Vous êtes sur la page 1sur 8

Simulation and Performance Evaluation of DAG

Construction with RPL


Olfa Gaddour

, Anis Koub aa

, Shaque Chaudhry

, Miled Tezeghdanti

, Rihab Chaari

, Mohamed Abid

CES Research Unit, National School of Engineers of Sfax (University of Sfax), Tunisia.

COINS Research Group, Al-Imam Mohamed bin Saud University (CCIS-IMAMU), Saudi Arabia

CISTER Research Unit, Polytechnic Institute of Porto (ISEP/IPP), Portugal.


Emails: olfa.gaddour@enis.rnu.tn, aska@isep.ipp.pt, shaque@coins.csrlab.org, miled@softmote.com,
rihab.chaari@ceslab.org, mohamed.abid@enis.rnu.tn
AbstractIn this paper, we simulate and analyze the perfor-
mance of the network formation process with the RPL (IPv6
Routing Protocol for Low Power and Lossy Networks), which
is a routing protocol specically designed for Low power and
Lossy Networks (LLN) compliant with 6LoWPAN (IPv6 Low
power Wireless Personal Area Networks). One motivation behind
this work is that RPL is the rst prospective candidate routing
protocol for low-power and lossy networks, which are a main
component in the next generation Internet-of-Things. RPL is
still under development, although having gained maturity, and
is open to improvements. Indeed, there is a need to understand
well its behavior, and investigate its relevance. Our analysis is
based on the ContikiRPL accurate and realistic simulation model
developed under Contiki operating system. The performance of
RPL is evaluated and analyzed for different network settings
to understand the impact of the protocol attributes on the
network formation performance, namely in terms energy, storage
overhead, communication overhead, network convergence time
and the maximum hop count. We argue through simulation that
RPL provides several features that make it suitable to large scale
networks.
Index TermsRouting, Low power and lossy networks (LLN),
IPv6 networks, Wireless Sensor Networks, Performance evalua-
tion, Simulation
I. INTRODUCTION
In the last few years, there has been an increasing trend to-
wards enabling the concepts of Internet-of-Things and Cyber-
Physical Systems, where thousands of tiny devices interacting
with their environments are inter-networked together and ac-
cessible through the Internet. To that end, several standard
protocols have been dened starting with the famous IEEE
802.15.4 Physical and Medium Access Control (MAC) sub-
layer protocol, then ZigBee as an underlying Network and
Application Layers protocol, and recently the 6LoWPAN
Network Layer protocol. 6LoWPAN represents a milestone
protocol that bridges the gap between low-power devices
and the IP world. It is an IP-based technology for Low-
Power Wireless Personal Area Networks (LoWPANs), such
as Wireless Sensor Networks, which combines IEEE 802.15.4
[2] and the IPv6 protocols [3]. This integration provides new
dimensions in the design of LoWPANs as it allows for a full
interoperability with the Internet.
Since the release of 6LoWPAN, routing has been considered
as a main research and development challenge. In fact, several
endeavors for specifying an efcient routing protocol for
6LowPAN-compliant low-power and lossy networks (LLNs)
have been driven, such as Hydro [11], Hilow [9], and Dymo-
low [8]. Recently, the IETF ROLL working group [3] came
up with the IPv6 Routing Protocol for Low power and Lossy
Networks (RPL) routing protocol in an attempt to standardize
the routing process for LLNs. The inherent LLNs of low data
rates, high probability of node and link failures, and scare
energy resources, have turned the design of RPL challenging
and different from previous routing proposals.
In the literature, several research works have evaluated the
performance of the RPL protocol from different perspectives
[4], [18], [19], [6], [5], [13]. Their goals were roughly to
understand and analyze the behavior of certain mechanisms
specied in the IETF draft and to propose some enhancements
to potential shortcomings and open issues. In this paper, we
present a new simulation study that focuses on the network
construction process. Our main objective is to investigate the
impact of various network parameters and attributes, including
number of nodes (i.e. network size), number of network roots
(i.e. DAG root), the objective function, and the packet loss
on the performance of the network construction process. Our
simulation study differs from previous efforts in two distinct
ways: (i.) in contrast to previous works based on NS-2 and
OMNET++ simulation models, we evaluate the performance
of RPL with COOJA/Contiki simulator, which provides an
accurate and a more realistic simulation model of RPL. In fact,
COOJA simulates the real code implemented on real motes.
In addition, the ContikiRPL code is open source and thus it is
easy to control the network parameters, (ii.) Our objective is
to evaluate the performance of the RPL network construction
process, which is different from the objectives of previous
studies, as described in the related works Section.
II. RELATED WORKS.
Several works have proposed simulation and experimental
models for the sake of experimenting with RPL and evaluating
its performance. In [12], the authors presented a performance
evaluation study of RPL through an OMNET++ simulation
model. They used ETX as a default link metric to build the
DAG. Simulation results showed the efciency of the trickle
timer in controlling the packet overhead and stabilizing the
network. In addition, the paper demonstrated the signicant
effect of the global repair duration on the number of control
packet overhead. However, the simulations were performed for
a small-scale network, preventing the generalization of results
for large-scale networks.
In [19], the authors compared between the P2P routing
based on RPL and the simple shortest path routing algorithm.
They demonstrated via NS-2 simulations of a large-scale net-
work that the RPL P2P routes are signicantly sub-optimal as
compared to the minimum cost (shortest) routes, mainly when
the source and destination are close to each other. However,
the authors did not propose a solution to this problem.
In [13], the authors proposed a performance evaluation
of the RPL protocol in the context of smart grid appli-
cations. Using OMNET++ simulation model of RPL, they
demonstrated the capability of trickle timer in RPL to bound
the control overhead and reduce communication latency. In
addition, they demonstrated that RPL quickly performs local
repair of link outage and it provides a path quality close to
an optimized shortest path for an outdoor environment. This
result is inconsistent with that found in [19], which open ways
for further investigation about path quality in RPL.
ContikiRPL is the rst real-world implementation of RPL
developed under Contiki [15] operating system. The authors
have developed a comprehensive framework for simulation,
experimentation, and evaluation of RPL routing mechanisms
under Contiki operating system. One of the main features of
ContikiRPL is that it provides a simple programming interface
for designing and evaluating objective functions. In [14], the
authors evaluated ContikiRPL in terms of power-efciency and
implementation complexity, and they proved that the packet
delivery ratio is high and the battery lifetime may last for
several years. The measured buffer size was shown to be low as
compared to the available memory space of TmoteSky motes.
In this paper, we have performed our performance evalua-
tion under the ContikiRPL implementation.
III. OVERVIEW OF THE RPL ROUTING PROTOCOL
RPL is a distance-vector (DV) and a source routing protocol
that is designed to operate on top of IEEE 802.15.4 PHY and
MAC layers. It targets collection-based networks, where nodes
periodically send measurements to a collection point. A key
feature in RPL is that it represents a specic routing solution
for low power and lossy networks [16] [17], which stand
for networks with very limited resources in terms of energy,
computation and bandwidth turning them highly exposed to
packet losses. In particular, RPL-enabled LLNs take into
account two main features (i.) the prospective data rate is
typically low (less than 250 kbps), and (ii.) communication is
prone to high error rates, which results in low data throughput.
A lossy link is not only characterized by a high Bit Error Rate
(BER) but also the long inaccessibility time, which strongly
impacts the routing protocol design. In fact, The protocol
was designed to be highly adaptive to network conditions
and to provide alternate routes, whenever default routes are
inaccessible.
RPL is based on the topological concept of Directed Acyclic
Graphs (DAGs). The DAG denes a tree structure that speci-
es the default routes in the LLN.
The basic topological component in RPL is the DODAG,
a Destination Oriented Directed Acyclic Graph, rooted in a
special node called DODAG root, as illustrated in Fig. 1. The
DODAG root has the following properties: (i.) it typically acts
as a Low Power and Lossy Border Routers (LBR), (ii.) it
represents the data sink within the directed acyclic graph, (iii.)
it is typically the nal destination node in the DODAG, since it
acts as a common transit point that bridges the LLN with IPv6
networks, (iv.) it has the ability to generate a new DODAG that
trickles downward to leaf nodes.
Rank=0
DAG
Root
1
DAG
Root 3
Internet, IPv6
DAG
Root
2
Instance 1 Instance 2
uAC1 uAC2 uAC3
Rank=3
Rank=1
Rank=2
Rank=3
Rank=1
2
3 4
3
6
Root 3
12
13
14
13
16
2
7
8 10
9
11

Fig. 1: A RPL network with 3 DODAGs in 2 instances


An RPL network is typically formed by multiple DODAGs
satisfying the same constraints or metrics, referred to as RPL
instance identied by a unique ID, called RPLInstanceID. An
RPL node may join multiple RPL instances, but must only
belong to one DODAG within each instance.
Each node in the DODAG is assigned a rank. It represents
the location of a node within the DODAG. The rank strictly
increases in the downstream direction of the DAG, and strictly
decreases in the upstream direction. This is illustrated in Fig.
1.
In the construction process of network topology, each router
identies a stable set of parents on a path towards the DODAG
root, and associates itself to a preferred parent, which is
selected based on the Objective Function (OF). The Objective
Function denes how RPL nodes translate one or more metrics
into ranks, and how to select and optimize routes in a DODAG.
It is responsible for rank computation based on specic
routing metrics (e.g. delay, link quality, connectivity, etc.)
and specifying routing constraints and optimization objectives.
The design of efcient objective functions is still an open
research issue. However, there exist a couple of internet drafts
with proposals for OF denition. In [7], the draft proposes
to the Expected Number of Transmission (ETX) required to
successfully transmit a packet on the link as the path selection
criteria in RPL routing. The route from a particular node to
the DODAG root represents the path that minimizes the sum
of ETX from source to the DODAG root. In [10], the draft
proposes Objective Function 0 (OF0), which is only based on
the abstract information carried in an RPL packet, such as
Rank. OF0 is agnostic to link layer metrics, such ETX, and
its goal is to foster connectivity among nodes in the network.
A. RPL Control Messages
There are 4 principle RPL control messages:
DODAG Information Solicitation (DIS): it is used to
solicit a DODAG Information Object (DIO) from an RPL
node. The DIS may be used to probe neighbor nodes in
adjacent DODAGs.
DODAG Information Object (DIO): it is issued by the
DODAG root to construct a new DAG and then sent
in multicast through the DODAG structure. The DIO
message carries relevant network information that allows
a node to discover a RPL instance, learn its conguration
parameters, select a DODAG parent set, and maintain the
DODAG.
Destination Advertisement Object (DAO): it is used to
propagate reverse route information to record the nodes
visited along the upward path. DAO messages are sent
by each node, other than the DODAG root, to populate
the routing tables with prexes of their children and to
advertise their addresses and prexes to their parents.
Destination Advertisement Object Acknowledgment
(DAO-ACK): it is sent as a unicast packet by a DAO
recipient (a DAO parent or DODAG root) in response to
a unicast DAO message.
B. DODAG Construction
The DODAG construction is based on the Neighbor Discov-
ery (ND) process, which consists in two main operation (1)
Transmission of DIO control messages issued by the DODAG
root to build routes in the downward direction from the root
down to client nodes, (2) Broadcast of DAO control messages
issued by client nodes and sent up to the DODAG root to build
routes in the upward direction.
In order to construct a new DODAG, the DODAG root
broadcasts a DIO message to announce its DODAGID, its
Rank information to allow nodes to determine their positions
in the DODAG, and the Objective Function. This message will
be received by a client node which can be either a node willing
to join or an already joined node.
When a node willing to join the DODAG receives the DIO
message, it (i.) adds the DIO sender address to its parent list,
(ii.) computes its rank according to the Objective Function,
such that the nodes rank is greater than that of each of its
parents, and (iii.) forward the DIO message with the updated
rank information. The client node chooses the most preferred
parent among the list of its parents as the default node through
which inward trafc is forwarded.
When a node already associated with DODAG receives an-
other DIO message, it can proceed in three different ways (i.)
discard the DIO message according to some criteria specied
by RPL, (ii.) process the DIO message to either maintain its
location in an existing DODAG or (iii.) improve its location
by getting a lower rank in the DODAG based on the Objective
Function and the path cost. Whenever a node changes its rank,
it must discard all nodes in the parents list whose ranks are
smaller than the new computed nodes rank to avoid routing
loops.
The owchart presented in Fig. 2 summarizes the operation
of a router in a DODAG.
Receive DO
the 1st time?
Satisfy criteria
Add the sender to
the parent list
yes
No
Compute the rank
based on OF
Forward DO to
others in multicast
Discard the
packet
Process the DO
Rank<
Own_Rank
No
yes
mprove its
location + get the
lower rank
Discard the
parents with lower
rank
No
yes
Maintain
location in
the DODAG
Receive a DO
Fig. 2: The operation of a router in a DODAG
After the construction of the DODAG, each client node
would have a default upward route through which it can
transmit its inward trafc at the destination of the DODAG
root. Obviously, the default route is formed by the most
preferred parent of each node.
IV. THE SIMULATION MODEL
A. Simulation Test-bed and Objectives
We simulated RPL under COOJA [2], a well-known sim-
ulator available under Contiki operating system [1]. We have
opted for Contiki because it accurately emulates the behavior
of real motes as the simulator uses the real code implemented
on motes. Contiki is an open source, highly portable, multi-
tasking operating system for memory-efcient networked em-
bedded systems and wireless sensor networks. Both interfaces
and plug-ins can easily be added to COOJA, enabling users
to quickly add custom functionality for specic simulations.
Our objective is to evaluate the performance of RPL , with
a special focus on network construction process. To this end,
we consider the default simulation test-bed that consists in
a typical wireless sensor network spanning over a surface
of (600 m x 600 m) with one DAG root and a number of
identical nodes (emulating Tmote Sky nodes). The nodes are
spread in the space so that they form a connected network. Two
types of topology have been considered: (i.) regular topologies
and (ii.) random topologies. The regular topology means that
nodes are quasi-uniformly distributed such that the inter-node
distance is around 50m. A random topology consists of nodes
randomly distributed. The location of the DAG root in the
network has a signicant impact on the performance, as it will
lead to different rank assignment to nodes. For that reason, we
consider two locations of DAG root: at the border, and in the
middle. The latter should result in a network with smaller
rank than the former. Fig. 3 presents an example of a regular
simulated DODAG topology where node1 represents the DAG
root, and the other nodes represent simple routers. Arrows
represent the parent-child links between the nodes.
Fig. 3: Example of a DAG topology composed of 50 nodes
During the simulation, the DAG root rst initiates sending
out DIO messages, and the DAG is gradually constructed. The
trickle time interval for emitting DIO messages assumes the
initial value of 1 second (Imin), and then, it is doubled 8 times
until it reaches the maximum value Imax. For any detected
change in the DODAG (e.g. unreachable parent, new parent
selection, new DODAG Sequence Number, routing loop, etc.),
the trickle timer is reset to Imin, prescribed in DIO messages
[17].
TABLE I depicts the different scenarios considered in our
simulation study. The radio channel is set to 26 and the
transmission power was set at 31 which is the maximum
available radio output power level in COOJA.
TABLE I: Experiment Scenarios
nodes Nbr OF DAGs Nbr
Sc1:Impact of Network Size [10..100] ETX 1
Sc2:Impact of OF [10..100] ETX,OF0 1
Sc3:Impact of DAGs Number [10..100] ETX [1,4]
Sc4:Impact of Packet loss [10..100] ETX 1
B. Performance Metrics
The performance metrics analyzed in this paper are the
following.
Average Power Consumption. It represents the average
energy consumed by all the nodes in the DAG. It is the
sum of the Low Power Mode (LPM) power, CPU power,
radio listen and radio transmit power.
Average Hop Distance. The hop distance is the number
of hops between each node and the DAG root. The
number of hops primarily depends on the metric that is
specied by the objective function.
DAG Convergence Time. It is the network setup time
when the last node in the network joins a DAG and gets
its IP address. The convergence time represent the time at
which the DAG is completely constructed and all nodes
in the network have joined a DAG.
Routing Table Size. It represents the number of routing
entries it has stored during the network formation process.
As the DAO messages help to feed the routing tables,
these latter grow up with time and network size. However,
in our simulation, we did not impose any constraint on
the maximum size of this table. We note that the routing
table size is not expressed in terms of Kbyte of memory
usage but measured in terms of number of entries for
each node. Each entry has next hop node and path cost
associated with the destination node.
Packet Loss. It occurs when one or more packets of data
traveling across the DAG fail to reach their destination.
The packet loss mainly occurs in the case of link or node
failure.
V. PERFORMANCE EVALUATION RESULTS
In this section, we study the DAG construction through four
parameters: the number of nodes, the objective function (OF),
the number of DAGs and the packet loss as shown in TABLE
I. We run each simulation for a sufcient time until obtaining
a stable network.
A. Impact of the number of Nodes in a DAG
1) Average Power Consumption: For assessing the en-
ergy consumption in each network, we observed the power
consumption of each node and we computed the average
power consumption of all nodes in the network. In Fig. 4a,
we compare the average power consumption between DAGs
where the root is placed in the middle and DAGs where the
root is placed in the border in regular and random networks.
According to this gure, we nd that the power consumption
grows linearly with the number of nodes in the networks
energy consumption, which means that the power consumption
is related to the number of nodes.
The analysis of the consumed power(Fig. 5) showed that
the main proportion of the power is due to the radio listen
power since RPL requires that all the nodes listen to the radio
in order to receive control messages.
In addition, by observing the nodes into their DODAG,
we concluded that the power consumption depends on the
number of neighbors (parent, set of parents, and children) of
each node. That means that a node consumes more energy
if it has to forward much data to its neighbors. Moreover,
we measured the power consumption for each rank and we
concluded that the position in the DAG does not inuence the
1.5
2
2.5
A
v
e
r
a
g
e

p
o
w
e
r

c
o
n
s
u
m
p
t
io
n

(
m
W
)
Regular topology + DAG root in middle
Regular topology + DAG root in border
Random topology + DAG root in middle
Random topology + DAG rot in border
0 10 20 30 40 50 60 70 80 90 100
0.5
1
Number of nodes in the DAG
A
v
e
r
a
g
e

p
o
w
e
r

c
o
n
s
u
m
p
t
io
n

(
m
W
)
(a) Average power consumption
6
7
8
9
10
11
A
v
e
r
a
g
e

h
o
p

c
o
u
n
t
Regular Topology + DAG root in the middle
Regular topology +DAG root in the border
Random topology + DAG root in the middle
Random topology + DAG root in the border
0 10 20 30 40 50 60 70 80 90 100
0
1
2
3
4
5
Number of nodes in the DAG
A
v
e
r
a
g
e

h
o
p

c
o
u
n
t
(b) Average hop count
50
60
70
80
90
100
C
o
n
v
e
r
g
e
n
c
e

t
im
e

(
s
)
Regular topology + DAG root in the middle
Regular topology + DAG root in the border
Random topology + DAG root in the middle
Random topology + DAG root in the border
0 10 20 30 40 50 60 70 80 90 100
0
10
20
30
40
Number of nodes in the DAG
C
o
n
v
e
r
g
e
n
c
e

t
im
e

(
s
)
(c) Convergence time
Fig. 4: Impact of the number of nodes in the DAG construction in different network topologies
Fig. 5: Distribution of power consumption for 10 nodes
power consumption. We observed also from the gure that
when the root is in the middle, the nodes consume less energy
than when the root is in the border and this is because of data
forwarding. A node spends more energy if it has to relay /
forward from other nodes as well. This is the main reason
that in a network with higher average hops (DAG root in the
border), the energy consumption is higher because the nodes
forward more data to the neighbors.
2) Average Hop Distance: Fig. 4b shows the average hop
distance assigned to the nodes after joining the DAG by
varying the position of the DAG root in regular and random
topologies. The number of hop count illustrates the rank
assigned to each node.
According to Fig. 4b and as expected, the average hop count
increases linearly with the network size. Observe that the The
hop count increases in a greater manner when the DAG is in
the border than the topologies where the DAG is in the middle,
since the DAG in the middle is nearer to the nodes.
3) Convergence Time: Fig. 4c illustrates the DAG conver-
gence time. During this time, all the nodes in the network join
the DAG. We have performed the simulations for regular and
randomly chosen topologies and we varied the position of the
root to be in the middle or in the border.
From this gure, it is clear that the convergence time
also increases linearly with the network size. During the
simulations, we noticed that all the nodes that have the same
hop distance join the DAG at almost the same time. Thus,
the convergence time is dependent on the maximum number
of hops within the DAG and the space in which the nodes
are deployed. The convergence time is independent from the
number of nodes, since a network deployed in a small surface
with a large number of nodes may have a low convergence
time because the nodes are placed in the transmission range
of the root. In addition, it depends on the distribution of the
nodes in the surface, a sparse network have a convergence time
greater than a dense network as the nodes in a dense DAG have
more relations between each other and they forward DIOs in
a fast way. It also depends on the position of the DAG (in the
middle or in the border) as shown in Fig. 4c.
4) Routing Table Size: The objective of this metric is to
observe the distribution of the number of routing table entries
per node. Figures 6a and 6b show the CDF of routing table
entries in regular networks in surfaces of 200200m and 600
600m respectively .
One can see from Fig. 6a that 90% of the nodes have less
than 5 entries in their routing table in almost the different
network sizes, which means that the RPL protocol responds to
the memory constraints of the LLN networks in the process of
DAG construction in the case of a sparse network. In addition,
we compared the routing table size of each rank and we found
that it is almost similar for each rank. However, in a sparse
network, the routing table size increases with the number of
nodes (refer to Fig. 6b). As a conclusion, the routing table size
depends on the density of the network: the more the number
of neighbors, the more the routing table contains entries.
B. Impact of the Objective Function
To compare the OF1 based on ETX and the OF0 which
minimizes the number of hops, we measured the average
number of hops that separates the nodes from the root and
the power consumption. At each simulation, we kept the same
topology (the position of the nodes is the same) and we only
changed the objective function.
When deploying a sparse network (600600m), we noticed
that the average hop count is constant when using the OF0
and the OF1 whatever the DAG size, and this is due to the
limited number of neighbors; the choice of the preferred parent
is then limited . However, we noticed that the average power
consumption when using OF0 is a bit lower.
50
60
70
80
90
100
C
D
F

in

p
e
r
c
e
n
t
a
g
e
N=30
N=40
0 2 4 6 8 10 12 14 16
0
10
20
30
40
Routing Table Size
C
D
F

in

p
e
r
c
e
n
t
a
g
e
N=40
N=50
N=60
N=70
N=80
N=90
N=100
(a) CDF of routing table size in 200*200m
50
60
70
80
90
100
C
D
F

in

p
e
r
c
e
n
t
a
g
e
N=20
N=30
N=40
N=50
N=60
N=70
N=80
N=90
0 1 2 3 4 5 6 7
0
10
20
30
40
Routing Table Size
C
D
F

in

p
e
r
c
e
n
t
a
g
e
N=90
N=100
(b) CDF of routing table size in 600*600m
Fig. 6: Impact of routing table size
1.5
2
2.5
3
A
v
e
r
a
g
e

h
o
p

c
o
u
n
t
OF0
OF1
0 10 20 30 40 50 60 70 80 90 100
0
0.5
1
Number of nodes in the DAG
A
v
e
r
a
g
e

h
o
p

c
o
u
n
t
(a) Average hop count for the two OFs
0 10 20 30 40 50 60
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Number of nodes in the DAG
A
v
e
r
a
g
e

p
o
w
e
r

c
o
n
s
u
m
p
t
io
n
comparison between the power consumption when using ETXand OF0 objective functions
ETX
OF0
(b) Average power consumption for the two
OFs
Fig. 7: Impact of the objective function
Thus, we considered a dense network deployed in a surface
of 100 100m where nodes are very close to each other so
that a node will have multiple choices for his best parent. The
results of this simulation is shown in Fig. 8.
1) Average Number of Hops: Fig. 7a shows the difference
between the two objective functions in term of the average
number of hops that separates the nodes and the DAG root. It
is clear from Fig. 7a that a DAG based on OF0 has always a
lower number of hops than a DAG whose objective function
is based on ETX. This difference increase signicantly when
the number of nodes increases.
2) Average Node Energy: Fig. 7b shows the difference
between OF0 and OF1 objective functions in term of node
energy. One can see that the average node energy in a DAG
whose objective function is computed with OF0 is lower than
that of a DAG with ETX objective function.
As a conclusion, the number of hops is decreased in the
DAG when using OF0 and the power consumption is also
decreased as compared to the OF based on ETX metric.
Regarding the convergence time, we have noticed through
simulations that the two objective functions allow the construc-
tion of the DAG in almost the same time period (we have not
drown the gure as the convergence time is the same).
C. Impact of the number of DAGs
In this scenario, we varied the number of the DAG roots
from 1 to 4 and we observed the impact of having multiple
DAGs on the convergence time, the power consumption and
the average hop count.
1) Convergence Time: Fig. 8a shows the impact of increas-
ing the number of DAGs on the time convergence for different
network sizes from 10 until 100 nodes.
The DAG convergence time decreases linearly when adding
DAG roots to the network. This is because the nodes are
divided between the DAGs and the nodes join process occurs
in parallel between the DAGs. However, it is almost constant
for small networks.
2) Average Node Energy: Fig. 8b shows the impact of
increasing the number of DAGs on the average power for
different network sizes. One can see that the average power
consumption of the nodes decreases signicantly when the
number of DAGs increase. This leads to the fact that nodes
are independent from those in other DODAGs, the increase
in the number of DAGs minimize the relations between the
nodes which results in lower power consumption.
3) Average Hop Count: Fig. 8c illustrates the impact of
increasing the number of DAGs on the average number of
hops.
It is clear from the gure that the average hop count
decreases signicantly when increasing the number of DAGs
for large scale networks. These results are expected as the hop
count is computed separately at each DAG.
25
30
35
40
45
C
o
n
v
e
r
g
e
n
c
e

t
im
e

(
s
)
N=10
N=20
N=30
N=40
N=50
N=60
N=70
N=80
N=90
1 1.5 2 2.5 3 3.5 4
0
5
10
15
20
Number of DAGs
C
o
n
v
e
r
g
e
n
c
e

t
im
e

(
s
)
N=90
N=100
(a) Average power consumption
1.4
1.6
1.8
2
2.2
A
v
e
r
a
g
e

P
o
w
e
r

C
o
n
s
u
m
p
t
io
n

(
m
W
)
N=10
N=20
N=30
N=40
N=50
N=60
N=70
N=80
N=90
N=100
1 1.5 2 2.5 3 3.5 4
0.8
1
1.2
Number of DAGs
A
v
e
r
a
g
e

P
o
w
e
r

C
o
n
s
u
m
p
t
io
n

(
m
W
)
N=100
(b) Average hop count
3.5
4
4.5
5
5.5
6
A
v
e
r
a
g
e

H
o
p

C
o
u
n
t
N=10
N=20
N=30
N=40
N=50
N=60
N=70
N=80
N=90
N=100
1 1.5 2 2.5 3 3.5 4
1
1.5
2
2.5
3
Number of DAGs
A
v
e
r
a
g
e

H
o
p

C
o
u
n
t
(c) Convergence time
Fig. 8: Impact of the number of DAGs
D. Control Packet Overhead
In this section, we investigate the control packet overhead
of the root, a router in the middle and a leaf node in the DAG
construction process composed of 50nodes as shown in Fig.
9.
Fig. 9a shows the statistics of control packets sent by the
DAG root during the DAG construction process. As shown in
the gure, the root sends 33 DIO messages after 4 seconds
to announce in broadcast that it is the root and to solicit the
other nodes to join the DAG. At t=11s, the router continues
sending other DIO messages in broadcast in order to solicit
other nodes. The DAG root then listens to the radios and sends
acknowledgement packets and receives DAO messages for the
uplink from the routers.
Fig. 9b shows the control messages sent by a router whose
rank is equal to 2 and have 4 child nodes. Observe that, in the
beginning, the node is unable to send any message before it
joins the DAG because it should receive DIOs from neighbors.
In 10s, the node joins the DAG and sends a high number of
control messages. Before sending the DIO messages, the node
joins the DAG, gets its IP address, and builds links with its
neighbors through sending the DIO messages. After that, the
node sends and receives packets in order to keep or to improve
its position in the DAG.
Fig. 9c shows the control messages sent by a leaf node
whose rank is 4. According to this gure, we notice that the
node sends 50 DIS messages in the beginning (at t=5s) in order
to solicit joining a DAG. When the node joins the DAG (at
t=21s), the node sends DIO messages for 3 seconds and stops
sending messages as there are no neighbors that can reply to
these messages.
E. Packet Loss
Fig. 10 shows the packet loss ratio space by varying the
space in which the nodes are deployed.
It is clear from the gure that the packet loss is high when
deploying a large scale network, as the probability of link
failure is higher. When measuring the packet loss, we noticed
that a dense network is immune against the packet loss. This is
due to the nodes in this network are very close to each other,
and the nodes nd in an easy way alternative parents. The
25
30
35
40
45
50
P
a
c
k
e
t

lo
s
s

r
a
t
io

(
%
)
600*600m
200*200m
0 10 20 30 40 50 60 70 80 90 100
0
5
10
15
20
Number of nodes the a DAG
P
a
c
k
e
t

lo
s
s

r
a
t
io

(
%
)
Fig. 10: Packet loss ratio
problem of packet loss is present mainly when the DAG is
deployed in a large space (e.g 600*600m), because the nodes
are far from each other, and it is difcult to nd an alternative
link to global or local repair. The problem become worse when
the failed link is close to the root, as all the nodes that are
related to the same branch will loose connectivity and thus,
will not be able to receive packets.
VI. CONCLUSIONS
In this paper, we have performed a simulation study to eval-
uate the DAG construction process in RPL routing protocol.
The main conclusion from this study is that RPL has several
benets for LLN networks as compared to the current routing
protocols (such as IS-IS, OSPF) as it minimizes the power
consumption of the nodes mainly in sparse networks, and
reduces the memory consumption of the nodes by reducing
the routing table size even in large scale networks.
From our simulations, we concluded that placing the DAG
root in the middle of the surface will decrease the hop distance
and thus the convergence time of the DAG. Moreover, we
observed that the OF0 allows lower hop distance and lower
power consumption than OF1 based on ETX, but OF0 does not
consider the network throughput, which opens the challenge of
designing new objective functions tailored to the application
requirements. We noticed also that the increase in the number
of DAGs improves the network performance as each node will
belong to a smaller DAG.
20
25
30
35
N
u
m
b
e
r

o
f

c
o
n
t
r
o
l
p
a
c
k
e
t
s
0 5 10 15 20 25 30
0
5
10
15
Time (s)
N
u
m
b
e
r

o
f

c
o
n
t
r
o
l
p
a
c
k
e
t
s
(a) Control packets of the root
20
25
30
35
40
N
u
m
b
e
r

o
f

c
o
n
t
r
o
l
p
a
c
k
e
t
s
0 5 10 15 20 25 30
0
5
10
15
Time (s)
N
u
m
b
e
r

o
f

c
o
n
t
r
o
l
p
a
c
k
e
t
s
(b) Control packets of a router
25
30
35
40
45
50
N
u
m
b
e
r

o
f

c
o
n
t
r
o
l
p
a
c
k
e
t
s
0 5 10 15 20 25 30
0
5
10
15
20
25
Time (s)
N
u
m
b
e
r

o
f

c
o
n
t
r
o
l
p
a
c
k
e
t
s
(c) Control packets of a leaf node
Fig. 9: Control packet overhead during DAG construction
Our analysis of packet loss showed that a sparse network
is exposed to very high packet loss as the nodes frequently
have not any alternative parents for an intermediate repair in
the case of node or link failure, and this raises the challenge
of improving the design of the repair mechanisms in RPL.
REFERENCES
[1] Contiki Operating System. [Online]. Available:
http://www.sics.se/contiki/
[2] Cooja Simulator. [Online]. Available:
http://www.sics.se/ fros/cooja.php
[3] The Internet Engineering Task Force (IETF), 2010.
[Online]. Available: http://www.ietf.org/
[4] C. Chauvenet, Tourancheau, B. Genon-Catalot,
D. Goudet, and M. P.-E. Pouillot, A communication
stack over PLC for Multi Physical Layer
IPv6 Networking, Smart Grid Communications
(SmartGridComm), 2010 First IEEE International
Conference on, November 2010.
[5] T. Clausen and U. Herberg, Comparative Study of
RPL-Enabled Optimized Broadcast in Wireless Sensor
Networks, Sixth International Conference on Intelligent
Sensors, Sensor Networks and Information Processing
(ISSNIP2010), Brisbane, Australia, December 2010.
[6] , Multipoint to Point and Broadcast in RPL, 13th
International Conference on Network-Based Information
Systems, pp. 1 6, September 2010.
[7] O. Gnawali and P. Levis, The ETX Objec-
tive Function for RPL, draft-gnawali-roll-etxof-00,
2010. [Online]. Available: http://tools.ietf.org/html/draft-
gnawali-roll-etxof-00
[8] K. Kim, S. Park, I. Chakeres, and C. Perkins, Dy-
namic MANET On-demand for 6LoWPAN (DYMO-low)
Routing, draft-montenegro-6lowpan-dymo-low-routing-
03, June 2007.
[9] K. Kim, S. Yoo, J. Park, S. D. Park, and J. Lee, Hierar-
chical Routing over 6LoWPAN (HiLow), draft-deniel-
6lowpan-hilow-hierarchical-routing-00.txt,, vol. 38, De-
cember 2005.
[10] E. P. Thubert, RPL Objective Function 0,
draft-ietf-roll-of0-03, 2010. [Online]. Available:
http://tools.ietf.org/html/draft-ietf-roll-of0-03
[11] A. Tavakoli, HYDRO: A Hybrid Routing Protocol for
Lossy and Low Power Networks, draft-tavakoli-hydro-
01, September 2009.
[12] J. Tripathi, J. de Oliveira, and J. Vasseur, A Perfor-
mance Evaluation Study of RPL: Routing Protocol for
Low Power and Lossy Networks, Information Sciences
and Systems (CISS), 2010 44th Annual Conference on ,
pp. 16, May 2010.
[13] , Applicability Study of RPL with Local Re-
pair in Smart Grid Substation Networks, Smart Grid
Communications (SmartGridComm), 2010 First IEEE
International Conference on, pp. 16, November 2010.
[14] N. Tsiftes, J. Eriksson, and A. Dunkels, Poster Abstract:
Low-Power Wireless IPv6 Routing with ContikiRPL,
IPSN10, Stockholm, Sweden., p. 1216, April 2010.
[15] N. Tsiftes, J. Eriksson, N. Finne, O. Fredrik, J. Hglund,
and A. Dunkels, A Framework for Low-Power IPv6
Routing Simulation, Experimentation, and Evaluation,
SIGCOMM10,New Delhi, India, pp. 479480, November
2010.
[16] J. Vasseur and A. Dunkels, Interconnecting Smart Ob-
jects with IP: The Next Internet. Morgan Kaufmann, 1
edition, 2010.
[17] T. Winter and P. Thubert, RPL: IPv6 Routing Protocol
for Low Power and Lossy Networks, IETF Internet-
Draft draft-dt-roll-rpl.txt., vol. 3, 2010.
[18] W. Xie, Goyal, M. Hosseini, H. Martocci, J. Bashir,
Y. Baccelli, and A. E. Durresi, Routing Loops in DAG-
based Low Power and Lossy Networks, Advanced Infor-
mation Networking and Applications (AINA), 2010 24th
IEEE International Conference on, pp. 888 895, April
2010.
[19] W. Xie, M. Goyal, H. Hosseini, J. Martocci, Y. Bashir,
E. Baccelli, and A. Durresi, A Performance Analy-
sis of Point-to-Point Routing along a Directed Acyclic
Graph in Low Power and Lossy Networks, 2010 13th
International Conference on Network-Based Information
Systems, pp. 111116, 2010.

Vous aimerez peut-être aussi