Content uploaded by Antonio Di Maio
Author content
All content in this area was uploaded by Antonio Di Maio on Nov 21, 2019
Content may be subject to copyright.
A Centralized Approach for Setting
Floating Content Parameters in VANETs
Antonio Di Maio, Ridha Soua, Maria Rita Palattella, Thomas Engel
University of Luxembourg
Luxembourg
Email: {name.surname}@uni.lu
Gianluca A. Rizzo
HES-SO Valais
Switzerland
Email: gianluca.rizzo@hevs.ch
Abstract—Floating Content (FC) has recently been proposed as
an attractive application for mobile networks, such as VANETs,
to operate opportunistic and distributed content sharing over a
given geographic area, namely Anchor Zone (AZ). FC perfor-
mances are tightly dependent on the AZ size, which in literature
is classically chosen by the node that generates the floating
message. In the present work, we propose a method to improve
FC performances by optimizing the AZ size with the support
of a Software Defined Network (SDN) controller, which collects
mobility information, such as speed and position, of the vehicles
in its coverage range.
I. INTRODUCTION
One of the main technical challenges of content dissemina-
tion in VANETs is related to the high dynamism of vehicular
topology and the volatility of inter-vehicular links, either
Vehicle-to-Vehicle (V2V) or Vehicle-to-Infrastructure (V2I)
communications, in which infrastructure typically comes in
the form of Road Side Units (RSUs). Such volatility hampers
the efficient spreading of the content. Floating Content (FC),
a push-based communication scheme [1], was proposed to
overcome the rapid vanishing of the disseminated content by a
vehicle in a specific area, when no support from infrastructure
is available. The goal of FC is to replicate a content in all the
nodes in a limited spatial region called Anchor Zone (AZ),
associated to that content. In this way, the content is stored
probabilistically in the AZ for the content lifetime, or until
its disappearance from the AZ. During such time, the content
is made available to nodes traversing the AZ by means of
opportunistic replication. More specifically, users that traverse
that AZ and do not possess the associated content, can have
a copy of it when they are within the transmission range of
any vehicle owning a copy of the original content. Hence, the
content ends up being available on a set of nodes within the
AZ and ”float” over time even after that the originating node
has left the AZ. When a vehicle exits the AZ and has a copy of
the associated content, it considers it as obsolete and deletes
it.
From a performance viewpoint, the main challenge in
designing the AZ size and shape, and in orchestrating content
replications, is to strike a balance between resources utilization
(e.g. storage and bandwidth) and utility of storing and repli-
cating the content. In general, this is performed by allocating
enough resources to achieve a minimum target value of an
application level performance parameter, such as the average
time to get content, or the percentage of nodes leaving the AZ
with a copy of the message (success probability) [2].
Existing works on FC focus on modeling content availability
(the percentage of nodes with content inside the AZ) and
success probability in function of various mobility metrics
such as node density, average speed, mean rate of node
encounters and arrival rate in the AZ [1], [2], [3]. These studies
assume that the generating node has perfect knowledge of
those parameters, without considering how such knowledge is
built. Of course, these parameters are essential for any practical
application of FC, as they strongly determine shape and size
of the AZ in function of the target performance. Decentralized
estimation of various features of mobility patterns for VANETs
has been already considered in literature [4], [5]. However, in
realistic settings, typically characterized by a high spatial vari-
ability in such parameters as mean node density, mean speed,
etc., the performance of decentralized estimation approaches
may suffer from low accuracy.
In general, all techniques for estimation of such parameters
suffer from some degree of uncertainty, which may lead to
conservative and inefficient AZ dimensioning, or to poor FC
performance. Conservative FC dimensioning has a cost on
the system in terms of storage and bandwidth resources,
but also mechanisms for mobility estimation consume system
resources. Pursuing a high degree of estimation accuracy might
put a higher burden on the system than a more conservative
dimensioning of FC parameters based on coarser estimations.
Such tradeoff depends on specific features of the mobility
model, on techniques used to estimate some parameters related
to it, and to node density distribution.
The ultimate goal of our work is to determine an optimal
mobility estimation method that, for a target value of success
probability, maximizes FC efficiency by striking a balance
between centralized and distributed approach.
In this paper, we take the first steps at tackling this issue: We
focus on FC in VANETs and we analyze how FC performance
and resource efficiency may benefit from the availability of
a centralized entity, which interacts with moving vehicles,
collects relevant mobility data, and build estimates of those
mobility parameters for optimal FC dimensioning. In partic-
ular, we consider an SDN-based architecture for VANETs
as proposed in [6]. RSUs are SDN-enabled controllers that
cooperate with vehicles and improve the management of
2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC)
978-1-5090-6196-9/17/$31.00 ©2017 IEEE 712
content delivery and dissemination by setting properly the size
of the AZ.
II. NE TWORK SCENARIO
As illustrated in Fig. 1, we consider a set of vehicles moving
in a geographical area, the Service Area (SA), containing a
certain number of RSUs. At any time, each vehicle may start
disseminating a certain content in a given AZ, contained in
the SA. In particular, we are interested in those setups where
the density and the coverage range of RSUs do not allow
a complete coverage of the SA. Vehicles communicate via
V2V between them and via V2I with the RSU, using in both
cases the IEEE 802.11p standard, a Medium Access Control
(MAC) protocol based on the DSRC spectrum at physical layer
[7]. We assume the AZ to be circular, though the analysis
can be extended to more complex shapes. When a vehicle
wants to create an AZ and start disseminating a content in
it, it must first determine the AZ radius and the content
lifetime, i.e. the time after which vehicles are allowed to
discard the content. AZ radius is determined in function of
target values of performance estimators such as availability
(the mean fraction of users in the AZ with a copy of the
floating content) and success rate (the mean fraction of users
getting out of the AZ with a copy of the floating content),
as defined in [2]. These estimators are strongly correlated to
nodes density, to their transmission range and their speed. In
addition to such performance parameters, we consider also
the FC efficiency, which takes into account also bandwidth
utilization. It is defined as the mean number of transmissions
per useful content replication (an useful content replication is
a transfer of content within the AZ, from a node with content
to a node without it), for a given mean success rate. In an ideal
setting where nodes are assumed to have perfect knowledge of
mobility, and in which such knowledge comes at no cost, FC
efficiency depends on node mobility patterns, on node density,
and on the ratio between the AZ and vehicle range radii. In
realistic settings, the message exchanges that implement the
specific mechanism for mobility estimation have to be taken
into account too.
In IEEE 802.11p VANETs, periodic beaconing is crucial
to enable network synchronization and packets transmission
orchestration. Vehicles access the channel to broadcast their
beacons at least once in the Control Channel Interval. In our
setting, we assume these beacons broadcast (to other vehicles
and to RSUs) information about the originating vehicle’s ID,
position, speed and direction, as well as a global timestamp.
The RSUs store these parameters in a Mobility Information
Table (MIT). An entry in the table is updated when the RSU
receives a beacon from the corresponding vehicle. Besides,
an entry in the MIT is deleted if it is not updated during a
specific period of time. If a vehicle wants to start floating a
specific message, and it is in the RSU range, it sends a request
to the RSU (possibly in the form of a Wave Service Message
(WSM) [7]) for the estimated measures of average speed and
density. The RSU sends back the requested data via WSM.
When the vehicle is outside the RSU range, it estimates the
TABLE I
AZ Model parameters
Parameter Definition
viID of the generic i-th vehicle in the SA
v∗ID of the seeder vehicle in the SA
t(vi)Timestamp of the mobility information sent by vi
p(vi)Position of the vehicle viat time t(vi)
s(vi)Speed of the vehicle viat time t(vi)
R(S, D)Radius of the AZ started by the vehicle v∗
RSU(vi)RSU ID to which viis connected
C(vi)Surface of the virange
C(RSU(vi)) Surface of the RSU(vi)range
B(p(v∗), δ)Closed disk centered in p(v∗), with radius δ
V(α)Set of vehicles (IDs) in a neighborhood area α
VMI T (RSU(vi)) Set of all the vehicles (IDs) in the MIT of RSU(vi)
VC(v∗)VMI T (RSU(v∗))
Vδ(p(v∗), δ)VC(v∗)∩V(B(p(v∗), δ))
VD(v∗)V(C(v∗))
DDensity of vehicles in v∗neighborhood
Dc(V(α)) Cardinality-based density
Dd(V(α), p(v∗)) Distance-based density
SAverage speed of vehicles in v∗neighborhood
AZ radius using a collaborative and distributed approach, as
classically proposed in literature.
III. MOBILITY-BASED AZ SHAPI NG M OD EL
In this section, we detail the proposed method to compute
the AZ radius (R), used by the vehicle that wants to start
disseminating the associated FC (seeder vehicle). The AZ
radius will be function of node density and average speed
in a neighborhood of the seeder. When the average speed
of the vehicles close to the seeder is high, it is expected
that the FC success rate drops due to the reduced time the
vehicles have to exchange information. In order to increase
the number of vehicles that get the content, it is possible
to enlarge the AZ radius, which will imply a higher success
rate. Conversely, when the density of vehicles is high, it is
likely that the message will reside in several vehicles which
tend to form small clusters and therefore keep floating for
more time. In this case, it is also important to optimize other
practical parameters aside the success rate, such as minimizing
message redundancy and memory occupation, by choosing
a smaller AZ radius. Table I summarizes the notation used
hereafter for describing the mobility-based AZ shaping model.
The relationship between AZ radius, average speed and vehicle
density [2], [3] is generally in the form of:
R(S, D) = wS
D(1)
where Sis average speed in a neighborhood of v∗,Dis
vehicle density in the same neighborhood, and wis a weight
which depends on the specific mobility model, to be either
empirically fixed or defined using some dynamic policies.
The radius is always computed locally by each seeder vehicle
v∗, with Sand Dobtained either from an RSU, if the
seeder is in the RSU range, or through estimations performed
collaboratively by vehicles. In both cases, the WSMs messages
2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC)
713
Fig. 1: An example scenario, showing ranges, neighborhoods and AZs in a generic SA. The thick lines represent the neighborhood boundaries
for each approach: C-based (orange), δ-based (purple) and distributed (yellow).
sent by each vehicle contain information about its current
position, speed, and timestamp:
p(vi)=(px(vi), py(vi)) ∈R2(2)
s(vi) = (sx(vi), sy(vi)) ∈R2(3)
t(vi)∈R+(4)
A. Centralized approach
When a seeder vehicle is in the range of an RSU, it can rely
on the mobility parameters estimated by the SDN controller.
Each RSU can build its own MIT, whose generic structure is
shown in Fig.1. The MIT is organized like a database table
in which the primary key is the vifield and therefore no
duplicates are allowed.
Considering that the RSU memory is finite, an update policy
must be implemented to keep in the MIT only the most up-
to-date information about the mobility of each vehicle. For
instance, an expiration time can be set so that entries older
than the specified time will be discarded. In the centralized
approach, a request-response mechanism is devised: the seeder
vehicle v∗must choose which kind of neighborhood to use
(C-based or δ-based, described in the following sections) and
send to RSU(v∗)the suitable information to determine the
values of D,Sand the set of vehicles that belong to the
selected neighborhood (neighbor vehicle set). The RSU(v∗)
will compute S(Eq.12) and D(Eq.11 or Eq.10) and send their
values back to the vehicle for determining the AZ radius.
1) C-based Neighborhood: In the simpler case, the whole
surface of the RSU range, C(RSU(v∗)), can be considered
as v∗neighborhood. Given that all and only the vehicles that
send their mobility information to the RSU will have a valid
entry in the MIT, they will also be the only ones included in
the neighbor vehicle set VC(v∗). Therefore:
VC(v∗) = VMIT (RSU(v∗)) ⊆V(C(RSU(v∗))) (5)
2) δ-based Neighborhood: If the range of the RSU is too
large, it is possible to define a custom-shaped neighborhood
around the seeder vehicle, exploiting the mathematical def-
inition of neighborhood. A two-dimensional closed ball (or
closed disk) [8] in a metric space (R2, d), centered in p(v∗)
and with radius δ, is defined as:
B(p(v∗), δ) = x∈R2:d(p(v∗), x)≤δ(6)
To physically shape the neighborhood as a circle, we assume as
distance function dthe euclidean (2-norm) distance in R2. Any
other p-norm distance could be used as well as the Chebyshev
distance (∞-norm) or a generic distance function to shape the
neighborhood in a custom way.
d(x, y) = kx−yk2(7)
Let V(B(p(v∗), δ)) be the set of vehicles in the closed disk
of radius δand centered on the position of the seeder p(v∗).
We define the δ-based neighbor vehicle set Vδ(p(v∗), δ)as:
Vδ(p(v∗), δ) = VC(v∗)∩V(B(p(v∗), δ)) (8)
The value of δshould be properly chosen. For instance,
δcan be heuristically set by the RSU for all the vehicles in
its range. In this case, δcould be such that the area of any
B(p(vi), δ)is a proportional fraction of the total extension of
C(RSUi). It is also possible to let the seeder vehicle determine
which is the most suitable value of δ, accordingly to its needs.
In this case, the vehicle must communicate the desired value
of δto the RSU, along with the value of p(vi). The seeder
2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC)
714
vehicle v∗might initially use a standard value (e.g. equal to
the expected or desired AZ radius) and subsequently adjust it
using information provided by the RSU. A special case may
consist in setting δequal to the radius of seeder range.
B. Distributed approach
It may happen that the seeder vehicle is outside the range
of any RSU in the SA, therefore it does not have any
infrastructure support for determining the neighbor vehicle set
VD(v∗). In this case, it can simply consider its own range as
neighborhood, resulting in:
VD(v∗) = V(C(v∗)) (9)
The seeder collects the mobility information by asking all the
vehicles in its range for their position and speed, in a dis-
tributed fashion. After having gathered the needed information,
the density and average speed values will be computed locally
by the seeder and not provided by the RSU as in the centralized
approach.
C. Density Function
We propose two different methods for computing the vehi-
cles density Din a given v∗neighborhood. The methods can
be applied in both approaches (centralized and distributed),
choosing the appropriate neighbor vehicle set Vamong VC,
Vδ, and VD.
The density can be computed as a function of the distance
between the seeder v∗and every other vehicle in the selected
neighbor vehicle set, as in
Dd(V, p(v∗)) = X
vi∈V
vi6=v∗
kp(v∗)−p(vi)k−1(10)
Note that Eq.10 returns a density score that combines informa-
tion about number of vehicles in the neighbor vehicle set and
information on how physically close they are to the seeder.
An alternative way for estimating the density consists in
considering the cardinality of the neighbor vehicle set V,
normalized by the area of the v∗neighborhood.
Dc(V, A) = |V|
A(11)
This definition of density does not necessarily require to
know the position of the seeder and it is strictly dependent
on the area of the chosen neighborhood. Differently from
Dd(V, p(v∗)), it is not continuous but discrete with a minimum
step of 1/A. We might prefer using Eq.11 instead of Eq.10
when the information about the neighborhood area is available.
Note that A is equal to the area of C(RSU(v∗)),
C(B(p(v∗), δ)), or C(v∗), when adopting, respectively, the
centralized C-based, centralized δ-based, or distributed ap-
proach.
D. Average Speed Function
We define the average speed S as the arithmetic mean of
the speed vector modules of the vehicles in a neighbor vehicle
set V.
S(V) = 1
|V|X
vi∈V
ks(vi)k2(12)
Similarly to the density, the same formula of the average
speed can be used in both centralized and distributed scenarios,
choosing the appropriate neighbor vehicle set of vehicles V.
IV. FUTURE WORK
The increasing availability of a large amount of informa-
tion in Vehicular Ad-hoc Networks is giving rise to interest
on Floating Content (FC). FC supports infrastructure-less
distributed content sharing in an Anchor Zone (AZ). The
performance of FC paradigm in function of specific mobility
patterns and for specific applications has been at the center of
recent investigations. However, the issue of how to estimate
the main mobility features, on which AZ dimensioning is
based, has not been considered so far. More specifically, the
issue of how to maximize FC efficiency by striking a balance
between estimation accuracy and conservative dimensioning of
FC service is still unsolved. In this paper, we take the first steps
at tackling this issue by proposing AZ dimensioning based on
centralized (SDN-based) estimation, as well as a dimensioning
based on information built collaboratively by vehicles through
a distributed approach. The next step of the investigation will
encompass the validation of the preliminary hypotheses and
the derivation of an overall resource consumption model, for
each of the proposed approaches. This work will lay the
foundation to determine a resource-optimal strategy for FC
dimensioning in VANETs.
REFERENCES
[1] J. Kangasharju, J. Ott, and O. Karkulahti, “Floating content: Information
availability in urban environments,” in Pervasive Computing and Commu-
nications Workshops (PERCOM Workshops), 2010 8th IEEE International
Conference on, March 2010, pp. 804–808.
[2] S. Ali, G. Rizzo, V. Mancuso, and M. A. Marsan, “Persistence and
availability of floating content in a campus environment,” in 2015 IEEE
Conference on Computer Communications (INFOCOM), April 2015, pp.
2326–2334.
[3] E. Hyyti, J. Virtamo, P. Lassila, J. Kangasharju, and J. Ott, “When
does content float? characterizing availability of anchored information in
opportunistic content sharing,” in INFOCOM, 2011 Proceedings IEEE,
April 2011, pp. 3137–3145.
[4] G. V. Rossi, K. K. Leung, and A. Gkelias, “Density-based optimal
transmission for throughput enhancement in vehicular ad-hoc networks,”
in 2015 IEEE International Conference on Communications (ICC), June
2015, pp. 6571–6576.
[5] S. Zemouri, S. Djahel, and J. Murphy, “A short-term vehicular density
prediction scheme for enhanced beaconing control,” in 2015 IEEE Global
Communications Conference (GLOBECOM), Dec 2015, pp. 1–7.
[6] I. Ku, Y. Lu, M. Gerla, R. Gomes, F. Ongaro, and E. Cerqueira, “Towards
software-defined VANET: Architecture and service,” Conference: Annual
Mediterranean Ad Hoc Networking Workshop MEDHOCNET, 2014.
[7] C. Campolo, A. Molinaro, and R. Scopigno, Vehicular ad hoc Networks.
Standards, Solutions, and Research. Springer, 06/2015 2015, editors.
[Online]. Available: http://bit.ly/1Fm4TEt
[8] M. O Searcoid, Metric Spaces. London: Springer London, 2007, pp.
71–82.
2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC)
715
