Content uploaded by Houda Labiod
Author content
All content in this area was uploaded by Houda Labiod
Content may be subject to copyright.
Proceedings
of
the
9th
International
Conference
on
Mobile
and
Wireless
Communications
Networks,
Cork,
Ireland,
September
19-21,
2007
Performance Analysis
of
Position-Based
Routing Approaches
in
VANETS
Miguel Garcia de la Fuente
Telecom Unit, Robotiker-Tecnalia
Zamudio -Spain
Email: mgarcia@robotiker.es
Abstract-This
article presents aperformance analysis
between two location-based routing protocols: SIFT (SImple
Forwarding over Trajectory), anovel, scalable, spatial-aware,
trajectory-based approach, and DREAM (Distance Routing
Effect Algorithm for Mobility), astable, largely tested position-
based scheme. The study was accomplished under arealistic
urban mobility model for VANETS (Vehicular Ad hoc
NETworks), within ahighly deployed evaluation network of up
to 1000 nodes. Classical ad
hoc
routing schemes
do
not perform
well in VANETS because they were not designed to handle
efficiently mobility handicaps. Position-based techniques perform
better in dynamic scenarios, but in some highly dynamic
scenarios, like VANETS, they do not always perform efficiently.
Trajectory-based protocols perform more efficiently in VANETS
since they
are
spatial-aware. We demonstrate that SIFT
performs better than DREAM in arealistic VANET scenario
concerning delivery ratio, control overhead, delivery delay, and
route length.
Index
Terms-
Vehicular ad hoc networks, routing,
geographical routing, trajectory based forwarding, position
based routing, mobility model, spatial aware.
1.
INTRODUCTION
AVANET (Vehicular Ad hoc NETwork) is aspecial
variety
of
Wireless Ad Hoc Networks, where nodes are
vehicles that move at high speed, and their movements are
constrained to roads layout and traffic rules. VANETS
provide communications among vehicles to carry out different
tasks like cooperative driver-assistance, traffic management or
commercial services. Network topology is strongly linked
to
road topology, since vehicles mobility
is
restricted to roads
layout. Nodes mobility
may
change network topology and
invalidate existing routes. With these special characteristics,
classical routing approaches for generic ad hoc networks
do
not perform efficiently
in
VANETS.
An important number
of
routing protocols have been
developed for wireless ad hoc networks. Depending on the
type
of
information used for routing, they can be classified
into two categories: topology-based [1] and position-based
[2]. Classical ad hoc protocols are topology-based
as
routing
decisions are based on existing links among nodes. OLSR
[14], DSR [13], AODV
[9]
or DSDV
[8]
are some examples
of
proactive and reactive protocols. All they are link table-
978-1-4244-1720-9/07/$25.00 ©2007 IEEE
Houda Labiod
Ecole Nationale Superieure des Telecommunication (ENST)
Paris -France
Email: labiod@enst.fr
driven since nodes build arouting table where they record a
valid path for each possible destination. Paths depend on
existing links, and links are dependent on network topology.
When nodes move, links change, and paths must be
recalculated. In dynamic scenarios, the control overhead
generated to calculate routes can be extremely high, resulting
in low-performance networks [7]. To prevent that, it is
required adifferent forwarding paradigm more suitable for
very dynamic scenarios.
In Position-Based (PB), routing decisions are based
on
the
geographical coordinates
of
nodes. Some examples
of
PB
protocols are GPSR [4], LAR [10], or DREAM [3]. PB
protocols are more efficient
[7]
for dynamic scenarios, like
VANETS. In
PB
methods, each node must create and
maintain updated alocation table, containing the geographical
position
of
its neighbours. To maintain these tables up-to-date,
when nodes move, they send acontrol message with its new
position.
PB
methods reduces appreciably control overhead,
since the information amount necessary to build alocation
table is smaller than to build alink table. Anyhow,
in
PB
protocols, control overhead may also cause low performance
in highly deployed networks.
Routing paths, defined
as
sequence
of
forwarding nodes,
are unstable due to topology changes, while geographical
routes, defined as lines, are quite stable due to the physical
characteristics
of
the service area. Trajectory-Based
Forwarding (TBF)
[5]
exploits this basic observation
proposing a
PB
routing scheme that requires the source node
to encode ageographical line, referred to
as
trajectory, into the
packet header. Since the sequence
of
forwarding nodes
is
not
specified, packets are routed hop-by-hop according to nodes
position with respect to the trajectory. The forwarding
schemes proposed
in
[5],
[6]
and [17] are based on point-to-
point transmissions. Differently from previously proposed
TBF schemes, SIFT [15] is based on broadcast transmissions
and does not require neighbours position knowledge, since the
forwarding decision is shifted from the transmitter to the
receiver. As each node does not need to know anything about
the rest
of
the network nodes, SIFT does not send any kind
of
control message, solving the problem
of
control overhead.
Hence, this routing strategy makes SIFT to become avery
suitable forwarding protocol for VANETS.
Research in VANETS
is
mostly simulation-based due to the
high cost
of
testing and evaluating routing protocol in real
16
Authorized licensed use limited to: Yuan-Ze University. Downloaded on January 12, 2009 at 02:32 from IEEE Xplore. Restrictions apply.
Proceedings
of
the
9th
International
Conference
on
Mobile
and
Wireless
Communications
Networks,
Cork,
Ireland,
September
19-21,
2007
environments. The main difference between classical ad hoc
networks and VANETS is the way nodes move. Thereby,
simulation-based evaluations in VANETS must make use
of
appropriate realistic mobility models.
In this paper we present aperformance comparison study
between SIFT, ascalable, spatial-aware, TBF approach, and
DREAM, astable, largely tested PB routing scheme. The
performance was evaluated within areal-world urban
scenario, deploying up to 1000 nodes that move according to
SSM (Stop Sing Model) [12], arealistic mobility pattern for
VANETS.
The remainder
of
this article is organized as follows. In
section II we describe the operating scheme
of
DREAM and
SIFT. Section
ill
exposes the mobility model used for this
study. The performance comparison results are analysed in
section IV. We conclude this work in section
V.
II. SIFT AND DREAM DESCRIPTON
In this section we give abrief description
of
the protocols
evaluated, SIFT and DREAM.
A. DREAM
DREAM is adirectional, restricted flooding, PB routing
approach.
It
makes use
of
several techniques to reduce control
overhead. DREAM implements 2algorithms: one to
disseminate location information packets, and another one to
disseminate data packets. The first one is based on arestricted
flooding scheme. Each node, periodically, sends location
packets to update the position tables
of
the other nodes. Nodes
restrict this flooding introducing atravel distance threshold,
that is, the maximum distance that alocation packet will
reach. This flood is also restricted by means
of
controlling the
frequency at which nodes send location updates. This
frequency is proportional to nodes mobility rate. The
algorithm used to disseminate data packets is directional
flooding. When node Swants to send adata packet to node D,
it checks its location table to find
D's
position. Based on this
information, Sselects from its neighbours those nodes that are
in the direction
ofD
and forwards the packet to them. Each
of
these nodes, in turn, do the same, forwarding the message to
those nodes in the direction
ofD
until D,
is
reached.
The comparison study carried out in [3] states that DREAM
could find aroute to agiven destination in 80%
of
the times.
End-to-end delay
of
DSR is between 25% and 250% longer
than in DREAM.
B.
SIFT
SIFT is anew scalable, spatial-aware, TBF approach.
Differently from previously proposed TBF schemes, SIFT
uses broadcast instead
of
point-to-point transmissions.
Wireless transmissions are broadcast in nature and allow
reaching possibly all active neighbours at the same time.
Moreover, forwarding decisions are shifted from the
transmitter to the receiver. Each node that receives apacket
takes the decision to forward it or not based only on its own
position, the last transmitter position and the trajectory. This
reduces control overhead down to 0, that is, SIFT sends no
control packet. Once received apacket, each node sets atimer
according to its position with respect to the trajectory and the
last transmitter. The closer to the trajectory and the farther
from the last hop anode is positioned, the shorter the timer is
set.
If
acopy
of
the same packet, forwarded by another node,
is received before the timer expires, the timer is stopped and
the packet is dropped. Otherwise, the packet is transmitted
when the timer expires. Therefore, the node with the shortest
timer will forward the packet. Packets include into the header
the trajectory and the coordinates
of
the last node that
forwarded the packet. Trajectories can be obtained from
digital maps. Since intermediate nodes get all the required
routing information from the packet header, they do not need
to know anything about its neighbours; hence, they exchange
no control packets. This issue is very interesting in highly
dynamic environments.
III.
MOBILITY
MODEL DESCRIPTION
Research in VANETS is mostly simulation-based due to the
high cost
of
testing and evaluating protocol implementations
in real environments. The main difference between classical
ad hoc networks and VANETS is the way nodes move.
Thereby, the simulation-based evaluations
of
routing
protocols for VANETS must make use
of
appropriate realistic
mobility models, in such away that evaluation results are
coherent with the performance that those protocols would
have
if
they were evaluated in areal-world VANET scenario.
The most popular simulation environments for ad hoc
networks are generic discrete-event simulators that were
designed for modelling generic communication networks.
These tools provide also generic mobility models that do not
address the special motion features
of
VANETS.
In these simulation tools, nodes commonly are placed
random and uniformly within the simulation field, and nodes
move according to acertain kind
of
random mobility model.
Random Waypoint, Manhattan Grid, Purse or Reference Point
Group are some examples
of
commonly used, generic
mobility models [11]. In models like Random Waypoint, each
node randomly selects awaypoint in the simulation area and
moves from its current location to the waypoint with arandom
and constant speed. Once anode has reached the waypoint, it
pauses for arandom amount
of
time before selecting anew
waypoint. However, in VANETS, nodes move according to
the following facts: A) nodes are not random and unifonnly
distributed, they are placed according to roads layout. B)
Nodes do not move according to arandom trajectory, nodes
routes are built on roads layout. C) Nodes speed is not random
and constant; speed is variable and depends on traffic
conditions, roads layout and traffic rules. Random movement
patterns have no similarity to the behaviour
of
vehicle
movements in real-world scenarios. This type
of
mobility
models is not appropriate to research in VANETS.
To perform areliable study, we made use
of
asimple
mobility model for VANETS that addresses movement
17
Authorized licensed use limited to: Yuan-Ze University. Downloaded on January 12, 2009 at 02:32 from IEEE Xplore. Restrictions apply.
Proceedings
of
the
9th
International
Conference
on
Mobile
and
Wireless
Communications
Networks,
Cork,
Ireland,
September
19-21,
2007
patterns
of
vehicles on an urban scenario. It is based on SSM
[12]. According to this model, initial and destinations nodes
position are chosen randomly according to the streets layout
of
agiven city map. The simulation field map represents a
simple grid-shaped urban scenario. Streets are placed creating
anon-uniform grid since streets are not equidistant. We
assume that streets have 2lanes, one for each direction, and
that vehicles are not allowed to pass each other. Thus, vehicles
movement depends on the preceding vehicle movement. Some
crossroads has astop sign; when avehicle approaches a
junction with astop sign, it must stop for aperiod
of
time,
which
is
proportional to the number
of
vehicles that are
moving near that junction. This way, the more crowded azone
is, the longer it takes to pass through it. Each node follows the
shortest path lying on the street topology towards its
destination and, once it is reached, nodes start again the
procedure choosing randomly another destination. Since
nodes speed is dependent on traffic conditions, only maximum
speed limit can be established for each vehicle.
IV. PERFORMANCE COMPARISON
This section presents the simulation results carried out with
the purpose
of
analysing the performance
of
DREAM and
SIFT. We studied the effects
of
critical factors such
as:
1)
nodes density and 2) distance between source and destination
nodes. The performance was compared through the following
parameters:
a)
delivery ratio, b) end-to-end delay, c) route
length in terms
of
number
of
hops and
d)
control overhead.
A.
Simulation scenario
The simulation scenario was implemented with Omnet++
Simulation Engine and its Mobility Framework [18]. Network
nodes are distributed within the simulation area according to
the mobility model described in section III. The simulation
area
is
asquare
of
1000m x1000m. The simulation map is a
simple
lOx
lOnon-uniform street grid. All the nodes are
provided with an IEEE 802.11bcommunications interface,
and all they operate with aradio range
of
100m. In all the
simulations, the source and the destination nodes do not move.
The source node sends 1message per second to the
destination node. The simulation lasts 2hours.
B.
Simulation results
1)
Network density
Through the frrst simulations we have studied the impact
of
nodes density on network performance. The source node was
placed at point Po(0,300), according to the Cartesian
coordinate system
of
the simulation area, and the destination
node was fixed at point Pd(200,500). For the rest
of
the nodes
the maximum speed limit was set up at 12m/s. Network
density
is
oscillating, from 100 to 1000 nodes.
Many studies [7], [16] assert that network density is
decisive on routing performance. This way, the higher the
density is, the higher the network connectivity is, and thence,
the more packets reach the destination. Based on this premise,
Figure 1shows that the delivery ratio
of
SIFT improves in
accordance with network density. DREAM delivery ratio also
improves in terms
of
network density; however, in this case,
performance decreases when the number
of
nodes deployed in
the network exceeds 500 devices.
If
network density is too
high, DREAM does not perform well because nodes send too
many control packets that get the channel overloaded and,
thus, many transmission errors occur, decreasing delivery
ratio.
Generally, VANETS protocol performance within realistic
mobility models is lower than the results showed
by
other
study [16] based on non-realistic models. This study affirms
that, under acertain value
of
density, the delivery ratio
of
SIFT and DREAM could reach 100%; however, for the same
value
of
density, and making use
of
arealistic mobility model
for VANETS, the delivery ratio in SIFT does not exceeds
40% and
in
DREAM this parameter does not exceeds 4%,
as
Figure 1shows. This low performance
is
due
to
the fact that,
in VANETS, protocol performance does not only depend on
network density, but also on nodes distribution. Generic
mobility models often assume that nodes are uniformly
distributed. But in VANETS that assumption
is
not correct
since distribution
is
restricted to roads topology. With anon-
uniform allocation, the probability
of
finding alow-
connectivity zone between source and destination nodes is
higher that with auniform distribution; thus, delivery ratio
decreases in arealistic environment. SIFT delivers more
packets than DREAM
in
any case under arealistic scenario
because SIFT
is
a spatial-aware protocol as it gets the routing
trajectories from digital maps, while DREAM uses asimple
greedy technique. Nodes distribution is akey factor; so that,
knowing that nodes are distributed according to streets layout,
arouting protocol aware
of
that layout will be more efficient
as it will be able to avoid low-connectivity regions. On the
other hand, DREAM is not aware
of
road topology and it
happens often that there
is
alow-connectivity zone between
source and destination nodes. Consequently, we obtain low
delivery ratio
in
DREAM, as showed in Figure
1.
Figure 2shows that DREAM
is
not capable
of
delivering
any data packet when the number
of
nodes
is
lower than 400;
the delay is infinite as no packet is delivered. SIFT delay
decreases when density increases; the higher density, the
better positioned is the next hop node with respect to the given
routing trajectory, and therefore, the shorter its timeout lasts.
This way, SIFT delay decreases when density increases. With
ahigh-deployed network, DREAM delivery delay
is
very high
due to the location-information dissemination procedure. This
process generates too much control traffic, which overloads
the channel.
In
an overloaded network, collisions occur, and
packets must be retransmitted. However,
if
density
is
not very
high, DREAM does not overload the channel and gets lower
delay than SIFT, which
is
timer-based. Thus, in low-deployed
but high enough connected networks, DREAM performs
better than SIFT regarding delivery delay.
Figure 3shows route length in terms
of
network density. In
this chart, the number
of
hops
is
represented with
an
infinite
value when packets do not reach the destination. When the
18
Authorized licensed use limited to: Yuan-Ze University. Downloaded on January 12, 2009 at 02:32 from IEEE Xplore. Restrictions apply.
Proceedings
of
the
9th
International
Conference
on
Mobile
and
Wireless
Communications
Networks,
Cork,
Ireland,
September
19-21,
2007
2Ot-------------------------i
~
~ ~
~
___
rof.....
_SIFT
_CREAM
Figure
2. End-to-end delay
as
function
of
density.
2)
Distance
The second simulations group performed in this work
attempted to evaluate the effects
of
the distance between
source and destination nodes, that is, how performance varies
in tenns
of
the distance that apacket must travel towards its
target. For these simulations, 437 nodes were distributed
within the simulation area. The source node was always
placed at point Po(0,300), and the destination node was placed
in several positions, in such away that distance between
source and destination nodes oscillates from 100 to 1100m.
For the rest
of
the nodes the maximum speed limit was set up
at 12m/s getting an average speed
of
4.7m/s.
Figure 5shows that delivery ratio decreases in accordance
with distance; the more distant the source and destination
nodes are, the less data packets are delivered. This result is
due to the fact that the probability
of
finding alow-
connectivity zone in the network is proportional to the number
of
intermediate hops that forward apacket. As showed in
Figure 3, the number
of
intermediate hops
is
always lower in
SIFT than in DREAM; thus, SIFT can deliver more packets
than DREAM.
In Figure 6we can observe that SIFT delivery delay
increases when destination node moves away. This is related
to the fact showed in Figure 5; delay is proportional to the
number
of
intermediate hops, and the number
of
hops is
proportional to the travelled distance towards the destination.
DREAM delivery delay is shorter than in SIFT when distance
is short, up to 400m. Within this limit, DREAM perfonns
better that SIFT because SIFT is timer-based and DREAM is
table-driven. However, when distance exceeds that threshold,
DREAM finds more low-connectivity zones between source
and destination nodes, and packets do not reach the
destination. It is represented in Figure 6with an infinite delay
value. The number
of
low-connectivity zones between source
and destination is the same for both protocols since both they
operate under the same network; however SIFT performs
better that DREAM because it can define an appropriate
routing trajectory to avoid low-connectivity regions.
Figure 7shows that the number
of
intermediate hops that a
data message passes through is increased proportionally to the
distance that the data packet must travel. In this chart, the
number
of
hops is represented with an infinite value when
packets do not reach the destination. In our scenario it occurs
with DREAM when the distance between source and
destination nodes is larger than 400m. SIFT performs better
that DREAM due to the same reasons already explained in
Figure 3.
Figure 8points out the same situation showed in Figure 4
with respect to the control overhead. Similar conclusions can
be
affirmed. With respect to DREAM, in this case, density
remains constant, but the larger the distance is, the more nodes
forward location packets generated by the location
information dissemination procedure
of
DREAM. So that,
control overhead is also proportional to distance between
source and destination nodes.
1000900
900 1000
__
DREAM
.......
SIFT
_~
-..-SIFT
800
800
700
500
800
nu
......
ofnocle.
300
300
200
200
o
f====~===~=====~===~==~!::::::===~===~==~==~
100
Overh
....
soooo
-..
-0-'-"'--'--'---"'-'"
-_
..•...
--
..
--
..
-.-
..
-.---00
-----
-00.
00
_,,_
._
__
_
".
_
_.
"
__
'
OOj
~
~l
11
40000
A"'" !
i:
~
i
~
2SXlO
.-A'
!
}20000
~~
I~::
_____
SJOO
-~
o
a---....-O=:~=:;;;;,...*----.----..----,Ir----+-----.--~~----..
100
Figure
4.
Control
overhead
as
function
of
density.
RoI*
IAngttI
1040
r-------
..
-.-----t--------------.-.--.----.------------.-.------.-
120t-------~f--------------__:"e/:_....j
/'
;.100
t-----------"I-----------~---::P------l
i
8O-t-------~-------~-~L..-------j
180+-------~\--_--------,,~--------l
c
040t---------..I.:::
...
=---...-=:::::::::~----------___l
Figure
1. Delivery
ratio
as
function
of
density.
Figure
3.
Route
length
as
function
of
density.
311+-----------------=---..,......-=:
__
......,::::..-1-3,5
t:
:51
I:
~I
'lit
iW
1J
5
~I
o0
0.-1OD~-.....-------..-::V----!OO--
..
----~-
..
-----41~
.....
rof.....
_81FT
_DReAM
DellnryD.lay
O,lDIIOCI
,.----.---00.-
00
--.--.----.--
-.----
-------.-------.---.00-.-
...
--.------
...
--
--------.-
0.7
1,01_
\.
0.0'"
\.
//
0.'
i
,~
, /
uj
i1,01200
'lL.
./
0.4~
1
0.01D01
0'-
~
ul
i=
~
-----
ul
/'
number
of
nodes is lower than 400, DREAM can deliver no
packet. SIFT chooses, at each intermediate step, the node that
is more distant from the last node that forwarded the packet.
Thus, in SIFT, the number
of
hops depends on transition radio
range, but not on density. So that, SIFT delivery delay is
constant, lightly decreased, when density increases because
the more nodes there are in the network, the better positioned
they are regarding the given trajectory, and the shorter the
intermediate timeouts are. However, DREAM hops number
depends on density since all the nodes that are within the
routing cone forward the given packet; hence, the number
of
intermediate hops is proportional to the number
of
total nodes
in the network.
Figure 4represents how control overhead changes in tenns
of
network density. SIFT exchanges no control messages;
therefore, control overhead is always
o.
However, DREAM
control overhead is proportional to density: the more nodes,
the more control packets are exchanged.
DeIlveryRlltio
19
Authorized licensed use limited to: Yuan-Ze University. Downloaded on January 12, 2009 at 02:32 from IEEE Xplore. Restrictions apply.
Proceedings
of
the
9th
International
Conference
on
Mobile
and
Wireless
Communications
Networks,
Cork,
Ireland,
September
19-21,
2007
Overhead
Delivery Delay
not uniform but constrained to roads layout. This distribution
generates low-connectivity zones within the network.
An
efficient routing protocol for VANETS must be able to avoid
routing packets through those regions. TBF approaches seem
to be an efficient solution to handle low-connectivity zones
drawbacks, as trajectories provide spatial-awareness. Classical
routing strategies like simple greedy, commonly used by
location-based protocols,
do
not perform properly since they
are not spatial-aware. In our future work, we will analyse the
impact
of
vehicles speed in TBF and PB protocols, comparing
them to dissemination protocol specifically designed for
VANETS environments.
REFERENCES
[1]
E.
Royer and
C.
Tob,
"A
Review
of
Current Routing Protocols for Ad
Hoc Mobile Wireless Networks", in IEEE Personal Communications,
April 1999.
[2]
M.
Mauve,
1.
Widmer and
H.
Hartenstein, "A Survey on Position-Based
Routing, in Mobile Ad-Hoc Networks", in IEEE Network Magazine,
Nov. 2001.
[3]
S.
Basagni, I. Chlamtac,
V.
Syrotiuk, and
B.
Woodward, "A Distance
Routing Effect Algorithm for Mobility (DREAM)", in ACM/IEEE
MOBICOM,1998.
[4]
B.
Karp and H.T. Kung, "GPSR: Greedy Perimeter Stateless Routing for
Wireless Networks", in ACM-IEEE MOBICOM, 2000.
[5]
D.
Niculescu and
B.
Nath, "Trajectory Based and Its Application", in
proc.
Of
ACM Mobicom '03, 2003.
[6]
B.Nath and
D.
Niculescu, "Routing on acurve", in ACM SIGCOMM
Computer Communication Review, vol. 33,
no.
1,
January 2003.
[7]
H.
FuBler,
M.
Kasemann, and
D.
Vollmer,
"A
Comparison
of
Strategies
for Vehicular Ad-Hoc Networks", in Dept.
of
Compo
Sc., Univ.
of
Mannheim, Tech. Rep. TR-3-2002, 2002.
[8]
C.
E.
Perkins and
P.
Bhagwat, "Highly Dynamic Destination-Sequenced
Distance-Vector Routing (DSDV) for Mobile Computers", in
SIGCOMM: Computer Communications Review, 24(4), 234-244,
October 1994.
[9]
C.
E.
Perkins and
E.
M.
Royer, "Ad Hoc On-Demand Distance Vector
Protocol", in
C.
E.
PERKINS (Ed), Ad Hoc Networking,
pp.
173-219.
Addison-Wesley, 2000.
[10]
Y.
Ko, and
N.
H.
Vaidya, "Location-Aided Routing (LAR) in Mobile Ad
Hoc Networks", in MOBICOM, 1998.
[11]
T.
Camp,
1.
Boleng, and
V.
Davies.
"A
survey
of
mobility models for ad
hoc network research" in Wireless Communications &Mobile
Computing, 2(5):483--502, 2002.
[12]
A.
Mahajan,
N.
Potnis,
K.
Gopalan, and
A.
Wang, "Evaluation
of
mobility models for vehicular ad-hoc network simulations", Technical
Report TR-051220 Florida State University, 2005.
[13]
D.
B.
Johnson and
D.
A.
Maltz, "Dynamic Source Routing in Ad-Hoc
Wireless Networks", in Mobile Computing, 1996.
[14]
T.
Clausen and
P.
Jacquet. "Optimized Link State Routing Protocol
(OLSR)",
in.
RFC 3626, IETF Network Working Group, October 2003.
[15]
A.
Capone,
M.
Garcia de la Fuente,
L.
Pizziniaco and I. Filippini, "SIFT:
An efficient method for Trajectory Based Forwarding", in IEEE
International Symposium on Wireless Communications Systems
(ISWCS'05), September 2005.
[16]
M.
Garcia
de
la Fuente and
H.
Labiod,
"A
Perfonnance Comparison
of
Position-Based Routing Approaches for Mobile Ad Hoc Networks",
submitted to IEEE Vehicular Technology Conference VTC-Fall, 2007.
[17]
M.
Yuksel, RPradhan and
S.
Kalyanaraman,
"An
Implementation
Framework for Trajectory-Based Forwarding in Ad-Hoc Networks", in
IEEE ICC'04, June 2004.
[18] OMNET++ Community Site -http://www.omnetpp.orgl
1000
_CREAM_SIFT
_CREAM
_SFT
!lOO
......
_lHItw
...
_
....
dII
..................
!
~
./
I
~
i
----..-----
i
~
i
....- !
o
.\---.......-----.---....---..-----.---...---.....,........----.--~
200
..
~~:--:~::---------------------J.
~1
0.018+-------I------------~-_--
-_---i
.::::~========~==================:~/~::======:=::::..,
fO,012
~
i
I~:
~+-
-==-
-==--,
-:==
-:=_
+~l../=~
-::;_~=_
"'"==_"-==_
-==_
-:=_-==_
-==_
-==-
-:=_
-==_
-==_-==_
-==_-==_
~~
o'OlM~.---r-=----__+------------------;
0.002_.._--....-
........
----------------
o
+----..------.----.--_-
__
--.....--.----....._--i
200
Rou1ItLength
60
--·------··--r-·---------··---·--··---··--······
-'.--'"
-_
..
---
---.-.--_
-
--.
__
_._.
-
..
-------
'j
140 II
~~
)
120
/
~
~
10~
Figure 6. End-to-end delay as function
of
distance.
Figure
7.
Route length as function
of
distance.
o
200
'-
~o
°2llO~--.---..;::~-_
.......-
---.7lllI-~t----------l
..
0IllI-~110D
d1
__
IHI
n_.nddll.-na1lonnod.s
_SIFT_CREAM
Figure 8.
Control
overhead as function
of
distance.
V. CONCLUSION AND FUTURE WORK
Ad hoc networks literature highlights that control overhead
is the most critical issue that must be faced. Classical ad hoc
routing schemes
do
not perform well
in
VANETS because
they need to send too much control packets, getting the
channel overloaded.
PB methods reduce appreciably control overhead since they
use more efficient routing techniques, like restricted or
directional flooding. However, simulations show that,
in
VANETS, those techniques may not be efficient enough. SIFT
seems to be more efficient for highly dynamic environments
because it solves the control overhead problem. SIFT,
performs better than simple location-based protocols
concerning delivery ratio, delivery delay, control overhead
and route length.
The other main difficulty than routing protocols for
VANETS should address is the fact that nodes distribution
is
45.-.--
..•...•
-.--.-
....
_
..
_.
.
__
..
__
..
_
..
__
..
__
~_~.-
•
Figure 5. Delivery ratio as function
of
distance.
20
Authorized licensed use limited to: Yuan-Ze University. Downloaded on January 12, 2009 at 02:32 from IEEE Xplore. Restrictions apply.
