This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Ann. Telecommun. (2012) 67:215–226DOI 10.1007/s12243-011-0253-z
Multi-radio multi-channel routing metrics in IEEE 802.11sbased wireless mesh networks
Abstract IEEE 802.11s is one of the emerging stan-dards designed to build wireless mesh networks whichmay serve to extend the coverage of access networks.The default IEEE 802.11s path selection protocol Hy-brid Wireless Mesh Protocol (HWMP) is based on theradio-aware airtime link metric (ALM) that outper-forms the hop-count metric in single channel multi-hop wireless networks. However, this metric may leadto capacity degradation when multiple channels and/ormulti-radio are used. To fully exploit the capacity gainof multiple channels use, new routing metrics havebeen proposed such as weighted cumulative expectedtransmission time, metric of interference and channelswitching, interference aware routing metric, exclusiveexpected transmission time, and normalized bottlenecklink capacity. These metrics distribute the data trafficload among channels and/or radios to reach the finaldestination. In this paper, we provide a qualitativecomparison study that considers the characteristics ofthese metrics. Indeed, we substitute ALM by thesedifferent metrics, and we evaluate the performance of
S. Ghannay (B) · S. M. Gammar · F. KamounCRISTAL Laboratory, National School of ComputerSciences, La Manouba, Tunisiae-mail: [email protected]
Wireless mesh networks (WMNs) [1] have recentlygained increasing attention and have emerged as atechnology with great potential for a wide range ofapplications. WMNs are composed of wireless nodeswhich form the wireless backbone access. One or sev-eral nodes which belong to the mesh infrastructure canbe configured as gateway to allow access to externalnetworks such as the Internet. Moreover, some of meshnodes can play the role of access point to allow meshnetwork access to client stations. In mesh network,nodes are either stationary or with low mobility andhave ample energy supply. Each node in a WMN canbe equipped with multiple radios and each of the radioscan be configured in a different frequency channel,and communication among mesh nodes and client sta-tions requires the use of routing protocols that mustbe combined with routing metrics to determine theappropriate route toward the destination(s).
IEEE 802.11s [2] is the future technology for a par-ticular category of WMNs derived from the worldwideused 802.11 standard. It proposes a default routingprotocol called the Hybrid Wireless Mesh Protocol(HWMP) which is also referred as a path selectionprotocol since it uses MAC addresses for route com-puting and forwarding procedures. This protocol uses aradio-aware link metric called the airtime link metric
216 Ann. Telecommun. (2012) 67:215–226
(ALM) which takes into account the quality of thelink by considering the link capacity and the expectedloss rate.
Allowing multiple channels use in the same net-work is often presented as a possible way to en-hance the overall network performance given that,with proper design, supporting multiple channels hasseveral benefits, including increasing system through-put, decreasing end-to-end delay, and achieving betterload balancing. It is well established that when meshnodes have multiple radios, the shortest path routingalgorithms do not perform well and even ALM facesperformance limitations in these networks as it doesnot consider the existence of multiple channels. Con-sequently, several multi-radio multi-channel routingmetrics have been specifically developed for multipleradio multiple channel mesh networks such as weightedcumulative expected transmission time (WCETT) [18],metric of interference and channel-switching (MIC)[20], interference aware routing metric (iAWARE)[19], exclusive expected transmission time (EETT) [22],and normalized bottleneck link capacity (NBLC) [21].The ultimate objective of these metrics is to increasethe capacity of the WMN by taking into account theimpact of interference, traffic load, and packet loss rateon the quality of the routing paths.
This paper aims at investigating the performance ofrouting metrics for multi-channel multi-radio WMNsboth qualitatively and quantitatively. In particular, weidentify and compare the relevant characteristics ofthese metrics. Obtained results allow us to identify theappropriate use case of each metric.
A simulation-based study which was carried allowsus to identify appropriate utilization of each metric de-pending on network conditions. To our knowledge, ourwork constitutes one of the first up-to-date attemptswhich investigates extensively by simulations severalrouting metrics for multi-radio multi-channel wirelessmesh networks.
The remainder of the paper is structured as follows:In Section 2, we give an overview of the current ar-chitecture of IEEE 802.11s, and we highlight its maincharacteristics and challenges. Section 3 presents anoverview of multi-channel in wireless mesh networks.We present the HWMP in Section 4. In Section 5,we survey existing multi-radio multi-channel WMNsrouting metrics designed which are investigated in thispaper. Qualitative comparison is detailed in Section 6,while a simulation-based performance study is pre-sented in Section 7. Section 8 concludes this paper andoutlines future directions.
2 Overview of the IEEE 802.11s standard
The essential motivation for the 802.11s standard wasto provide a means by which a wireless backbonecould be built with minimal configuration effort. Thiscould be used in scenarios such as office buildings,home networking, and apartment blocks. The IEEE802.11s working group specifies an extension to theIEEE 802.11 MAC to solve the interoperability prob-lem by defining the whole architecture and the requiredprotocols and mechanisms. The 802.11s draft standarddefines three key components [2] (Fig. 1):
– Mesh portal point (MPP): The MPP is connectedto the wired gateway. A number of MPPs canexist in any 802.11s-based mesh network. Routingprotocols determine which MPP should be used togain access to the wired infrastructure. The MPPsperiodically broadcast a specific signaling messageto announce their presence to other nodes in themesh.
– Mesh point (MP): The MP has less sophisticatedfunctionality. In essence, it acts as a layer 2 routeras it determines how to route packets through themesh backbone toward the destination.
– Mesh access point (MAP): The MAP behaves sim-ilar to the MP, except that it also provides a legacy802.11 interface with which legacy nodes (STAs)can associate to the system.
All the above-mentioned elements could have morethan one radio interface, so each node has to be ableto handle more than one radio channel and implement
specific mechanisms to coordinate between channels.IEEE 802.11s WMN have several characteristics:
– Coverage: The IEEE 802.11s standard aims atbuilding a small (to medium)-scale WLAN meshnetwork. Practically, each MAP can be connectedto many STAs enabling the entire network to ac-commodate several hundred terminals. MultipleWLAN mesh networks can also be interconnectedto further expand network scale.
– Fixed topology: Mobility depends on the type ofmesh nodes. In fact, mesh point usually have min-imal mobility, while mesh clients can be station-ary or mobile nodes. This characteristic imposesspecific requirements for the design of routingmetrics. Moreover, the tree-like structure of thenetwork needs also the design of new protocolsunlike ad hoc network where the mobility of nodesrequires the design of routing metrics that canefficiency maintain connectivity.
– Power consumption: It depends on the type of meshnodes. Mesh routers usually do not have strict con-straints on power consumption. That is why routingmetric and protocol does not worry on energy.
– Distributed control: One of the most attrac-tive features of wireless mesh networks is self-organization. Various functions, such as mediumaccess control and routing, are carried out in afully distributed manner with minimal human in-tervention. They are not subject to any centralizednetwork management processes such as practicedin wired networks.
3 Multi-channel in WMN
Nodes in mesh network can be equipped with multipleradios. Each radio can have a particular channel toenhance network capacity. Hence, it enables nodes totransmit and receive simultaneously. Besides, a nodecan transmit on two channels simultaneously with tworadios. Multi-channel can improve robustness, connec-tivity, and performance.
Several researchers [4] have proposed MAC proto-cols based on IEEE 802.11 for utilizing multiple chan-nels. In our previous work [5], we have studied multi-channel solutions and we have classified them in threecategories.
The first category consists on channel allocation pro-posals done at the MAC level independently to theother layers. In [6], the authors propose to statically
configure the interfaces of different nodes on pre-known channels. A second possibility [7, 8] that canbe considered as an improvement of the first one [6]is to frequently switch the interfaces of a node amongdifferent channels. This approach requires synchroniza-tion between nodes. A hybrid approach [9, 10] keepssome interfaces of each node fixed, while others canswitch among channels.
The second category consists on a channel allocationapproaches done by a modified MAC collaboratingwith upper layers. An approach called MESTIC [11]aims to improve the aggregate throughput of the net-work. The MESTIC is a fixed, rank-based, polynomialtime greedy algorithm for centralized channel assign-ment, which visits every node once. The node’s rankcomputation depends on its link traffic characteristics,topological properties, and number of its network in-terface cards (NICs).
Finally, the third category concerns channel allo-cation methods implemented in a new layer result-ing from a common-layer design between MAC andnetwork layer. Promising works have been proposedin [12, 13]. Raniwal et al. [12] proposes a centralizedsolution. This solution is “load-aware channel assign-ment” approach. In fact, given the initial routes forthe node pairs and thus the traffic load on each virtuallink, the radio channel assignment algorithm assigns aradio channel to each interface, such that the amountof bandwidth made available to each virtual link is noless than its expected load.
An enhancement of the previous described solutionis a distributed load channel assignment solution [13].The approach mechanism is composed by two phases.The first phase that is the initial step consists on adistributed route discovery update protocol developedto establish routes between multi-channel WMN nodesand wired gateway. Then, the second and periodicphase is the distributed load aware channel assignmentwhere each node separates its set of interfaces into UP-NIC that will be affected by their parents and DOWN-NIC that involves only the node itself for affectation.
4 Routing in IEEE 802.11s WMN
As already mentioned, the most innovative elementspresent in 802.11s stays in the idea to move the routingfunctionality, typical for the IP layer, on the Data-Link layer. The routing scheme which is mandated inthe standard is the HWMP. This is an hybrid routingprotocol which combines the flexibility of on-demand
218 Ann. Telecommun. (2012) 67:215–226
routing with a proactive topology extension. The on-demand primitives are based on those of the Ad HocOn-Demand Distance Vector (AODV) routing proto-col. The proactive component is based on a distance-vector protocol. A routing metric which can be usedwith either of these routing protocols is also definedin the standard. This is the ALM, which reflects theutilization of a wireless interface. The ALM metric isdefined as follows:
Ca =[
O + Bt
r
]1
1 − ef(1)
where r is the transmission bit rate, ef is the frame errorrate for a test frame of size Bt, and O is a constantgiving the channel access overhead. ALM estimates thequality of a link by taking into account the packet lossprobability as well as the bit rate of the link. HWMPhas two modes of operations as discussed below.
4.1 Radio-metric Ad Hoc On-Demand DistanceVector
In radio-metric AODV, if a source MP needs a route toa destination, it broadcasts a path request (PREQ) mes-sage with the destination MP address specified in thedestination field and the ALM metric initialized to 0.When an MP receives a PREQ, it creates a path to thesource or updates its current path whether the PREQcontains a greater sequence number or the sequencenumber is the same as the current path and the PREQoffers a better metric than the current path. Wheneveran MP forward a PREQ, the metric field in the PREQ isupdated to reflect the cumulative metric of the path tothe PREQ’s source. After creating or updating a pathto the source, the destination MP sends a unicast pathreplay (PREP) message back to the source.
4.2 Proactive RANN mechanism
The proactive tree-based routing is applied when aroot node (for example, an MPP playing the role ofa gateway to the Internet) is configured in the meshnetwork. The root MP periodically propagates a routeannouncement (RANN) message into the network.The information contained in the RANN is used toannounce and update by intermediate nodes the valueof the path metric toward the root MP. Upon receptionof a RANN, each MP that has to create or refresh apath to the root MP sends a unicast PREQ to the rootMP via the MP from which it received the RANN.The unicast PREQ follows the same processing rules
defined in the on demand mode. The root MP sendsa PREP message in response to each received PREQ.The unicast PREQ creates the reverse path from theroot MP to the originating MP while the PREP createsthe forward path from the MP to the root MP.
5 Multi-radio multi-channel routing metrics
The hop-count metric sets up paths based on the num-ber of hops and ALM tries to avoid overloaded links,but both of them have poor performances multi-radiomulti-channel WMNs. For that reason, current researchefforts are looking for other metrics to overcome theselimitations. Several metrics have been specifically de-signed for multi-radio multi-channel WMN. Those whoare considered the most promising are detailed andcompared in the remaining of this paper.
5.1 WCETT
WCETT [18] is an interference-aware routing metricexpressed for a path p as follows:
WCETT (p) = (1 − α)∑
link l∈p
ETTl + α max1≤ j≤k
X j (2)
where α is a configured parameter subject to 0 ≤ α ≤ 1.
ETTl = SB
× ETXl (3)
where S is the size of the probe packet and B is thebandwidth of the link l. To calculate ETT for each link,an MP needs to measure both terms ETX and B. ETXis defined as follows:
ETX = 11 − ef
(4)
where ef is the frame error rate. X j is estimated asfollows:
X j =∑
Hop i is on channel j
ETTi (5)
The first part of the metric expresses the expectedtransmission time on a path p while the second partX j is the sum of transmission time of hops on channelj . The total path throughput will be dominated by thebottleneck channel which has the largest X j . The bestpath is that having the less WCETT.
Ann. Telecommun. (2012) 67:215–226 219
5.2 MIC
The main objective of the MIC metric [20] is to protectflows against the inter-flow and intra-flow interference.MIC is composed of two metrics: interference-awareresource usage (IRU) and channel switching cost(CSC). It is computed from a path p as follows:
MIC (p) = 1N × min (ETT)
∑link l∈p
IRUl +∑
node i∈p
CSCi
(6)
where N is the number of nodes and min(ETT) is theminimal delay between a pair of nodes in the network.
IRU is the product of delay of the link l and the totalnumber of neighbors that may be interfered with by thetransmission activities of link l computed as follows:
IRUl = ETTl × Nl (7)
CSC captures the intra-flow interference because itgives a bigger weight to paths with consecutive linksusing the same channel. To express intra-flow interfer-ence, CSC at a node X is determined as follows:
CSCX = w1 if CH(prev (X)
) �= CH (X) (8)
CSCX = w2 if CH(prev (X)
) = CH (X) (9)
0 ≤ w1 < w2 (10)
where prev(X) is the channel used by the previous hopof node X and CH(X) is the channel that node X usesto transmit to the next hop. The relationship w1 < w2
captures the fact that due to intra-flow interference,using the same channel at node X and prev(X) imposesa higher cost than using different channels.
5.3 iAWARE
The iAWARE metric [19] considers link-quality vari-ation. This metric uses the signal-to-noise ratio (SNR)and the signal-to-interference and noise ratio (SINR)to continuously updating the routing metrics accord-ing to the variations of the neighboring interference.The iAWARE metric estimates the average time themedium is busy because of the transmissions from eachinterfering neighbor. The higher the interference, thehigher the iAWARE value.
iAWARE (p) = (1 − α)∑
link l∈p
iAWAREl + α max1≤ j≤k
X j
(11)
The metric for a given link l is:
iAWAREl = ETTl
IRl(12)
IRl = min (IRl (u) , IRl (v)) where l = (u, v) (13)
IRl (u) = SINRl (u)
SNRl (u)(14)
In order to exploit the channel diversity and to findpaths with less intra-flow interference, X j is defined asfollows:
X j =∑
Hop i is on channel j
iAWAREi (15)
5.4 EETT
EETT [22] is adopted for large-scale multi-radio meshnetworks. It was proposed to select multi-channelroutes with the least interference for long paths tomaximize the end-to-end throughput. For a given linkl, the metric is expressed as follows:
EETTl =∑
Link i ∈ IS(l)
ETTi (16)
where IS(l) is the interference set which includes thelink it self. The metric for a given path p is computed asfollows:
EETT (p) =∑
Link l ∈ p
EETTl (17)
EETTl of a link l represents the busy degree of thechannel used by link l. If there are more neighboringlinks operating in the same channel as link l, link l mayhave to wait for a longer period before doing a trans-mission on that channel. As a result, a path with a largerEETT indicates that it has a more severe interferenceand needs more time to complete the transmission overall links composing the path.
5.5 NBLC
The main idea behind the NBLC [21] metric is toincrease the system throughput by evenly distributingtraffic load among channels and among nodes. TheNBLC metric is defined as follows:
NBLC (p) = minlink i ∈ p
(RLCi
CEBTi,p
)γ L (18)
RLCi = Tm − maxj∈{i,Ii}
(BusyPeriod j [n]
)(19)
RLC is the residual link capacity which means the per-centage of free-to-use channel air time on an outgoinglink. It is the difference between a measurement period
220 Ann. Telecommun. (2012) 67:215–226
(Tm) and the maximum of busy period of interferingneighbors located in Ii. The busy period is the timespent in transmission, reception, or carrier sensing.
CEBTi,p =∑
i∈{Ix∩P}ETTi (20)
The cumulative expected busy time (CEBT) of alink i on a path p is the sum of the ETT values forthe path’s links operating on the same channel as iand interfere with the link i. γ is a tunable parameterimplicitly indicating the probability of a packet beingdropped by an intermediate node in path p of length L.
6 Qualitative comparison of routing metrics
Several papers have compared routing metrics. Camp-ista et al. [14] analyzes the state-of-the-art in WMNmetrics and propose a taxonomy for WMN routingprotocols. It classifies them into four categories: ad hoc-based, controlled-flooding, traffic-aware, and oppor-tunistic. Moreover, [14] presents performance resultsobtained with different metrics such as hop count, ETX,ETT, etc. by varying the number of mesh nodes ina WMN testbed. This work did not evaluate multi-channel metrics such as WCETT. In addition, [14] didnot vary some important parameters such as the trafficload. Finally, this work evaluates the performance ofthe OLSR protocol whereas HWMP is the default pro-tocol in the upcoming standard IEEE 802.11s.
The book chapter in [15] identifies different cate-gories of routing metrics proposed for wireless meshnetworks and describes the rationale of each category.The presentation of the metrics is highly comparative,with emphasis on the strengths and the weaknessesof both individual and whole families of metrics. Thiswork did not contain a performance study of thesesrouting metrics.
In the next section, we will present a comparativestudy of multi-radio multi-channel routing metrics anda simulation-based performance analysis, with empha-sis on the use case of each metric.
In Table 1, we provide a comparison of the maincharacteristics of the metrics discussed in the previoussection. All routing metrics consider some commonfactors such as the path length (PL), the data rate (DR),and the packet loss rate (PLR). Multi-radio multi-channel metrics consider also channel diversity (CD).Medium instability (MI) is considered by iAWAREwhile link load (LL) is considered by NBLC.
In addition, each metric has a specific feature.WCETT considers intra-flow interference. The intra-flow interference is the use of same channel on multiple T
able
1Q
ualit
ativ
eco
mpa
riso
nof
mul
ti-r
adio
mul
ti-c
hann
elm
esh
rout
ing
met
rics
AL
MW
CE
TT
MIC
iAW
AR
EE
ET
TN
BL
C
Fac
tors
PL
PL
PL
PL
PL
PL
DR
(B)
DR
(B)
DR
(B)
DR
(B)
DR
(B)
DR
(B)
PL
R(E
TX
)P
LR
(ET
X)
PL
R(E
TX
)P
LR
(ET
X)
PL
R(E
TX
)P
LR
(ET
X)
CD
(Xj)
CD
(Nla
ndC
SC)
CD
(Xj)
CD
(EE
TT
l)C
D(C
EB
T)
MI
(SN
R)
LL
(bus
ype
riod
)Sp
ecif
icfa
ctor
s–
Intr
a-fl
owIn
tra-
flow
(IR
U)
and
Lin
kqu
alit
yva
riat
ion
(IR
)In
ter-
flow
Res
idua
llin
kca
paci
ty(R
LC
)In
terf
eren
ce(X
j)in
ter-
flow
(CSC
)In
tra-
flow
inte
rfer
ence
(Xj)
Inte
rfer
ence
(EE
TT
l)In
tra-
flow
inte
rfer
ence
(CE
BT
)in
terf
eren
ceA
dvan
tage
sSi
mpl
ede
sign
Cha
nnel
dive
rsit
ySo
me
load
bala
ncin
gE
xter
nali
nter
fere
nce
Mor
ech
anne
ldiv
ersi
tyL
oad
bala
ncin
gan
dch
anne
ldiv
ersi
tyan
dm
ediu
min
stab
ility
for
long
path
sD
isad
vant
ages
Doe
sno
tcon
side
rC
onsi
der
inte
rfer
ence
even
Can
caus
ero
ute
Doe
sno
tcon
side
rD
oes
notc
onsi
der
inte
rfer
ence
the
node
isno
tinv
olve
din
stab
ility
intr
a-fl
owin
terf
eren
cein
ter-
flow
inte
rfer
ence
inan
ytr
ansm
issi
onR
outi
ngst
rate
gyA
llO
nde
man
dA
llbu
tcom
plic
ated
On
dem
and
All
On
dem
and
Dis
tanc
eve
ctor
for
link
stat
eD
ista
nce
vect
orD
ista
nce
vect
orA
ppro
pria
teM
ono-
radi
oN
etw
ork
wit
hlo
ngpa
ths
Hig
hnu
mbe
rof
Hig
htr
affi
cra
teL
arge
-sca
lene
twor
kL
inks
heav
ilylo
aded
use
case
netw
ork
and
low
load
tran
smit
ters
Ann. Telecommun. (2012) 67:215–226 221
consecutive hops of a path. It increases the bandwidthconsumption of the flow at each of the nodes alongthe path. Besides, MIC captures intra- and inter-flowinterference. Indeed, a flow through wireless links is notonly consuming bandwidth of the nodes along a pathbut it also contends for bandwidth with the nodes thatare in the neighboring area of its path. Consequently,iAWARE considers intra-flow interference and linkquality variation. In fact, it uses the SNR which isobtained from the device driver of the wireless inter-face, and it is considered as one of the most importantmetrics used to measure the quality of link. Moreover,iAWARE takes into account the external interferencewhich is the kind of interference resulting from non-cooperating communication entities external to themesh network and operating on the same frequencyband. As far as is concerned, EETT captures the inter-flow interference while NBLC accounts for the trafficload within a links interference range and uses theresidual link capacity of a path to judge its goodness.
Each metric has both advantages and disadvantages.ALM is simple but it does not consider interference,and hence, it cannot balance the traffic load in meshnetworks. On the other hand, WCETT only considersintra-flow interference, but it may create severe inter-flow interference especially for dense areas since thistype if interference is not considered. The MIC metrictakes channel diversity into account. However, it con-siders interference even if this node is not involved inany transmission. The iAWARE metric captures themedium instability and aims at selecting good qualitypaths. Nevertheless, iAWARE is sensitive to link trafficand presence of interfering traffic among neighboringnode. For that reason, iAWARE can affect route stabil-ity. It may cause frequent changes of established pathsand disrupt normal network operation. The EETT met-ric gives more channel diversity for long path, whereasit does not consider intra-flow interference. The NBLCmetric allows the load balancing in order to increase thesystem throughput by distributing traffic load amongdifferent nodes using different frequency channels. Onedisadvantages of NBLC is that it does not considerinter-flow interference.
It is worth mentioning that routing protocols mayimpose different requirements for the design and theuse of the considered routing metrics. In particular,the nature of the routing protocols (on-demand orproactive) is important to know in order to determinewhich metrics are more convenient to adopt given thatthe update of these metrics is done through the controlmessages of the routing protocols. Another importantfactor to consider is the isotonicity of the metric whichmeans that a metric should ensure that the order of
the weights of two paths is preserved when they arefollowed or preceded by the same link. Non-isotonicmetrics cannot be used with link-state routing protocolswhich means that they are not an efficient algorithmsuch as Bellmand-Ford and Dijkstra to calculate freeloop paths. ALM, MIC, and EETT can be used byall routing strategies as they are isotonic, but MICuses a complicated decomposition to become isotonic.However, WCETT, iAWARE, and NBLC are non-isotonic, and they can be used only by on-demandrouting protocols.
Based on all the observations made above, we cansay that each metric has its appropriate use cases whichdepend on network conditions such as density, trafficload, and available frequency channels. For instance,ALM can be used in mono-radio mesh networks,whereas WCETT is more suitable for multi-radio multi-channel mesh networks. On the other hand, MIC ispreferred in networks where there is a high numberof transmitters. Hence, iAWARE can lead to goodperformance in network that have high traffic rate.Large-scale networks will favor the EETT metric whileNBLC can be used in networks with high links load.The simulation-based performance study that will begiven in the next section will confirm these remarks.
7 Simulation-based comparison of routing metrics
To compare the performance of multi-radio multi-channel metrics, we use the Qualnet simulator [23].Every node has two radios and each radio can beconfigured to one of three available frequency chan-nels. The sources of the flows are randomly located, andall flows are CBR flows with 512 Byte packets. Eachpacket is sent every 50 ms. The transmission range is250 m and the carrier sensing range is 500 m. The valueof α in equation 2 and 11 was set to 0.5. There areN static nodes deployed in a density of one node per90 × 90 square. We evaluate the average applicationthroughput and the average end-to-end packet delay.
The bandwidth expressed by all the routing metrics iscalculated using the technique of packet pair [18]. Thepacket loss rate ef is calculated using the forward pf andreverse pr delivery ratio of the link to account for dataas well as ACK frames.
ef = 1 − (1 − pf) × (1 − pr) (21)
In order to compute each multi-radio multi-channelpath metric when the PREQ or the RANN propagatethrough the network, we need to compute the link met-ric for each traversed link and to know the frequency
222 Ann. Telecommun. (2012) 67:215–226
channel in each link is operating. Therefore, we slightlymodified the PREQ packet in case of on-demand modeand the RANN packet in case of proactive RANN inorder to carry the link metric and the channel of eachtraversed link.
In this experiment, we consider two main scenarios.The first one is the ad hoc scenario in which we evaluatethe performance of the on-demand mode of HWMPprotocol based on AODV. In the second scenario, weconsider a wireless backhaul network, in which wechoose a randomly node as the gateway to the wirednetwork. All CBR connections are destined to thisparticular node called the root and the routing protocolis the proactive RANN.
Figure 2a–c shows the average application through-put as a function of the per-flow data rate, the numberof CBR connections, and the number of nodes, respec-
tively, using the AODV protocol. The performance ofALM for all scenarios is not surprising either, since themetric only considers link loss rate and link bandwidthwhen selecting a path. It does not attempt to select paththat are channel diverse. In fact, ALM did not explicitlyconsider the impact of contention due to traffic fromnearby nodes. The contending traffic affects the link intwo ways. First, it may increase the packet loss as wellas the end-to-end delay due to collisions and secondreduces the available bandwidth.
In Fig. 2a, NBLC and iAWARE have the highestapplication throughput. In fact, NBLC accounts for thetraffic load within a link interference range and usesthe residual capacity of a path. Concerning iAWARE,it captures the interfering traffic when computing theweights of the links so that it chooses links that are lessaffected by the interfering nodes.
50000
100000
150000
200000
250000
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Ave
rage
thro
ughp
ut(b
its/s
)
Per flow rate (Mbits/s)
ALMWCETT
MICiAWARE
EETTNBLC
20000
40000
60000
80000
100000
120000
10 15 20 25 30 35
Ave
rage
thro
ughp
ut(b
its/s
)
Number of CBR connections
ALMWCETT
MICiAWARE
EETTNBLC
40000
50000
60000
70000
80000
90000
100000
20 30 40 50 60 70 80 90 100
Ave
rage
thro
ughp
ut(b
its/s
)
Number of nodes
ALMWCETT
MICiAWARE
EETTNBLC
(a)
(c)
(b)
Fig. 2 Average CBR application throughput using AODV by varying a the per-flow data rate, b the number of CBR connections, andc the number of MPs
Ann. Telecommun. (2012) 67:215–226 223
The EETT and MIC metrics have approximately thesame throughput. The reason for that is that MIC ex-presses intra-flow interference, but to express the inter-flow interference, it considers all interfering neigh-bor regardless of their state (transmission or other).Whereas, EETT accounts for the ETT of interferingneighbor which express more precisely if the node istransmitting, but it does not consider the intra-flowinterference.
In Fig. 2b, the MIC metric has the highest through-out. Note that MIC incorporates inter-flow interferenceby scaling up the ETT of a link by the number ofneighbors interfering with the transmission on that link.In fact, the term IRU in the MIC metric assigns links
with less interfering neighbors with a lower metric. Inthis figure, the number of transmitters increases andleads to more interferers. Moreover, MIC balancesnetwork load and reduces both intra-flow and inter-flow interference. NBLC and iAWARE have also goodperformance as they account for link quality and loadbalancing.
In Fig. 2a, b, the WCETT metric has poor perfor-mance compared to the other multi-channel metricsbecause the weighted average in WCETT implicitlyassumes that the network is not too heavily loaded.In Fig. 2c, the results obtained by almost all multi-radio multi-channel metrics are similar. It can be ob-served then that both EETT and WCETT metrics have
0
2
4
6
8
10
12
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
End
-to-
end
pack
et d
elay
(s)
Per flow rate (Mbits/s)
ALMWCETT
MICiAWARE
EETTNBLC
0
0.5
1
1.5
2
10 15 20 25 30 35
End
-to-
end
pack
et d
elay
(s)
Number of CBR connections
ALMWCETT
MICiAWARE
EETTNBLC
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
20 30 40 50 60 70 80 90 100
End
-to-
end
pack
et d
elay
(s)
Number of nodes
ALMWCETT
MICiAWARE
EETTNBLC
(a)
(c)
(b)
Fig. 3 Average end-to-end packet delay using AODV by varying a the per-flow data rate, b the number of CBR connections, andc the number of MPs
224 Ann. Telecommun. (2012) 67:215–226
good performance in several configurations. Indeed,EETT is suitable for networks which have long pathsand different channel distributions. Moreover, as men-tioned before, the benefit of WCETT is more interest-ing for shortest paths, but from these results, we canconclude that even on paths that are long, WCETT isable to provide acceptable performance. As expected,the MIC metric has low throughput which can be ex-plained by the fact that MIC favors links incident onnodes with less number of interfering neighbors what-ever the degree of interference caused by neighbors.This results in finding paths along the boundary of thenetwork where nodes have less number of neighborsand finding longer paths.
Figure 3a–c shows the average end-to-end packetdelay as a function of the per-flow data rate, the num-ber of CBR connections, and the number of nodes,respectively, using the on-demand mode of HWMPprotocol. In Fig. 3a, b, NBLC introduces the lowestend-to-end packet delay. As noted before, since theNBLC metric favor less congested route, it has shorterqueuing delay for packet at intermediate nodes than theother metrics. In addition, NBLC captures the channelutilization which is an important factor. Indeed, as thechannel utilization increases, the cost becomes moreexpensive since packet delay and loss ratio at the chan-nel increase and become more sensitive to traffic burst.iAWARE and MIC introduces less delay than WCETT
40000
50000
60000
70000
80000
90000
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Ave
rage
thro
ughp
ut(b
its/s
)
Per flow rate (Mbits/s)
ALMWCETT
MICiAWARE
EETTNBLC
(a)
20000
30000
40000
50000
60000
70000
80000
90000
10 15 20 25 30 35
Avr
eage
thro
ughp
ut(b
its/s
)
Number of CBR connections
ALMWCETT
MICiAWARE
EETTNBLC
(b)
20000
30000
40000
50000
60000
70000
80000
20 30 40 50 60 70 80 90 100
Avr
eage
thro
ughp
ut(b
its/s
)
Number of nodes
ALMWCETT
MICiAWARE
EETTNBLC
(c)
Fig. 4 Average CBR application throughput using RANN by varying a the per-flow data rate, b the number of CBR connections, andc the number of MPs
Ann. Telecommun. (2012) 67:215–226 225
and EETT since they account for interfering neighborsand intra-flow interference which reduce packet de-lay, delay jitter, and packet loss rate. Finally, WCETTand EETT induce more latency than the other multi-channel metrics except for large networks having lowtraffic load.
In the second scenario, we evaluate the performanceof the proactive RANN protocol. Figure 4a–c presentsthe average application throughput as a function of per-flow data rate, the number of CBR connections, andthe number of nodes. The topology structure imposesthat links that are close to the root would processhigher traffic load, and they should then given morebandwidth than others. This means that these links
should use a radio channel that is shared among a fewernumber of nodes. Metrics that capture link load suchas NBLC, MIC, and iAWARE have best performancein term of highest average throughput and smallestaverage end-to-end packet delay as shown in Fig. 5a–c since they satisfy most of the requirement of routingmetric design.
8 Conclusion and future directions
In this article, we focused on the routing issue in multi-channel IEEE 802.11s-based WMNs. We survey pathselection metric and protocol designated for IEEE
0
2
4
6
8
10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
End
-to-
end
pack
et d
elay
(s)
Per flow rate (Mbits/s)
ALMWCETT
MICiAWARE
EETTNBLC
(a)
0
0.5
1
1.5
2
2.5
3
3.5
10 15 20 25 30 35
End
-to-
end
pack
et d
elay
(s)
Number of CBR connections
ALMWCETT
MICiAWARE
EETTNBLC
(b)
0
0.5
1
1.5
2
2.5
20 30 40 50 60 70 80 90 100
End
-to-
end
pack
et d
elay
(s)
Number of nodes
ALMWCETT
MICiAWARE
EETTNBLC
(c)
Fig. 5 Average end-to-end packet delay using RANN by varying a the per-flow data rate, b the number of CBR connections, and c thenumber of MPs
226 Ann. Telecommun. (2012) 67:215–226
802.11s future standard and routing metrics designatedfor multi-channel multi-radio WMN (WCETT, MIC,iAWARE, EETT, NBLC). All metrics, except ALM,take channel diversity into account. For that reason,they offer best performance in term of applicationthroughput and end-to-end packet delay compared toALM. However, each metric considers a specific factorwhen judging the goodness of a path such as traffic load,medium stability, etc. Moreover, every metric has anappropriate use case which depends on network con-ditions. From the simulation results, we identified theconditions of applicability of each metric. For instance,iAWARE is preferred when there are a lot of changesin interfering traffic. NBLC is more suitable if networklinks are heavily loaded due to the high number ofconnections or the high data rate of a considerablenumber of communications. On the other hand, theMIC metric is more appropriate if there is a largenumber of interfering neighbors who transmit at thesame time as the sender. Finally, WCETT or EETTcan be used if the network is large and the traffic islow. ALM metric is well adapted to mono-radio mono-channel mesh networks due to its simple design.
Overall we can conclude from the comparison studyconducted in this paper that NBLC, iAWARE, andMIC are the most appropriate metrics since IEEE802.11s mesh networks are designed to support highmixed (internal and external) traffic load. One of thefuture directions that we aim to investigate is the rela-tion between routing and multichannel. Indeed, chan-nel assignment depends on the load on each link, whichin turns depends on routing. Therefore, routing andchannel assignment are interdependent in multi-radiomulti-channel WMN. Hence, the next step is to developan algorithm to joint routing and channel allocation inWMN.
References
1. Akyildiz IF, Wang X, Wang W (2005) Wireless mesh net-works: a survey. Comput Netw 47(4):445–487
2. IEEE P802.11s/D8.0 (2009) Draft standard for informationtechnology telecommunications and information exchangebetween systems local and metropolitan area networksspecific requirements—part 11: wireless LAN medium accesscontrol (MAC) and physical layer (PHY) specifications
3. Wang X, Lim AO (2008) IEEE 802.11s wireless mesh net-works: framework and challenges. Ad Hoc Netw 6(6):970–984
4. Crichigno J, Wub MY, Shu W (2007) Protocols and architec-tures for channel assignment in wireless mesh networks. AdHoc Netw 6:1051–1077
5. Faouzi F, Ghannay S, Filali F (2010) Channel allocation androuting in wireless mesh networks: a survey and qualita-tive comparison between schemes. Int J Wirel Mobile Netw(IJWMN) 2(1):132–150
6. Draves R, Padhye J, Zill B (2010) Routing in multi-radio,multi-hop wireless mesh networks. In: Proc of MobiCom
7. Bahl P, Chandra R, Dunagan J (2004) SSCH: slotted seededchannel hopping for capacity improvement in IEEE 802.11ad-hoc wireless networks. In: Proc. of MobiCom ’04
8. So J, Vaidya N (2004) Multi-channel MAC for ad hoc net-works: handling multi-channel hidden terminals using a singletransceiver. In: Proc of ACM international symposium onmobile ad hoc networking and computing (MOBIHOC)
9. Pradeep K, Vaidya NH (2006) Routing and link-layer pro-tocols for multi-channel multi-interface ad hoc wireless net-works. In: ACM SIGMOBILE mobile computing and com-munications review, vol 10, pp 31–43
10. Wu SL, Lin CY, Tseng YC, Sheu JP (2000) A new multi-channel MAC protocol with on-demand channel assignmentfor multi-hop mobile ad hoc networks. In: International sym-posium on parallel architectures, algorithms and networks(ISPAN)
11. Skalli H, Ghosh S, Das SK, Lenzini L, Conti M (2007) Chan-nel assignment strategies for multiradio wireless mesh net-works: issues and solutions. IEEE Commun Mag 1:86–95
12. Raniwal A, Gopalan K, Chiueh T (2004) Centralized channelassignment and routing algorithms for multi-channel wire-less mesh networks. In: SIGMOBILE, mobile computing andcommunications review
13. Raniwala A, Chiueh P (2005) Architecture and algorithms foran IEEE 802.11-based multi-channel wireless mesh network.In: Proc of IEEE infocom
15. Parissidis G, Karaliopoulos M, Baumann R, SpyropoulosT, Plattner B (2009) Routing metrics for wireless meshnetworks. In: Guide to wireless mesh networks. Springer,London, pp 199–230
16. Perkins C, Belding-Royer E, Das S (2003) Ad hoc on-demanddistance vector (AODV) routing. IETF RFC 3561
17. Bahr M (2006) Proposed routing for IEEE 802.11s WLANmesh networks. In: Proc of ACM international conference.The 2nd annual international workshop on wireless internet
18. Draves R, Padhye J, Zill B (2004) Routing in multi-radio, multi-hop wireless mesh networks. In: Proc. of ACMMobiCom
19. Subramanian AP, Buddhikot MM, Miller S (2006) Interfer-ence aware routing in multi-radio wireless mesh networks.In: Proc. of the IEEE workshop on wireless mesh networks(WiMesh)
20. Yang Y, Wang J, Kravets, R (2005) Designing routing met-rics for mesh networks. In: Proc. of the IEEE workshop onwireless mesh networks (WiMesh)
21. Liu T, Liao W (2008) On routing in multichannel wire-less mesh networks: challenges and solutions. IEEE Netw22(1):13–18
22. Jiang W, Liu S, Zhu Y, Zhang Z (2007) Optimizing routingmetrics for large-scale multi-radio mesh networks. In: Proc.of international conference on wireless communications, net-working and mobile computing (WiCom)