Performance Analysis of the Actuator Discovery Protocol ... - Fethi Filali

Institut Eurécom
Department of Mobile Communications
2229, route des Crêtes
B.P. 193
06904 Sophia-Antipolis
FRANCE
Research Report RR-06-159
Performance Analysis of the Actuator Discovery Protocol
for Mobile Sensor and Actuator Networks
March 30 th , 2006
Muhammad Farukh Munir and Fethi Filali
Tel : (+33) 4 93 00 82 03
Fax : (+33) 4 93 00 82 00
Email : {Muhammad-Farukh.Munir, Fethi.Filali}@eurecom.fr
1 Institut Eurécom’s research is partially supported by its industrial members: Bouygues Télécom,
Fondation d’entreprise Groupe Cegetel, Fondation Hasler, France Télécom, Hitachi, Sharp, ST
Microelectronics, Swisscom, Texas Instruments, Thales.

Performance Analysis of the Actuator Discovery Protocol
for Mobile Sensor and Actuator Networks
Muhammad Farukh Munir and Fethi Filali
Abstract
This paper presents the performance analysis of our proposed energydelay
aware coordination protocol, ADP [1] (Actuator Discovery Protocol)
for mobile SANETs (Sensor and Actuator Networks). We have presented
our analysis on ADP for static SANETs in an earlier work and it is shown
that ADP provides optimal attachment for a sensor node to the actuator with
minimum cost. For the sensor-actuator coordination, the proposed ADP can
guarantee ordering, synchronization and eliminates the redundancy of actions
[1]. In this work, we present our analysis on the dynamics of ADP
to handle mobile sensor and actuator applications. The interval chosen to
validate the acquired paths through ADP due to mobility can be a direct
mapping of application’s energy and delay constraints. Our NS-simulations
show that the total-energy consumed to validate the paths for time-stringent
traffic is actually less compared to the overall energy consumed during arbitrary
transmission of sensor data to-wards the sink. We have utilized a
random way-point mobility model in our ns-simulations and analyze the performance
of ADP in terms of energy estimates for maintaining paths (toward
the optimal-actuator) in order to keep a promised QoS for time-stringent applications.
The proposed ADP has also achieved a good balance for sensor
energy-consumption while keeping a low end-to-end delay (optimal number
of ’message-exchange’ to keep the paths updated).
Keywords: SANETs (Sensor and Actuator Networks), self-organization,
coordination protocols, energy and delay aware coordination protocols.
ii

Contents
1 Introduction and Related Work 1
2 The ADP for Mobile SANETs 2
2.1 Overview of the ADP . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 ADP Dynamics for Mobile SANETs . . . . . . . . . . . . . . . . 4
3 Experimental analysis of the ADP 5
4 Conclusion and Future Work 8

List of Figures
1 AttachRequest by sensors at the start of ADP . . . . . . . . . . . . . . . . . . 3
2 Actuator-replies (AttachReply) for corresponding AttachRequest messages . . . . . . 4
3 The occurrence of Failure due to mobility and Recovery (exchange of ’Hello’ msgs) Procedure. 4
4 CDF for Attachment-Delay for Static Complex Network Configuration . . . . . . . . 6
5 Mean Path Length for Static Complex Network Configuration . . . . . . . . . . . 6
6 Average Cluster Size over time for Static and Mobile SANETs . . . . . . . . . . . 7
7 Cluster Densities avg_speed = 1-m/s, pause_time 20s, Tx_range = 11 m, distance d (between
any two nodes) = 10 m . . . . . . . . . . . . . . . . . . . . . . . . 8
8 Cluster Densities avg_speed = 2-m/s, pause_time 10s, distance d (between any two nodes)
= 10 m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
9 Average Cluster Size for Mobile SANET Configuration Vs. Time . . . . . . . . . . 10
10 Delay Comparison for Static and Mobile SANET Configuration . . . . . . . . . . 11
11 Energy-Consumption for Static and Mobile SANETs . . . . . . . . . . . . . . . 11
iv

1 Introduction and Related Work
The advent in technology led to the emergence of SANETs giving a distributed
control to the management, communication and coordination aspects of the network
functioning formerly referred to as WSNs (Wireless Sensor Networks). A
SANET consists of a group of sensors and actors (actuators), that are deployed to
perform distributed sensing and actuation tasks, linked up by a wireless medium.
The sensor nodes (small, cheap devices with limited computation) are deployed for
the collection of data through a sensing mechanism, while actuators (resource rich,
better processing capabilities and stronger transmission power) take decisions and
then perform appropriate actions upon the environment. Most of the sensor network
research tends to focus on specific hardware with efficient, ingenious communication
protocol algorithms and system control architectures that address the
specific resource constraints, which did not lead to the emergence of a generic architecture
for the sensor networks [2, 3, 4]. Despite of the sufficient work done
in SANETs [5, 6, 7, 8], an effective and distributed coordination mechanism for
efficient sensing, dissemination of information, and to perform right and timely
actions (actuated by the actuators) is required. These networks can be the integral
part of systems such as: disaster/crime prevention, real-time military applications,
environmental and health monitoring to smart spaces [9].
This paper analyzes our proposed self-organizing framework for the mobile
SANETs. In this regard, we have implemented ADP (Actuator Discovery Protocol)
as an application layer coordination protocol. which provides optimal attachment
to a sensor node in the network with the nearest and most relevant actuator
node. A sensor node anywhere in the network finds the nearest actuator using the
proposed ADP during the initial deployment phase. Afterwords the data is relayed
to the attached-actuator through a multi-hop discrete path unless the network undergoes
a remarkable topology change, which covers the basics of sensor-actuator
coordination. We have employed a periodic exchange of “Hello” messages to keep
the paths updated. The update interval can be modified according to the applications’
requirements. Even in case of mobile SANETs, the mobility of the nodes
is not expected to be very high as opposed to their sensing duration during monitoring
a zone, which would be a realistic deployment situation for most mobile
applications. ADP is very simple to accommodate the limitations of small sensor
nodes and is energy efficient. The functionality of ADP is independent of the
routing protocol and is distributed to cope with large scale deployments. For the
actuator-actuator coordination, we employ the use of the assumed separate wireless
interface to communicate with the neighboring actuators so that they can perform
long-range data communication without any involvement of sensor nodes to effectively
coordinate and to perform the actuation process. The examples of actuation
covers a wide range of possibilities and is typically application dependent.
For example, targeting an enemy holding a snipper in a battle field can be an
interesting case to consider. The actuation process has to localize the position of
the enemy and actuate the destruction process. But the important constraint in this
1

case is the latency because the sensor data can be no more valid at the time of
actuation in case of increased latency. Which makes the assumption of a separate
wireless interface for actuator-actuator coordination quite relevant.
Note that in this work, we study the generic architecture which is robust and entirely
distributed with a little varying-degree of centralized control for the actuators
on their local clusters. In prior work done for SANET coordination protocols, an
excess of energy is consumed whenever sensors have some sensed data to transmit
to the nearest sink, even at the cost of variable delay and regardless of the fact that
events were rarely detected or an any-cast transmit strategy is followed. Instead
of trying to achieve improvements for a particular application, we addressed both
the static and dynamic network configurations with different natures of mobility,
changes to the discrete paths, energy and delay efficient path-recovery procedures,
and avoid using the GPS system for position localizations. In addition, the energy
consumption by the sensors (in order) to keep the updated-paths is highly dependent
on mobility. In this regard, the nodes can adaptively tune their update-interval
based on their mobility, which makes the ADP highly adaptive to any kind of network
configuration with optimum energy consumption and a promised QoS. It
also has the ability to adapt any nature of mobility at the expense of minimum total
network energy-consumption. This feature has the tendency to lean the SANETs
to-wards minimum residual energy sensor and actuator networks. We believe that
the proposed ADP leads toward optimal total network energy consumption providing
a guaranteed QoS for both densely deployed and stringent real-time traffic
networks.
The remainder of the paper is structured as follows. In Section 2, the dynamics
of the proposed ADP for mobile SANETs are explained in detail. Section 3
presents the experimental results. Section 4 summarizes the paper with a brief
conclusion and outlines the future work.
2 The ADP for Mobile SANETs
As we have already explained in the earlier section that there is no sensor/actuator
network structure and architecture, and the basic goals of the previously done work
relied mostly on the considered application. In this work, we propose an architecture
which is tailored toward a standard behavior for most deployment scenarios,
aiming to satiate the time-stringent requirements and effective energy utilization in
a purely distributed fashion.
2.1 Overview of the ADP
The learning-phase starts when the sensors during the initial deployment stage locate
the nearest actuator using a one-hop broadcast. The deployment of the sensor
and actuator nodes can be randomly chosen or follow any pre-defined distributions.
The finding of the "optimal-actuator attachment" for each sensor node is
2

done through the proposed novel protocol called ADP.
When a sensor node is turned on, it should first determine to which actuator
node it has to be associated. For this end, a sensor node transmits a broadcast message
named AttachRequest(cost, M j , C ) to all its one-hop neighbors as shown in
Figure 1. A neighboring node upon receiving an attach-request message checks
that it has sent an attach-request in the period T n (application specific), if it has
already sent a broadcast to its neighbors, it has to wait for a reply until timeout.
Otherwise it does the same unless the probe reaches the nearest actuator. The reply
message named AttachReply i (cost, M j , A i ) from the actuator follows the probe
and terminates at its origin defining a discrete path to the sensor node. If a node
receives multiple paths then it chooses the shortest one (hop-count), and certainly
if it receives more than one path containing the same hop-count, it chooses the
earliest reply.
ACTOR NODE
Sensor Nodes
AttachRequest
Figure 1: AttachRequest by sensors at the start of ADP
We don’t take into account the distance between any node and its associated
neighbors with the reason being that the energy required to transmit to a node
in its sensing radius is a constant (no power control assumed for transmissions).
ADP produces loop-free paths to the actuator nodes, as stated below. The next-hop
selected by a sensor with ADP has a defined optimal path to the actuator node.
As depicted by Figure 2, now a sensor node has a specified path to route its
sensed data to the actuator nodes by simply forwarding it to only one of its neighbors
(immediate next node in the path to the actuator), and the actuator also keeps
the defined path to the node (building its tree structure for the localized cluster).
The background for this consideration takes us to the lack of a GPS system for
position localizations in our study, which consumes a lot of sensor nodes’ energy.
In a similar fashion all the nodes reserve an optimal path to their nearest actuators,
forming a local cluster, thus giving us the initial deployment in the form of welldistributed
small clusters.
3

Attach−Replies
Actor−Node
Sensor−Node
Attach−Reply
Paths
not chosen due
more hop−count
Figure 2: Actuator-replies (AttachReply) for corresponding AttachRequest messages
2.2 ADP Dynamics for Mobile SANETs
The Failure and Recovery phase monitors a periodic update of the optimal paths
and if any particular path is not valid anymore, the sensors send a one-hop broadcast
to their neighbors and acquire another optimal actuator to route their further
sensed data. The sensor then informs the optimal actuator to join its local cluster
and similarly the actuator informs the neighboring actuators of the modification for
effective actuation process. The process of updating a path in case of node failure
or mobility is shown in Figure 3.
At T2, an Actor updating the neighboring Actors
New Cluster
found at T2
Node moving from
One Cluster to another at T1
Figure 3: The occurrence of Failure due to mobility and Recovery (exchange of ’Hello’ msgs) Procedure.
In this work, we have taken into consideration two possible deployment scenarios.
(i) Both sensors/actuators are static: For the static-case, if there is a node
failure in the middle of the path, the farther nodes can do a one-hop broadcast
to all the neighbors in their sensing range, which can reply with their respective
4

hop-counts toward the actuator. The sensor node selects the shortest path to the
actuator and an update message is sent (piggybacked along with the data-packet, if
necessary) to the actuator for updating the correlation tree.
(ii) Static actuators and mobile sensors: In mobile-sensor configuration, the
path to the attached-actuator can be no more valid at any instant in time due to the
mobility pattern of the nodes, and can also be the result of an accidental-failure.
The nodes that have no more defined paths sends a one-hop broadcast. If a reply
is received, the same procedure will follow as for the static case. Otherwise the
one-hop neighbors send a path-search probe to their one hop neighbors, and the
procedures can terminate after acquiring a path to the actuator. In-fact, if this is the
case, the neighboring nodes that don’t have a defined path has already sent a pathupdate
request after time-out in most practical scenarios. All the nodes have to
authenticate the validity of their paths after a certain defined interval T o which can
be a trade-off between the mobility model and the path-update procedure. In the
mobile-sensor case the constant dissemination of energy to keep the paths updated
can be a direct mapping to the applications’ requirements. The dry analysis for
the proposed scheme intuitively satisfies the efficient consumption of energy and
meeting the required constraints for real-time traffic. In contrast the total energy
consumed by the ADP to keep the paths updated is less than the total energy consumed
by the sensors using any of the coordination mechanisms proposed earlier
for the reason that they were application dependent. With our ADP, we can simply
maintain a trade-off between a predefined latency and sensor energy consumption
by managing the interval T o , which is mobility dependent and hence making this
an implementation issue. The alternate-path obtained is also a loop-free optimal
path to the actuator node. The Actors coordinate among each other to keep the
updated networking paradigm for effective actuation process.
3 Experimental analysis of the ADP
We finally implemented the ADP in ns-2 [21] as an application layer protocol. We
have considered a wide range of network topologies and test-validated the performance
of ADP with all considered topologies. Due to the posed space limitation,
we present our results for a complex random network (static and mobile). We
consider a complex random network configuration with 100 nodes (5% actuators),
where both sensor and actuator nodes are static. We have also evaluated the average
number of sensors per actuator (local cluster) for complex random network
topology. The local clusters have an equilibrium in terms of number of sensors,
and the calculated mean turns out to be 19 sensors per actuator, which proves that
the paths chosen by the sensors to their respective actuators are the optimal paths.
The CDF for the delay distribution shows that over 65% of the nodes lie in the
average-delay (according to our findings) range as shown in Figure 4 along with
the mean number of nodes per cluster. We have also shown, in Figure 5, the mean
path length as a function of network size. The mean path length, which is related
5

to the end-to-end latency for both cases: Fixed number of actuators, and secondly
actuators increasing by 5% with network size. However, the increase is more gradual
with the increasing number of actuator nodes, which would be a more practical
assumption for the SANETs.
90
80
70
Number of Sensors
60
50
40
30
20
100 Nodes in the Network & 5% Actuators
10
0
0.5 1 1.5 2 2.5 3 3.5 4
Delay (in ms)
Figure 4: CDF for Attachment-Delay for Static Complex Network Configuration
8
7
Fixed Number of Actors
Actors increasing by 5%
6
Mean Path Length
5
4
3
2
1
0
0 50 100 150 200 250
Nodes in the Network
Figure 5: Mean Path Length for Static Complex Network Configuration
In Figure 6, we have obtained the average local cluster densities as a function
of time for both static and mobile SANETs. The cluster average shows the optimal
equilibrium obtained in terms of efficient energy utilization inside each local
cluster. This phenomenon leads the network toward minimal total network energyconsumption,
if cost functions are well driven. In analyzing the performance of
6

ADP, the interesting measurement of a pre-promised delay for any type of traffic
comes from the increased optimal energy consumption, which was usually not the
case for previous work.
25
Mobile-Case, path updates and recovery
Static-Case, no updates required
20
Number of Sensors
15
10
5
0
0 5 10 15 20 25 30 35 40 45 50
Time (in sec.)
Figure 6: Average Cluster Size over time for Static and Mobile SANETs
Similarly in the dynamic node configuration, we have again considered a random
complex network with 100 nodes (5% actuators), where the sensors are mobile
and actuators have defined positions. We haven’t thought of any particular mobile
application of SANETs, so we focus our major goal to-wards optimizing the global
energy-consumption while maintaining an optimal delay for stern real-time traffic.
We have applied a random way-point mobility model in our ns-simulations and analyze
the performance of ADP in terms of energy estimates for maintaining paths
(to-wards the optimal-actuator) in order to keep a promised QoS for time-stringent
applications. The proposed ADP has also achieved a good balance toward energyconsumption
while keeping a low end-to-end delay (optimal number of ’messageexchange’
to keep the paths updated). The mobility of the sensors is kept slow
(mean speed = 1-m/s) and the duration of stay at one place is kept more (pause
time = 20s) for the appropriate sensing of the deployed environment. The Cluster
densities for two different mobility patterns vs. time are shown in the Figure 7 and
8 respectively. In both cases, the clusters tend to acquire an equilibrium in their
sensor density based on the mobility pattern. Similarly the average cluster size
over time for both cases is shown in Figure 9.
We computed matrices for energy consumption and delays for mobile SANETS.
Figure 11 shows the increased energy consumption for updating the paths that are
lost due to the mobility in the network as opposed to energy consumed for static
networks. The update as explained before comprises of only one ’Hello’ message
exchange between any two sensors that are part of the multi-hop path to-wards the
optimal actuators. It is worth mentioning that the delay has been considerably low,
7

40
35
30
Actor 1
Actor 2
Actor 3
Actor 4
Actor 5
Number of Sensors
25
20
15
10
5
0
0 100 200 300 400 500 600 700 800 900 1000
Time (in sec.)
Figure 7: Cluster Densities avg_speed = 1-m/s, pause_time 20s, Tx_range = 11 m, distance d (between any
two nodes) = 10 m
as shown in Figure 10 even in the course of mobility due to the periodic update
of multi-hop paths. The total network energy consumed during the lifetime of the
SANETs is considerably low with ADP compared to previous proposed coordination
protocols.
4 Conclusion and Future Work
For sensor-actuator coordination, the proposed ADP can acquire a promised QoS
for time stringent traffic at the cost of optimal energy consumption. For actuatoractuator
coordination, the proposed unified framework using AODV can be exploited
by different applications to always select the best networking paradigm
that is possibly available according to the sensed information and to perform the
necessary operations in an efficient way. The issues related to node-failures and
mobility are well handled by the periodic exchange of ’Hello’ messages in all deployment
scenarios. The energy consumption for obtaining new paths in-case of
failure is also energy efficient and is driven by the collaborative nature of SANETs.
For the moment, we are working on the design of a delay-energy sensitive
routing protocol to work in collaboration with our ADP, to make the architectural
and coordination platform sound for the future of SANETs. We are also working
on MAC optimization based on the proposal for the efficient utilization of sensorenergy
and to satiate the real-time problems experienced so far in the SANETs.
References
[1] M. Farukh Munir and F. Filali, A Novel Self Organizing Framework for SANETs, to appear in Proc, of
8

40
35
30
Actor 1
Actor 2
Actor 3
Actor 4
Actor 5
Number of Sensors
25
20
15
10
5
0
0 100 200 300 400 500 600 700 800 900 1000
Time (in sec.)
Figure 8: Cluster Densities avg_speed = 2-m/s, pause_time 10s, distance d (between any two nodes) = 10 m
the 12th European Wireless Conference (EW’06), April 2006.
[2] V. Paruchuri, A. Durresi, and L. Barolli, “Energy Aware Routing Protocol for Heterogeneous WSNs”, in
Proc. of DEXA, August 2005.
[3] Q. Fang, J. Gao, and L. J.Guibas, “Locating and Bypassing Routing Holes in Sensor Networks” in proc.
of IEEE Infocom March 2004.
[4] J-H. Chang, and L. Tassiulas, “Maximum Lifetime Routing in WSNs”, IEEE/ACM Transactions on Networking,
Vol. 12, No. 4, August 2004.
[5] Akyildiz, I.F, Kasimoglu, I.H, “sensor and actor networks: research challenges”. Article Ad Hoc Networks,
Vol. 2, No. 4, pp 351-367, October 2004.
[6] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci, “WSNs: a survey”. Article Computer Networks,
Vol. 38, No. 4, pp 393-422, March 2002.
[7] T. Melodia, D. Popili, V. C. Gungor and I. F. Akyildiz, “A distributed coordination framework for WSANs:
in Proc. of MobiHoc, May 2005.
[8] W. Hu, N. Bulusu, and S. Jha, “A Communication Paradigm for Hybrid SANs”, International Journal of
Wireless Information Networks, Vol. 12, No. 1, pp 47-59, January 2005.
[9] E. Yoneki, J. Bacon, “A survey of WSN technologies: research trends and middleware’s role” Technical
Report, No. 646, September 2005.
[10] V. P. Mhatre, C. Rosenberg, D. Kofman, R. Mazumdar, N. Shroff, “A Minimum Cost Heterogeneous
Sensor Network with a Lifetime Constraint”, IEEE Transactions on Mobile Computing, Vol. 4, No. 1, pp.
4-15, January 2005.
[11] Xiang Ji, and Hongyuan Zha, “Sensor Positioning in Ad-hoc WSNs Using Multidimensional Scaling”, in
Proc. of IEEE Infocom, March 2004.
[12] Yi Shang, and Wheeler Ruml, “Improved MDS-Based Localization”, in Proc. of IEEE Infocom March
2004.
[13] G. Sarma, and R. Mazumdar, “Hybrid SNs: A Small World”, in proc. of MobiHoc, May 2005.
9

30
av.speed = 2-m/s, pause-time = 10s
av.speed = 1-m/s, pause-time = 20s
25
Number of Sensors
20
15
10
5
0
0 5 10 15 20 25 30 35 40 45 50
Time (in sec.)
Figure 9: Average Cluster Size for Mobile SANET Configuration Vs. Time
[14] A. Cerpa, J. L. Wong, M. Potkonjak, and D. Estrin, “Temporal Properties of Low Power Wireless Links”,
in Proc. of MobiHoc, May 2005.
[15] I. F. Akyildiz, M. C. Vuran and O. B. Akan, ” On Exploiting Spatial and Temporal Correlation in WSNs”,
in Proc. of IWWAN, May 2004.
[16] H. Gupta, V. Navda, S. R. Das, and V. Chowdhary, “Efficient Gathering of Correlated Data in Sensor
Networks” in Proc. of MobiHoc, May 2005.
[17] G. Xing, C. Lu, Y. Zhang, Q. Huang, and R. Pless, “Minimum Power Configuration in WSNs”, in Proc.
of MobiHoc, May 2005.
[18] F. Kuhn, T. Moscibroda, and R. Wattenhofer, “Initializing Newly Deployed Ad Hoc and Sensor Networks”,
in Proc. of MobiHoc, May 2004.
[19] J. L. Bredin, E. D. Demaine, M. T. Hajiaghayi, and D. Rus, “Deploying Sensor Networks with Guaranteed
Capacity and Fault Tolerance”, in Proc. of MobiHoc, May 2005.
[20] B. Liu, P. Brass, O. Dousse, P. Nain, and D. Towsley, “Mobility Improves Coverage of Sensor Networks”,
in Proc. of MobiHoc, May 2005.
[21] The Network Simulator - ns (version 2.27). http://www.isi.edu/nsnam/ns/.
10

4.5
4
3.5
3
Delay (in ms)
2.5
2
1.5
1
Static-Case, 100 Nodes, 5% Actuators
Mobile-Case, 100 Nodes, 5% Actuators
0.5
0
0 20 40 60 80 100
Sensors in the Network
Figure 10: Delay Comparison for Static and Mobile SANET Configuration
0.006
0.0055
Energy Consumption (in joules)
0.005
0.0045
0.004
0.0035
0.003
0.0025
0.002
ADP for Static SANET, no Updates
ADP for Mobile SANET, updates and path-recovery
0.0015
0.001
0 20 40 60 80 100
Sensors in the Network
Figure 11: Energy-Consumption for Static and Mobile SANETs
11