arXiv:1203.2730v1 [cs.NI] 13 Mar 2012
ON USING MULTIPLE QUALITY LINK METRICS WITH DESTINATION SEQUENCED
DISTANCE VECTOR PROTOCOL FOR WIRELESS MULTI-HOP NETWORKS
N. Javaid‡, A. Bibi‡, Z. A. Khan$, K. Djouani†,♯
‡Department of Electrical Engineering, COMSATS, Islamabad, Pakistan.
$Faculty of Engineering, Dalhousie University, Halifax, Canada.
†F’SATIE, TUT, Pretoria, South Africa.
♯LISSI, Universit´e Paris-Est Cr´eteil (UPEC), France.
In this paper, we compare and analyze performance of ﬁve
quality link metrics for Wireless Multi-hop Networks (WMhNs).
The metrics are based on loss probability measurements; ETX,
ETT, InvETX, ML and MD, in a distance vector routing pro-
tocol; DSDV. Among these selected metrics, we have im-
plemented ML, MD, InvETX and ETT in DSDV which are
previously implemented with different protocols; ML, MD,
InvETX are implemented with OLSR, while ETT is imple-
mented in MR-LQSR. For our comparison, we have selected
Throughput, Normalized Routing Load (NRL) and End-to-
End Delay (E2ED) as performance parameters. Finally, we
deduce that InvETX due to low computational burden and link
asymmetry measurement outperforms among all metrics.
Index Terms—DSDV, OLSR, ETX, Inverse ETX, ML,
MD, ETT, IBETX, ELP, distance vector,loss probabilities
A routing protocol is responsible for signiﬁcant performance
from the underlying wireless network. A routing link met-
ric is a key component of a routing protocol. As, it ﬁnds all
possible end-to-end routes and also the fastest route. Mini-
mum Hop-count; non-quality link metric is the most popular
and is IETF standard metric . It is appropriately used by
Wireless Ad-hoc Networks, where the objective is to ﬁnd new
paths as fast as possible in the situations where quality paths
cannot be found quickly and/or can not efﬁciently work be-
cause of higher rates of node mobility. Moreover, hop-count
is the simplest to calculate and it avoids any computational
burden on the routing protocol. It is obvious from its equa-
tion: Hop CountPe2e
Quality Link Metrics (QLMs) are ﬁrstly introduced in ,
for successful delivery of data packets in static networks. Ef-
ﬁciency of static WMhNs depends upon low routing latency,
minimized routing load, and less end-to-end delay (E2ED).
To achieve proﬁcient performance of a protocol in such net-
works a realistic QLM is needed.
Several QLMs have been proposed, like, Expected Trans-
mission Count (ETX), Expected Transmission Time (ETT)
, Interference and Bandwidth Adjusted ETX (IBETX) ,
Expected Link Performance (ELP) , Minimum Loss (ML)
, Minimum Delay (MD) and Inverse ETX (InvETX) .
The metrics, ETX, ML, MD and InvETX have already been
implemented  with a proactive routing protocol, Optimized
Link State Routing (OLSR)  using Link State routing tech-
nique. While, ETX, IBETX, and ELP are implemented with
Destination Sequenced Distance Vector (DSDV)  based
on distance vector routing algorithm. Distance vector rout-
ing uses the next hop information during exchanging the rout-
ing information, whereas link state information contains the
whole topological information. In this paper, we implement
ETX, ML, MD, ETT and InvETX in distance vector protocol
DSDV. Original ETX is implemented with DSDV.
In this paper, we have implemented ﬁve QLMs; ETX, In-
vETX, ETT, ML and MD with DSDV which is based on dis-
tance vector algorithm. Previouse implementation of QLMs
in routing protocols have not considered routing load for per-
formance measurements. We, in our previous work  have
considered routing load while analyzing the performance of
OLSR which is a link state based proactive routing protocol.
Now, in the same way, we are evaluating the performance of
those link metrics along with ETT in this paper. Moreover,
ML and MD are implemented in OLSR, and ETT is imple-
mented with MR-LQSR, on the other hand, we implement
these metrics along with InvETX in DSDV.
2. QUALITY LINK METRICS
 is the very ﬁrst work launching the idea of quality routing
by proposing ETX. In this section, we discuss ﬁve QLMs,
among all are based on ETX except MD. A detailed study on
ETX-based metrics can be found in .
(1) ETX: Forward and reverse loss rates and link asym-
metries of the links are measured by calculating the loss prob-
abilities of links by broadcasting probe packets. In this ap-
proach, each node is supposed to periodically send out a broad-
cast probe packets only to the neighbors without any retrans-
mission. Nodes track the number of successfully received
probes from each neighbor during a sliding window time;
10 seconds, and include this information in their own probes.
Nodes can calculate reverse loss probability; dr, directly from
the number of probes they receive from a neighbor in the time
window, and they can use the information about themselves
received in the last probe from a neighbor to calculate forward
(2) InvETX: remarkably avoids the computational over-
head and thus achieves least delay . ETX calculates the
inverse of probability of success (product of forward and re-
verse probabilities) but as the names implies, InvETX directly
I nvET XPe2e=X
(3) ETT: of a link as a ”bandwidth-adjusted ETX” is de-
ﬁned in . Authors consider the link bandwidth to obtain
the time spent in transmitting a packet. They start with ETX
and divide by with link bandwidth. Let Sdenote the size of
the packet and Bthe bandwidth (raw data rate) of the link l.
ET Tl=E T Xl×tl(3)
ET Tl=E T Xl×SF
ET Tl=E T Xl×(SF
Forward and reverse loss rates of links is measured in
ETX portion of ETT. These lost rates are calculated through
broadcast prob packet as in . The problem of determining
the bandwidth of each link is more complex. For the measure-
ment of bandwidth, ETT uses the technique of packet pairs,
i.e, after every minute, each node is supposed to send two
back-to-back probe packets to each of its neighbors. First
probe of size of 137 bytes, while the second probe packet
is comparatively heavier and is of 1137 bytes. Upon receiv-
ing these two probes, neighbor measures the time difference
between the reception of the ﬁrst and the second probe and
acknowledges the sender with the difference. To estimate the
bandwidth, sender takes the minimum of 10 consecutive sam-
ples and then divides the size of the second probe packet by
the minimum size sample.
(4) ML: is based on ETX with the aim of selecting the
path with the minimum loss probability. It uses the probabil-
ity of successful transmissions, and not the inverse probabil-
ity, as in ETX. Another difference of ML with ETX is that
ETX ﬁnalizes the end-to-end route with two or three links,
whereas, like InvETX and IBETX, ML also considers the
longer paths. The whole route’s probability is given by the
product of the links’ probabilities instead of the sum of their
inverse probabilities (like ETX). It has the advantage of elim-
inating the routes with high loss rate, 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 .
(5) MD: With MD metric, routing table computation is
based on the total minimum transmission delay. The trans-
mission delay measurements come from a variant of a link
capacity estimation technique, known as Ad-hoc Probe. The
technique takes into account the differences in clock synchro-
nization, thus providing a more reliable measurement. A dis-
advantage is that this metric considers routes which have nodes
sharing a collision domain with many other nodes, and this
tends to degrade the communication on such routes. The Ad-
Hoc Probe algorithm uses packet-pairs to measure the packet
dispersion . The formula used to calculate packet disper-
sion Tfrom the packet pair sample is given by:
T=Trecv2,i −Trecv 1,i (8)
T= (Trecv2,i −Tsend,i −δ)−(Trecv 1,i −Tsend,i −δ)(9)
Where δis clock offset of nodes, Tsend,i is packet sending
time stamped by sender, whereas Tr ecv1,i, and Tr ecv2,i are
receiving time of each packet stamped by receiver node.
3. IMPORTANT ISSUES REGARDING QLMS
Here, we discuss the direct inﬂuences of mathematical design
of QLMs on the performance of routing protocol implement-
ing it and indirect affects on efﬁciency of the underlying net-
work being operated by respective protocol.
(1) Link asymmetry: Link asymmetry can be used by
QLMs to check the loss ratios of a link in both directions;
forward and backward. If the asymmetry of the link is not
determined correctly then route entry is downgraded to uni-
directionality questionable. If a route request is received over
such a link, the node delays forwarding it while it issues a
direct, one-hop unicast route request back to the questionable
neighbor. If acknowledgement is received back to the sender,
then node forwards the original route request and conﬁscates
the blacklist entry, otherwise, request is dropped by node. In
ETX, invETX and ML account link loss ratios, in both direc-
tions of link.
(2) Low computational overhead: For routing metric,
necessary computations should be considered that must not
consume memory, processing capability and the most impor-
tant; battery power . They discuss the case of three widely
used routing link metrics for wireless routing protocols: ETX,
invETX, ETT and ML, as in Fig.1.
(3) Low routing overhead (routing latency and rout-
ing load): Routing overhead is discussed in detail in .
Wireless networks have limited bandwidth Computing a link
metric in such a way that it generate extra routing overhead
reduce packet delivery.
2000 3000 4000 5000 6000 7000 8000 9000 10000
No of nodes
100 200 300 400 500 600 700 800 900 1000
100 200 300 400 500 600 700 800 900 1000
100 200 300 400 500 600 700 800 900 1000
Fig. 1. Simulation Results for Modeled Framework
(4) Trade-offs: Generally, a protocol achieves higher through-
put values at the cost of increased routing latency in the case
of static networks. Whereas, in mobile networks, where link
breakage is frequent cause more routing load to obtain better
throughput from the network . A QLM in a speciﬁc sce-
nario supposed to a suitable trade-off between routing latency
and routing load to achieve optimal performance.
4. SIMULATION SETUP AND PERFORMANCE
EVALUATION OF QLMS
This section provides the details concerning the simulation
environment. We compare the performance of ETX, InvETX,
ML, ETT and MD in NS-2. The window wused for link
probe packets is chosen to be of size 10s. The wireless net-
work consists of 50 nodes randomly placed in an area of 1000m
x 1000m. The 20 source-destination pairs are randomly se-
lected to generate Continuous Bit Rate (CBR) trafﬁc with
a packet of size 640bytes. To examine the performance of
QLMs under different network loads, the trafﬁc rate is varied
from 1 to 10 packets per second. For each packet rate, the sim-
ulations are run for ﬁve different topologies for 900seach and
then their mean is used to plot the results. Wireless networks
suffer from bandwidth and delay. Because of on-demand na-
ture, the reactive protocols are best suited to cope with these
issues for mobile scenario where change in topology is fre-
quent. We are dealing with static networks where proactive
protocols work at their best because of getting the picture of
whole topology and independent of the data generation. Per-
formance of QLMs has been evaluated and then compared
with three performance parameters; throughput, E2ED, and
(1) Throughput: Among selected ﬁve QLMs; ETX (eq.1),
InvETX (eq.2), ETT (eq.6), ML (eq.7) and MD (eq.9), MD is
not considering link asymmetry. While ETT, ETX, and ML
are introducing computational overhead, as depicted in Fig.1.
Forward and reverse probes are sent by ETX, ETT, InvETX
and ML to check link asymmetry periodically. Asymmetric
links and computation overhead can increase drop rate by in-
troducing computational delay and delivering data to unreli-
able path due to lack of correct assumption about link status.
ETX due to computational overhead increases delay, there-
fore, produces lower throughput value. ETT uses extra probes
after every minute by sending two packets of 137 bytes and
1137 bytes to estimate the bandwidth. This not only intro-
duces routing load due to extra probes at routing layer but
also computational overhead is introduced by estimating the
bandwidth from these probes. In higher network loads, net-
work is more sensitive for routing load and delay. Computa-
tional overhead and routing load lead to more drop rate in high
data trafﬁc rates. Lack of link asymmetry in MD produces
more drop rates. ML uses product of drop rates for a link
as well as for a complete path, and introduces computational
overhead (as shown in Fig.1). InvETX produces lowest com-
putational overhead (Fig.1) by taking the product of forward
and reverse delivery rates of a link and selecting the maxi-
mum InvETX value of a path which is sum of individual links
of a path. So, InvETX achieves high throughput, as obvious
from Fig.2. Unlike MD, InvETX calculates link asymmetry
as well as contains low routing load (Fig.4) by avoiding the
use of extra probes after every minute like in ETT.
(2) E2ED: is the time a packet takes to reach the desti-
nation from the source. We have measured it as the mean of
Round Trip Time (RTT) taken by all packets. Computational
delay and longer paths cause latency for end-to-end path cal-
culation. MD estimates the paths based upon one way de-
lay, while ETT selects paths with shorter delay based on the
bandwidth estimation. Therefore in medium and low network
loads, both metrics produce lower delay (Fig.3), while in high
network loads, ETT due to computational overhead and MD
due to lack of measuring the link asymmetry augments E2ED.
InvETX as compared to ETX and ML has lowest computa-
tional overhead thus selecting the paths with low delay, as
shown in Fig.3.
(3) NRL: is the number of routing packets transmitted by
a routing protocol for a single data packet to be delivered suc-
cessfully at destination. ETT and MD both are supposed to
calculate the paths with low latency. In ETT, bandwidth es-
timation is considered to select a path with low latency. MD
produces the lowest value of routing load (Fig.4) because it
generates delivery measurement probes only in forward di-
rection. On the other hand, ETX, InvETX and ML send both
forward and reverse delivery probes to check link asymmetry
thus producing more routing load as compared to MD. The
highest NRL among the selected metrics is produced by ETT,
because it uses forward and reverse delivery probes with a pair
of packets for calculating low E2E path. Each node is sup-
posed to send two back-to-back probe packets to each of its
neighbor after every minute. First probe packet is small and
having the size of 137 bytes, while the second probe packet
is comparatively heavier and of 1137 bytes. Upon receiving
these two probes, neighbor computes the time difference be-
tween the reception of the ﬁrst and the second probes and
sends acknowledgement value back to the sender. For band-
width estimation, sender takes the minimum of 10 consec-
utive samples and then divides the size of the larger probe
packet by the minimum sample value.
5. MODELING ROUTING OVERHEAD BY DSDV
WITH SELECTED QLMS
For computing routes by using QLMs, loss probabilities may
be required. In OLSR, while computing the loss probabilities,
modiﬁed HELLO messages are used. Whereas, DSDV is sup-
posed to send extra small probes for measuring the loss prob-
abilities and this leads to more routing load. As in this work,
we are considering the routing load as a metric to evaluate
the performance; therefore, we ﬁrst discuss the route mainte-
nance operation of DSDV. DSDV periodically exchanges link
state updates with its neighbors to maintain the recent infor-
mation about connectivity in the network. Moreover, routes
are updated thorough trigger updates also. The periodic route
updates are ﬂooded with full dump period, while trigger up-
date ﬂooding takes place through incremental dumps only
when a link is broken in an active route. Although, the trigger
route update operation may appear surplus because of the em-
ployment of link state monitoring periodically, it has certain
advantages. Monitoring the link status periodically leads to
routing loops which are eliminated in trigger route updates us-
ing the latest sequence numbers. To keep updated all nodes in
a network with topological information, each routing protocol
has to exchange routing packets generating routing overhead.
For this overhead the protocol has to pay some cost in the
form of energy consumed per packet. So, we deﬁne this cost
in the equation given below to calculate the routing overhead
in terms of packet cost. First two costs in eq.10 (eq.10a, 10b)
are the same as we have deﬁned in . We deﬁne third part
(eq.10c) for this work. CDS DV
E−metric shows the cost of those
packets which are to be sent for measuring a QLM.
E−total =CDS DV
E−per +CDSD V
E−tri +CDS DV
Expressions for CDSDV
E−per and CDSD V
E−tri are same as in 
and can be further studied from .
C(DSD V )
E−P er =Zτ1
(Perr davg +davg
E−T ri =Zτ2
(1 −Pnlb)nPerr davg +davg
(Perr )i+1 i
E−metric =CET X,I nvE T X,ML
E−QLM , CE T T
E−QLM , CM L
CET X,I nvET X ,ML
E−QLM = (αdf+αdr)×τNL (11)
E−QLM =CET X,Inv ET X,M L
E−QLM +ZτNL αs−probes +αl−probes (12)
E−QLM = 2 ×αdf×τN L (13)
Where, αdfand αdrare the rates of forward and reverse
probe deliveries. τN L is the total network life time or total
simulation time. Similarly, for ETT, αs−probes and αl−probes
are the rates of exchange of small and large probes.
Routing protocols are responsible for ﬁnding efﬁcient route
selection mechanism for reliable communication by select-
ing optimal routes. There are two types of link metrics; non-
quality link metrics and quality link metrics. In this paper, we
have compared and analyzed the performance of ﬁve qual-
ity link metrics which are based on loss probability measure-
ments; ETX, ETT, InvETX, ML and MD. For comparison,
we have selected distance vector routing algorithm protocol;
DSDV. We implemented ML, MD, InvETX and ETT in DSDV
and computed computational burden of loss probability mea-
surements in ETX, InvETX and ML. It is analyzed that ETX
and ML produce more computational burden when compared
with InvETX. MD does not measure the link asymmetry, thus
fails to achieve appreciable throughput for the operating pro-
tocol. InvETX due to low computational overhead and ac-
curate link asymmetry measurement outperforms in DSDV
among ﬁve selected quality link routing metrics.
 Javaid, N. et al., ”Performance Study of ETX Based
Wireless Routing Metrics”, IC4, pp. 1-7, 2009.
 De Couto et al., ”A High-Throughput Path Metric for
Multi-Hop Wireless Routing”, ACM Mobicom, 2003.
 Richard D. et al., ”Routing in Multi- Radio, Multi-Hop
Wireless Mesh Networks”, ACM Mobicom, 2004.
 Javaid, N. et al., ”Interference and Bandwidth Ad-
justed ETX in Wireless Multi-hop Networks”, 53rd
IEEE Globecom Workshop SaCoNAS-II, 2010.
 Usman A. et al., ”An Interference and Link-Quality
Aware Routing Metric for Wireless Mesh Networks,”.
IEEE 68th VTC, 2008.
 D. Passos, et al., ”Mesh network performance measure-
ments”, I2TS, 2006.
 W. Cordeiro et al.,” Providing quality of service for
mesh networks using link delay measurements”, 16th
 N. Javaid, et al. ”Identifying Design Requirements for
Wireless Routing Link Metrics”, 54th IEEE Globecom
 Qayyum, A. et al., Multipoint relaying: An efﬁcient
technique for ﬂooding in mobile wireless networks,
 C. E. Perkins et al., ”Highly dynamic Destination-
Sequenced Distance- Vector routing (DSDV) for mobile
computers,” SIGCOMM CC, pp. 234-244, 1994.
 Moreira, W., et al., ”Using multiple metrics with the op-
timized link state routing protocol for wireless mesh net-
works”, SBRCSD, Maio (2008).
 N. Javaid, et al., ”Modeling Routing Overhead Gen-
erated by Wireless Proactive Routing Protocols”, 54th
IEEE Globecom Workshop SaCoNAS-II, 2010.
 Ph.D. Thesis, Analysis and Design of Link Metrics for
Quality Routing in Wireless Multi-hop Networks, Uni-
versity of Paris-Est, 2010.