal of reducing
the computational complexity mentioned earlier is achieved.
Both metrics, ETX and MD, make use of Dijkstra’s algorithm to determine the
best route, that is, the one with the least cost among the possible routes found. For this
example, we wish to determine to best route between nodes 0 and 9. The best route,
through nodes 0 - 4 - 5 - 9, ended up to be the same for both ETX (7.0833) and MD
(0.931694).
Since this proposal aims to select routes which combine the best quality values of
the considered metrics, after the application of AHP and Pruning the best selected route
should be the same as for ETX/MD or another one even better. It is important to point out
that even though ETX was considered for Pruning, priority of metrics was not taken into
914
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
account and anprpcm in Step 5 remained unchanged. So, applying the remaining steps
of the solution in [Alkahtani et al. 2006], the route obtained was also through nodes 0 -
4 - 5 - 9 with a maximum total score of 0.0066343. Hence, we also attain the goal of
combining multiple metrics with OLSR in order to provide selection of the best route.
6. Conclusions and Current Work
The main goal of this paper was to provide QoS routing for Wireless Mesh Networks
through the application two simple techniques, Analytic Hierarchy Process and Pruning,
normally used in wired scenarios, and that together may improve OLSR’s performance
through the usage of multiple metrics.
Although the solution presented by [Alkahtani et al. 2006] was proposed for a
wired scenario, it does have great potential for application in a wireless network context.
However, the complexity of this solution may be high, due to the many matrix computa-
tions which are a result of the number of possible paths between a source and destination.
In WMNs, this complexity may become even greater, since most, or even all, of the nodes
are able to ”sense” the other ones present in the network.
With that in mind, the Pruning technique comes into play to lessen the number
of matrix computations, and thus reducing even more the complexity of the solution pre-
sented in [Alkahtani et al. 2006]. By applying Pruning, links with metric values that are
greater than or equal to a certain threshold will not be taken into account, and as a result
less computation will be needed. [Costa et al. 2002] successfully applied this technique,
which really improved their results.
Unlike only having improvements either on throughput or transmission delay,
when only a single metric is used for routing, this proposal guarantees improvements
based on both metrics at the same time, using the already known techniques of AHP and
Pruning, which together enhance OLSR’s route selection.
Currently, efforts are being undertaken to conclude the NS-module implementa-
tion of the proposal presented here, which will make it possible to really evaluate its
performance.
References
Akyildiz, I. F., Wang, X., and Wang, W. (2005). Wireless mesh networks: a survey.
Computer Networks, 47(4):445–487.
Alkahtani, A. M. S., Woodward, M. E., and Al-Begain, K. (2006). Prioritised best ef-
fort routing with four quality of service metrics applying the concept of the analytic
hierarchy process. Computers and Operations Research, 33(3):559–580.
Aslam, N., Phillips, W., and Robertson, W. (2004). Composite metric for quality of
service routing in olsr. Canadian Conference on Electrical and Computer Engineering,
2:759–762.
Badis, H., Munaretto, A., Agha, K., and Pujolle, G. (2003). Qos for ad hoc networking
based on multiple metrics: bandwidth and delay. Fifth IFIP International Conference
on Mobile and Wireless Communications Networks.
Bruno, R., Conti, M., and Gregori, E. (2005). Mesh networks: commodity multihop ad
hoc networks. IEEE Communications Magazine, 43(3):123–131.
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
915
Clausen, T. and Jacquet, P. (2003). Optimized link state routing protocol (olsr). RFC
3626, Experimental.
Cordeiro, W., Aguiar, E., Moreira, W., Abelem, A., and Stanton, M. (2007). Providing
quality of service for mesh networks using link delay measurements. 16th Interna-
tional Conference on Computer Communications and Networks, pages 991–996.
Cormen, T. H., Leiserson, C. E., Rivest, R. L., and Stein, C. (2001). Introduction to
Algorithms. The MIT Press, 2nd edition.
Costa, L. H. M. K., Fdida, S., and Duarte, O. C. M. B. (2002). Developing scalable
protocols for three-metric qos routing. Computer Networks: The International Journal
of Computer and Telecommunications Networking, 39(6):713–727.
DeCouto, D. S. J., Aguayo, D., Bicket, J., and Morris, R. (2003). A high-throughput path
metric for multi-hop wireless routing. 9th Annual International Conference on Mobile
Computing and Networking, pages 134–146.
Faccin, S., Wijting, C., Kneckt, J., and Damle, A. (2006). Mesh wlan networks: concept
and system design. IEEE Wireless Communications, 13(2):10–17.
Held, G. (2005). Wireless Mesh Networks. Auerbach Publications, 1st edition.
IEEE (1999). INSTITUTE OF ELECTRICAL AND ELECTRONIC ENGINEERING. IEEE
802.11, 1999 Edition (ISO/IEC 8802-11: 1999): Information Technology - Telecom-
munications and Information Exchange between Systems - Local and Metropolitan
Area Network - Specific Requirements - Part 11: Wireless LAN Medium Access Con-
trol (MAC) and Physical Layer (PHY) Specifications.
Jain, R. (1991). The Art of Computer Systems: Techniques for Experimental Design,
Measurement, Simulation, and Modeling. John Wiley & Sons, 1st edition.
Kowalik, K. and Davis, M. (2006). Why are there so many routing protocols for wireless
mesh networks? Irish Signal and Systems Conference.
Lee, M., Zheng, J., Ko, Y., and Shrestha, D. (2006). Emerging standards for wireless
mesh technology. IEEE Wireless Communications, 13(2):56–63.
Nandiraju, N., Nandiraju, D., Santhanam, L., He, B., Wang, J., and Agrawal, D. P. (2007).
Wireless mesh networks: Current challenges and future directions of web-in-the-sky.
IEEE Wireless Communications, 14(4):79–89.
NS (2007). The Network Simulator Manual - The VINT Project. URL:
http://www.isi.edu/nsnam/ns/.
Passos, D., Teixeira, D., Muchaluat-Saade, D., Magalh˜
aes, L., and Albuquerque, C.
(2006). Mesh network perfomance measurements. 5th International Information and
Telecommunication Technologies Symposium.
Saaty, T. L. (1980). The Analytic Hierarchy Process: Planning, Priority, Setting and
Resource Allocation. McGraw-Hill, Inc, 1st edition.
Wang, Z. and Crowcroft, J. (1996). Quality-of-service routing for supporting multimedia
applications. IEEE Journal on Selected Areas in Communications, 14(7):1228–1234.
Zhang, Y., Luo, J., and Hu, H. (2007). Wireless Mesh Networks: Architectures, Protocols
and Standards. Auerbach Publications, 1st edition.
916
26° Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos
