You are on page 1of 8

2017 VII Brazilian Symposium on Computing Systems Engineering

μSDN: An SDN-based Routing Architecture for


Wireless Sensor Networks
Leonardo Francisco da Silva Santosa, Francisco Ferreira de Mendonça Júniorb, Kelvin Lopes Diasc
Center of Informatics (CIn)
Federal University of Pernambuco (UFPE)
Recife, Brazil
E-mail: {lfss3a , ffmjb, kldc}@cin.ufpe.br

Abstract— Recently, the SDN paradigm, which splits the control and (industrial, automation, and smart homes) frequently changing
data planes, initially defined for wired networks, has been considered the firmware or operating systems of all sensors belonging to a
as a solution for the management of WSNs (Wireless Sensor WSN may be infeasible and counterproductive. Nevertheless,
Networks). However, the adoption of OpenFlow protocol, the most the process of changing the routing protocol in a WSN may
widely deployed SDN standard, directly into the WSNs may require become cumbersome once it depends on using the
novel / customized hardware or incur significant signaling overhead. reprogramming method available for each platform. The
This paper proposes a novel routing architecture for WSN, based on primary mechanisms for reprogramming WSN are: manual
SDN principles, by extending existing protocols, which allows its reprogramming, which consists on rewriting the firmware on
deployment in legacy motes. A testbed based evaluation demonstrates
each node, and remote reprogramming, also called Over the Air
the comparable performance results regarding signaling overheads
(OTA), which consists of using the wireless interface. Each of
and energy consumption to AODV and LQRP, routing protocols
widely used in WSN and extended in this papers as a means of those methods has an associated operating cost, which is closely
demonstrating the feasibility of a smooth transition to the future all- related to consumption of computing resources and energy, as
SDN WSNs. Although a case study has been conducted on a well as the time spent during the reprogramming of network.
proprietary platform, the paper shows a viable path for similar Such time may include the period where the system is entirely
extensions on other solutions to benefit from flexible management or partially inactive so it may become problematic for critical
and programmability provided by the SDN paradigm. applications.
The problems related to the network management and
Keywords—Wireless Sensor Networks, Software-Defined Networks,
Programmable Networks, Internet of Things.
routing are also present in wired networks. Future Internet
design has driven the emergence of a new architectural paradigm
I. INTRODUCTION designed to create flexible networks, by allowing tests and
innovations without affecting production traffic [3]. The
Wireless Sensor Networks (WSN) are a fundamental component
Software-Defined Networking (SDN) paradigm proposes the
of IoT (Internet of Things) to get information about
separation between the data and the control planes, turning
environments and individuals, thus acting as a bridge between
network equipment (e.g. switches) into simple packet
physical and digital worlds [1]. WSN’s nodes are constrained
forwarding devices. The control plane regulates the behavior of
devices with restrictions with regard battery lifetime, processing
the entire network, and it is logically centralized software
power, storage, and communication. Hence, WSN present
running on commodity hardware. [3]
tightly coupling between hardware and software, which leads to
the design of application-specific sensor networks. Some works have naturally shifted efforts to provide SDN
solutions for WSN, but differently from wired counterparts,
Recent applications require multi-hop communication, and
constrained devices and their specific routing requirements over
even Internet connections, so energy consumption becomes a
wireless links turn OpenFlow (OF) enabled sensors a
major concern for the survivability of applications and WSN.
questionable solution. On the one hand, OF signaling from
Research efforts attempt to create energy-efficient routing
conventional WSN’s sensors to a central SDN controller for
policies to optimize the utilization of wireless network
routing decision may be unpractical and require powerful
interfaces, which consumes most of the energy of sensor nodes.
hardware, which may be unfeasible for sensor nodes. On the
Thus, proposals for multi-hop routing protocols have emerged.
other hand, legacy systems without hardware modification may
Nonetheless, most of those policies still require adjustments to
not interoperate with pure SDN solutions to improve flexibility
fit the scenarios where WSNs work, so the application can find
and adaptability.
the balance between performance and consumption of both
computing resources and energy [2]. Thus, the benefits SDN provide to WSN still require
complete evaluation, and this work presents evaluations
Therefore, in the current scenario, the manager/administrator
methods and results to demonstrate SDN may be suitable for
of WSN still needs to test distinct routing protocols and choose
WSN. The improvements achieved with the utilization of SDN
the most suitable for both the application and the characteristics
in wired networks, the unresolved concerns, and the requirement
of the production network. In large deployment scenarios (smart
for a complete evaluation, motivated this work.
city sensors for smart grids, forest fire surveillance, natural
disaster monitoring,) and even in indoor environments

2324-7894/17 $31.00 © 2017 IEEE 63


DOI 10.1109/SBESC.2017.15
This paper proposes an SDN-based routing approach, named Nodes. Despite suggesting OpenFlow as the communication
μSDN, which extends a legacy platform and well-known WSN protocol between the Master Nodes and the Center Nodes, the
routing protocols. μSDN can reduce the time required to paper does not provide any implementation or testbed. Although
reprogram an entire WSN, as it makes all the modifications in a they did not implement their application, they evaluate it
single node (the Controller). regarding energy consumption, lifecycle and load balancing
using mathematical modeling using MATLAB.
The remainder of this article is organized as follows: Section
2 presents related work. Section 3 describes implementation Work in [10] proposed a hybrid architecture for Software-
details. Section 4 delineates the evaluation of the proposal and defined WSN, which kept some elements of the legacy network.
discusses the results. Finally, Section 5 concludes this paper and For instance, the dedicated router nodes could understand route
highlights plans for future work. requests from the AODV protocol, based on policies defined in
the OpenFlow daemon. Nevertheless, this proposal requires the
II. RELATED WORKS router nodes to have a different, specialized and even a powerful
This section provides background on SDN-based solutions hardware. So, to make this strategy work, some nodes should be
for WSN in general, as well as ones focusing on routing. replaced, or still, it may be necessary to add more nodes to the
Software-Defined Wireless Networks (SDWN) [4] is a seminal network, but they did not evaluate such burden.
work applying SDN to WSN. The authors proposed The work in [11] advocates the combination of context
requirements for SDWN, including support to duty cycles, data awareness and routing policy for WSN, based on SDN. By
aggregation, flexible rules, mobility, link unreliability and node context, authors include information such as network status and
failure. Despite the proposal, mostly based on a network node conditions. The policy-based approach makes routing, as
operating system, they provide neither an implementation well as the packet forwarding, dynamic and flexible, once it can
platform nor evaluation. act according to the context. Despite the discussions, this work
The work in [5], Sensor OpenFlow (SOF), presented was limited to the description of the architecture, considering the
technical challenges for the implementation of SDN in WSN and use of OpenFlow, but did not provide an implementation or
examined some possible solutions. The paper claims there are evaluation of the proposed solution.
some key problems for the realization of SOF: the creation of The author in [12] presented “a Software Defined
flows, the allocation of a communication channel between the Networking (SDN) solution for WIreless SEnsor networks”
network devices and the controller, the overload introduced by (SDN-WISE). Their goal was to control some messages through
signaling, the traffic generation, in-network processing and an OpenFlow-based architecture, containing state array,
backward compatibility. Those concerns exist in current neighbors array, and flow table. They also corroborated the
implementations of SDN to WSN, but now emerges a new requirements in [4]. Although they described the architecture
discussion about standardization [6]. The proposed solution and platform, their controller required a physical connection to
relies on the adoption of specialized flow tables, which take into a computer or a dedicated embedded system, that is, they kept
account specific features of WSN such as addressing, duty the need for specific hardware.
cycling, upward routing, and data aggregation. Although the
work adopts OpenFlow, the authors only presented a conceptual Not all SDN solutions for WSN consider OpenFlow as
architecture. Moreover, some problems arise from this platform protocol, primarily because it fits high-end networks and
such as: the presence of several adaptations in the protocols, lack devices, and may not be suitable for resource-constrained WSN.
of proposals for backward compatibility and lack of any To this end, [13] suggested a general architecture with five
evaluations for the platform and no description of a scenario layers: physical, medium access control, network operating
where it can work. system, middleware, and application. The control logic runs in
the middleware layer so it can gather information from the entire
The study in [7] also gives some theoretical contributions network. An adjacency matrix represents the network topology,
while not provide any implementation details. The authors claim but the authors did not specify the SDN protocol nor the WSN
SDN-based WSN, which they call Software-Defined Sensor platform, and therefore, provided no evaluation about the impact
Networks (SDSN) or Programmable Sensor Networks (PSN), that such layered architecture may impose to a constrained
are different from wired networks once there is no sensible WSN.
change in hardware and communication channels.
The work of [14] presents an overview of the concepts that
Flow-Sensor [8] provides IaaS (Infrastructure as a Service) involve the use of the SDN paradigm in WSN. Also, they point
through virtualization of WSN aiming to make sensing services to some key technologies for implementing Software-Defined
available through the Internet. Despite adopting OpenFlow, Sensor Network (SDSN), including software-defined radio and
there is no specification on how they implemented or adapted it OTA-based reprogramming techniques. The authors propose a
for WSN. It is worth to mention there is neither comparison nor generic architecture of an SDSN and the description of a sensor
evaluation of energy consumption, memory utilization, and node prototype, necessary for the presented architecture to
processing overhead, some of the most important metrics for operate. Although it is a complete framework, it seems
WSN. burdensome to incorporate these functionalities into the
Work in [9] categorizes nodes as follows. Master Node runs commercially available sensors, as it may require the design of
the network routing strategy and controls the Center Node, new hardware for the motes. Even with this new hardware
which forwards the data packets inside the network and has no offering more processing power and benefits of the SDN
routing functions. They designate the other sensor as Normal

64
paradigm in several layers, there is a need to replace the entire tackle the requirements for a centralized controller view as well
WSN infrastructure, and that may impose economic issues. as to implement the exchange of messages between data and
control planes. AODV is a reactive routing protocol for ad hoc
Finally, [15] developed TinySDN, which is an SDN networks whose main features are a quick response to network
framework for WSN based on TinyOS, a widely accepted OS changes, low energy consumption, and low memory footprint.
for WSN. It is one of the few related works that specifies an AODV fits better to WSN with nodes demanding dynamic route
operating system. Also, the architecture has hardware requests. LQRP is a variant of AODV which adds the link
independence and presents the possibility of an extension based quality and the battery level of the nodes as routing criterion.
on multiple controllers. They implemented and tested the
solution using the COOJA network simulator. Despite, being a SDN may fit the requirements for being accepted as a
distributed controller solution, it did not describe any standard for IoT, once it is hardware and application
mechanisms to ensure that each of the controllers is aware of the independent and still demonstrates a step towards backward
status of the entire network. It is also not possible to identify any compatibility [2]. Figure 1 depicts the architecture of μSDN with
communication between controllers. Thus, each controller could three layers: the Data Plane, the Communication Channel, and
have only a partial view of the network so forcing the Control Plane.
programmer to make the changes in a decentralized way.
The authors in [2] extend the work of [15] by providing
relevant theoretical guidelines on how to integrate SDSN to IoT
through standardization of routing protocols, programming
interfaces, and backward compatibility. They also point out the
possibility of using Routing Protocol for Low power and Lossy
Networks (RPL), a widely known routing protocol for IoT, to
address some standardization issues.
Most of the proposals discussed in these related works are
limited to conceptual model, which is a recurrent weakness.
Many works did not provide a real implementation, and some of
those works are limited to simulation environments. Another
common drawback is the requirement for changes in the Fig. 1. Architecture of μSDN
hardware of the sensor nodes, either part of it or even the entire
network, to implement SDN in WSN. Table I summarizes the A. Communication channel
information about related works.
μSDN uses the signaling messages of AODV and LQRP to
TABLE I. RELATED WORKS create a communication channel between the controller and the
other network elements. Such mechanism also promotes
Implementation [15] communication between data and control planes, with almost no
impact on the performance of the network, and avoids the
Simulation [8]
requirement for a second network interface.
Modelling [8] [9] μSDN considers minor changes in the signaling procedure
Theoretical Contribution [4] [5] [11] [14] [7] [2] of those routing protocols, such as the modification of messages
and the definition of new types of messages. The modifications
OpenFlow [5] [8] [9] [11] [12] follow:
Non-OpenFlow [13] [14] [15] x Rule Request Message (RULE_REQ): SDN-enabled
Operating System and Platform [15] [12] nodes send this message to the controller-enabled node
to request a rule to communicate to another node, which
New-hardware Required [10] [14] [12] is not in the list of neighbors.
x Rule Request Response Message (RULE_RESP):
III. μSDN have different functions depending on the type of the
The μSDN is a WSN routing architecture which reuses the node. When the controller sends this kind of message, it
signaling procedure of existing protocols as a communication is the reply to RULE_REQ and contains the requested
channel between control and data planes. Routing control is rule, if it exists. On the other side, when a SDN-enabled
centralized, flexible and does not require any hardware node sends this message to another SDN-enabled node,
replacement or modification. μSDN requires only firmware it starts the neighbor discovery, and the receiver appends
modification at the controller to implement new routing policies the information about the sender to the neighboring list.
after small extensions to the software of all WSN nodes. This
way, the software of ordinary nodes remain unchanged x Link Quality Request Message (LQREQ): This type
regardless of future reprogramming needs for the entire network. of message works the same way as LQRP determines.
All sorts of nodes periodically exchange this kind of
μSDN extends AODV (Ad hoc On-demand Distance message so they can estimate the quality of the links
Vector) [16] and LQRP (Link-Quality Routing Protocol) [17] to between them.

65
x Link Quality Request Response Message (LQREP):
Just like LQREQ, messages of this type also remain
backward compatible. They are replies to LQREQ and
carry information used which helps estimate the quality
of the link between two nodes.
x Network View Update Message (NVUPDT): SDN-
enabled nodes send this type of message to the controller
so that it can update Network View. The message
contains the quality of the link between two nodes
(resulting from exchanging RULE_RESP messages) and
the value for the battery level of the node that is sending
the message.
B. Data Plane (a) Node B requests the controller for a route
towards node C
The data plane has two functions: query and request
Forwarding Rules. SDN-enabled nodes keep these rules in tables
and use them to forward the messages according to the logic
defined by the control plane at the controller. Table II portrays
the structure of the Forwarding Table. The column Orig (Origin
or Source) stores the address of the source of the packet and the
column Dest (Destination) stores the address of the final
destination. NextHop refers to the address of the neighbor to
which the SDN-enabled node should forward the packet so it can
reach its destination. Lifetime indicates the time interval (in ms)
that the rule should remain active and stored in the Forwarding
Table. The column TimeOut contains the timestamp (in ms)
after that the SDN-enabled node should remove the rule from
the Forwarding Table and corresponds to the sum of the time of (b) Controller responds to request and node
arrival and the value of the lifetime. At last, the rightmost C stores the rule in the Forwarding Table
column, Action, stores a code (integer) indicating an action to
be applied to the packet. Number 1 indicates that the SDN-
enabled node should forward the packet and number 2 indicates
it should discard the packet.

TABLE II. RULES TABLE

Orig Dest NextHop LifeTime TimeOut Action


TimeIn
5CF1 30C0 5CCF 30000 1
+LifeTime

C. Control Plane
The Controller represents the logical/control plane. Its role
is converting all forwarding decisions into rules, sending them (c) B sends data to A
to the data plane (i.e., sensor nodes) using the signaling
messages implemented as extensions of AODV and LQRP
protocols. The control plane also includes the Network View and
the Routing Algorithm set by the network manager. The Routing
Algorithm may or may not employ the Network View to set the
forwarding rules.
The Network View reproduces the status and topology of the
network along with some information, such as the connections
between nodes; the qualities of the links between nodes; and the
battery level of each node. Any routing algorithm set by the
Network Manager, which should determine the next hop in a
multi-hop network, may use the Network View. In the current
implementation, the Network View is a graph represented by an
adjacency matrix. Each row and each column header, in the (d) A forwards data to C
matrix, accounts for one of the nodes, while the values filling the
fields in the matrix represents the quality of the link between Fig. 2. Exchange of messages containing the rules in μSDN

66
nodes or even the lack of connection between them. In the static environment, they present important differences regarding
matrix, there is also a “battery vector” listing the battery level the lifetime of the rules and the interval between messages of
for all network nodes. neighbor discovering. Table III contains results from a
comparison between the versions of μSDN.
D. Routing Mechanisms
The algorithms, planes, and protocols previously described AODV and LQRP represented the legacy protocols for
compose the core routing mechanisms for both single hop and comparison. AODV is the native protocol of SunSpot platform
multi-hop communications. Furthermore, once the Network and uses the hop count as a criterion for routing. LQRP is a
Manager sets the routing protocol, the rules are not immutable version of AODV, which calculates the costs of routes and, in
and he or she can rewrite the protocol at any time only by addition to the number of hops, also considers the lifetime of
replacing the firmware of the Controller. Thus, the whole nodes and the link quality between them.
network will behave according to the new rules.
TABLE III. USDN CONFIGURATION PARAMETERS
Single-hop communication works similarly to regular
Interval Between
AODV and LQRP operations and results from the “Neighbor Lifetime of Rules Neighbors
Discovery” process. Periodically, each node broadcasts (ms) Discovery
RULE_RESP messages to the neighbors. In the message, the Protocol
Messages (ms)
fields “next hop” and “destination address” are the addresses of
the sender itself. All nodes that receive those messages store the μSDN v1 60 x 103 40 x 103
rule in their forwarding tables. Single-hop communications can μSDN v2 30 x 103 15 x 103
start as soon as there is enough information to fill the forwarding
tables of all nodes.
As soon as two nodes have information about each other IV. PERFORMANCE EVALUATION
after Neighbor Discovery, they exchange LQREQ and LQREP The performance assessment had two stages. The first stage
messages to estimate and store the quality of the link between aims to investigate the improvements that the reprogramming
them. After that, they can broadcast NVUPDT messages to the time generated in the network when network manager uses
Controller and update the Network View. μSDN. The second stage aims to determine how the signaling
A node requires multi-hop communications when the introduced by μSDN affects the performance of the network. For
application requires communication with another node that is this study, we used SunSpot1 platform which is a WSN platform
not in their communication range or for which there is no rule developed by Sun Microsystems. Its main feature is the
registered in the forwarding table. Thus, it needs to start the possibility to use Java as a programming language, which gives
process of obtaining a route, and then it sends a message of type it a high degree of abstraction and allows rapid development of
RULE_REQ to the Controller. As soon as the Controller applications. Although Sun has finished supporting the platform;
receives this message, it executes the Network Management some developers still use it for testing and research purposes.
Algorithm (NMA) that determines the best next hop from one The nodes have an ARM920T 180Mhz processor, 512KB of
node to another, based on the Network View, and replies with RAM, 4MB of storage, 3.7V and 720mAh lithium-ion batteries
RULE_RESP. With this feedback, the requesting node registers and 802.15.4 radios, which are widely adopted in Zigbee and
the new rule, so it can efficiently send data packets. Figure 2 6loWPAN networks.
displays an example of multi-hop communication where node B
wants to communicate with node C. We assumed a complete A. Reprogramming
setup for single hop communication and that the controller Both reprogramming methods, OTA and manual (through
(CTLD) has a complete network view. In step (Figure 2a) node USB interface), are carried out ten times. As depicted in Table
B sends a RULE_REQ message to controller soliciting the IV, reprogramming time when network manager uses OTA is
corresponding rule for the destination C. After that, in step approximately ten times greater than the reprogramming time
(Figure 2b), the controller replies to B with RULE_RESP, which due to a manual approach. The USB interface, used in the
contains the next hop that node B can use to reach C. In the manual reprogramming, provides more throughput than the
following steps, B sends data to A (Figure 2c), which is the wireless interface, used in the OTA reprogramming. The
intermediate node, and A forwards the data to C (Figure 2d). maximum throughput provided by 802.15.4 wireless interface is
E. Versions 250 kbps [18] while the USB interface can achieve transmission
throughput up to 480 Mbps in the 2.0 specification [19].
We developed two versions of μSDN to help validate our
proposal against legacy protocols. μSDNv1 is the first version The rightmost column of Table IV shows an estimate of
of the solution created in this work and uses pre-programmed reprogramming times for a network with five nodes, so when
static rules as a criterion for routing, whose are specifically set network manager uses μSDN, they can significantly reduce the
up to this test scenario. μSDN v2 is a dynamic version which reprogramming time. With μSDN, the reprogramming is
uses an algorithm based on the Network View considering the approximately 22% of the time spent by the manual
battery level of nodes to determine the next hop for packet reprogramming of the entire network, and roughly 2% of time
forwarding. Although both versions may work similarly in a spent by reprogramming via OTA. The reduction happens

1
http://www.sunspotdev.org/

67
because μSDN allows the modification of the entire network 1) Transmitter
routing through the reprogramming of a single node, which The performance indicators at the transmitter are energy
plays the role of Controller. consumption and the number of sent packets.

TABLE IV. REPROGRAMMING TIME A NETWORK WITH FIVE NODES Figure 4 compares four protocols according to the average
values for energy consumption and the number of packets sent.
Number Individual Total Figure 4a displays that the protocols which induce the highest
Reprogramming
of Time Reprogramming utilization are μSDN v2 and AODV, consuming respectively
Method
Updates Rescheduling Time about 11% and 15% more energy than the lowest energy
USB 5 44,3 s 221,5 s consumption protocol, which was the LQRP.
OTA 5 431 s 2155 s μSDN v1 consumes less battery than the μSDN v2 due to
reduced frequency of updates in the Network View, which
μSDN 1 44,3 s 44,3 s
reduces the signaling traffic. One can notice that the average
consumption remains close to all other experiments and
B. Signaling and Network Performance indicates that the energy consumed by μSDN remains close to
At this stage, the evaluation focuses on the ability of μSDN AODV. With regard the number of packets sent, displayed in
to keep the network performance similar to the traditional Figure 4b, all routing solutions generated an average of 290
protocols. That may help to establish it as a viable solution, so packets sent during the test period.
we proposed a hypothetical scenario for the evaluation. The
scenario has four nodes and a Base Station, which may refer to
a smart home application with light, temperature and presence
sensor. It is worth mention that the sensors are not transmitting
data all the time and they may work as routers and form multi-
hop networks. One of the nodes (transmitter) gathers
information about temperature and lighting and sends it to a
second node (receiver). The other two nodes are routers between
the transmitter and the receiver, while the base station is the
Controller.
The nodes stand on a flat surface, according to Figure 3, and
have had their transmission powers adjusted so the transmitter (a) Energy consumption
would not be able to transmit packets directly to the receiver,
forcing the communication over multiple hops. The application
running on the nodes sends approximately 30 messages per
minute to the receiver.

(b) Packets Sent

Fig. 4. Results in the Transmitter

2) Receiver
In the Receiver, the performance indicators are energy
consumption and the number of messages received.
Fig. 3. Evaluation Scenario The results, depicted in Figure 5a, show that AODV caused
the highest average energy consumption, but still, the values the
The results of the performance evaluations will be presented consumption of all protocols remain close to each other. The
depending on the roles played by the nodes. We evaluated maximum difference among them is only 0,015 mAH,
packet transmission in the transmitter; forwarding operation in indicating that μSDN v2 does not cause a significant increase in
the intermediate node; and message reception in the receiver. energy consumption and it is only slightly greater than LQRP,
The average of metrics considers a 95% confidence interval and which does not require extra signaling for communicating to a
there is no overlapping between compared metrics. controller entity as SDN proposal does. Note that μSDN v1
reaches the lowest energy consumption.

68
Regarding the number of received packets, AODV, μSDN The number of packets forwarded is a metric present only in
v1and μSDN v2 got the same results, while LQRP protocol intermediate nodes. This metric, depicted in Figure 6b, considers
achieved a performance slightly below the others (receives more the number of packets that the node could forward during each
messages). Figure 5b displays these results. The similarities in test. In this aspect μSDN v2 had the best performance, with an
the number of packets received indicate that even using the average of 301 packets forwarded, while μSDN v1 allowed
μSDN, the performance of this metric remained similar to that forwarding of 290 packets, LQRP forwarded 284 packets, and
of other protocols. AODV forwarded 278.7. The values, result from the capacity
that the controller has to determine the next hop in both versions
The results, especially regarding energy consumption, of μSDN, that is, the control plane centralization, which allows
indicate the feasibility of using the static μSDN forwarding rules faster convergence and route set up for μSDN versions once
in scenarios where the network tends not to have a dynamic network view is available, thus increasing number of forwarded
behavior. At the same time, it is possible to conclude that data packets.
dynamic version of μSDN can achieve better results than
traditional hop-only based protocols (AODV), as well as quite As the total number of packets sent and energy consumption
similar gains to those of energy-aware protocols, such as LQRP. separately analyzed cannot determine the protocol efficiency, it
was necessary to establish a relationship between them to
ascertain the effectiveness of the protocols. The following
equation provides such value:
NFP
EE =
EC,

In the equation, EE is the Energy Efficiency, NFP is the


Number of Forwarded Packets, and EC is the Energy
Consumption (in mAh).
Figure 6c demonstrates that the results regarding energy
efficiency are similar, just like other metrics. However, even
(a) Energy consumption
with a small advantage, μSDN v2 is the protocol that performed
best in this metric, at a rate of approximately 61 packets
forwarded by mAh consumed.
V. CONCLUSIONS AND FUTURE WORK
This paper presented an original proposal, as we developed
a new routing architecture, based on SDN. Such architecture
allows the separation between the Data Plane and the Control
Plane without causing significant impact to the network and yet
keeping the legacy hardware. The implementation also
demonstrates important steps towards complete backward
compatibility once we have to reuse some types of messages
(b) Packets Received from legacy protocols.
The performance evaluation demonstrated that, with μSDN,
Fig. 5. Results in the receiver
the energy consumption remained at levels close to traditional
protocols with low power consumption, despite an increase in
3) Intermediate overhead of signaling compared to AODV and LQRP. That
The Intermediate nodes are the ones enabling multi-hop allows us to state that the μSDN is a viable solution by which
communications. Therefore, nodes of this type suffer more the network manager can modify routing policies throughout the
impact regarding the signaling procedure. The main network only by changing the firmware of the Controller. Thus
performance indicators for the intermediate node are energy it is possible to reduce the total amount of unnecessary reinstalls,
consumption; the number of forwarded packets; and the and yet the total time to reprogram the network.
relationship between the number of sent packets and the required
energy. Future works may follow some strands regarding possible
improvements and further evaluations. We still can investigate
Figure 6a demonstrates a small difference in energy strategies for μSDN reuse more messages from legacy protocols
consumption among the protocols. The results assert that μSDN by using optional headers; create an API that allows the
can deliver the benefits SDN promises and yet keeps modification of routing policies at the application level; expand
consumption close to those observed in AODV and LQRP the information in the Network View; include support for
results even in intermediate nodes. Both LQRP and μSDN virtualization of WSN and multiple controllers. Further
versions require signaling for link quality discovery, while evaluation may include scalability and resilience of μSDN; and
AODV only relies on simple packet forwarding, what explains the construction of the Network View regarding convergence
its slightly better performance at the intermediate node.

69
time and the number of required messages; the reduction of networks," ACM SIGCOMM Computer Communication Review, 38(2),
overload inflicted to the network. pp. 69-74, 2008.
[4] S. Costanzo, L. Galluccio, G. Morabito and S. Palazzo, "Software
defined wireless networks: Unbridling sdns," 2012 European Workshop
on Software Defined Networking (EWSDN), pp. 1-6, October 2012.
[5] T. Luo, H. Tan and T. Quek, "Sensor OpenFlow: Enabling software-
defined wireless sensor networks," Communications Letters, IEEE,
16(11), pp. 1896-1899.
[6] K. M. Modieginyane, B. B. Letswamotse, R. Malekian and A. M. Abu-
Mahfouz, "Software defined wireless sensor networks application
opportunities for efficient network management: A survey.," Computers
& Electrical Engineering, 2017.
[7] Y. Choi, Y. Choi and Y. G. Hong, "Study on coupling of software-
defined networking and wireless sensor networks," Eighth International
Conference on Ubiquitous and Future Networks (ICUFN), 2016. IEEE,
(a) Energy consumption pp. 900-902, 2016.
[8] A. Mahmud, R. Rahmani and T. Kanter, "Deployment of Flow-Sensors
in Internet of Things' Virtualization via OpenFlow," Third FTRA
International Conference on Mobile, Ubiquitous, and Intelligent
Computing (MUSIC), 2012. IEEE, pp. 195-200, June 2012.
[9] Z. Han and W. Ren, "A Novel Wireless Sensor Networks Structure
Based on the SDN," International Journal of Distributed Sensor
Networks, 2014.
[10] A. Yuan, H. Fang and Q. Wu, "OpenFlow based hybrid routing in
Wireless Sensor Networks," 2014 IEEE Ninth International Conference
on Intelligent Sensors, Sensor Networks and Information Processing
(ISSNIP), pp. 1-5, April 2014.
[11] S. Shanmugapriya and M. Shivakumar, "Context Based Route Model
for Policy Based Routing in WSN using SDN approach," BGSIT
National Conference on Emerging Trends in Electronics and
(b) Packets Forwarded
Communication, 2015, 2015.
[12] L. Galluccio, S. Milardo, G. Morabito and S. Palazzo, "SDN-WISE:
Design, prototyping, and experimentation of a stateful SDN solution for
WIreless SEnsor network," IEEE Conference on Computer
Communications (INFOCOM), p. 513–521, 2015.
[13] A. De Gante, M. Aslan and A. Matrawy, "Smart wireless sensor
network management based on software-defined networking," 27th
Biennial Symposium on Communications (QBSC), pp. 71-75, 2014.
[14] D. Zeng, T. Miyazaki, S. Guo, T. Tsukahara, J. Kitamichi and T.
Hayashi, "Evolution of software-defined sensor networks," IEEE Ninth
International Conference on Mobile Ad-hoc and Sensor Networks
(MSN), pp. 410-413, 2013.
[15] B. T. de Oliveira, L. B. Gabriel and C. B. Margi, "Tinysdn: enabling
multiple controllers for software-defined wireless sensor networks,"
(c) Efficiency
IEEE Latin America Transactions, 13(11), pp. 3690-3696, 2015.
Fig. 6. Results at the Intermediate node [16] C. Perkins, E. Belding-Royer and S. Das, "Ad hoc on-demand distance
vector (AODV) routing," RFC 3561, 2003.
REFERENCES [17] A. Floros, M. Zennaro and B. & Pehrson, "Routing Protocols for SUN
Spots: Analysis, Comparison and Implementation of new Protocols,"
Skolan för information-och kommunikationsteknik, Kungliga Tekniska
[1] L. Atzori, A. Iera and G. & Morabito, "The Internet of things: A högskolan., 2009.
survey," Computer networks, 54 (15), pp. 2787-2805, 2010. [18] J. Zheng and M. J. Lee, "Will IEEE 802.15. 4 make ubiquitous
[2] B. T. de Oliveira, R. C. A. Alves and C. B. Margi, " Software-defined networking a reality?: a discussion on a potential low power, low bit
Wireless Sensor Networks and Internet of Things standardization rate standard," IEEE Communications Magazine, 42(6), pp. 140-146,
synergism," EEE Conference on Standards for Communications and 2004.
Networking (CSCN), pp. 60-65, 2015. [19] H. Wang, "USB Connector". Patent US Patent No. 7,155,545, 2006.
[3] N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson,
J. Rexford and J. Turner, "OpenFlow: enabling innovation in campus

70

You might also like