EDGE CONTROLLER PLACEMENT FOR NEXT GENERATION WIRELESS SENSOR NETWORKS

International Journal of Computer Science & Information Technology (IJCSIT) Vol 14, No 5, October 2022
DOI: 10.5121/ijcsit.2022.14502 17
EDGE CONTROLLER PLACEMENT FOR NEXT
GENERATION WIRELESS SENSOR NETWORKS
Mohamed S. Abouzeid, Heba A. El-khobby,
Mahmoud A. A. Ali and Mohamed E. Nasr
Electronics and Electrical Communication Department,
Faculty of Engineering-Tanta University, Tanta, 31527, Egypt
ABSTRACT
Nowadays, Fog architecture or Edge architecture is becoming a popular research trend to distribute a
substantial amount of computing resources, data processing and resource management at the extreme edge
of the wireless sensor networks (WSNs). Industrial communication is a research track in next generation
wireless sensor networks for the fourth revolution in the industrial process. Adopting fog architecture into
Industrial communication systems is a promising technology within sensor networks architecture. With
Software Defined Network (SDN) architecture, in this paper, we address edge controller placement as an
optimization problem with the objective of more robustness while minimizing the delay of network
management and the associated synchronization overhead. The optimization problem is provided and
modelled as submodular function. Two algorithms are provided to find the optimal solution using a real
wireless network to get more realistic results. Greedy Algorithm and Connectivity Ranking Algorithm are
provided. Greedy algorithm outperforms connectivity ranking algorithm to find the optimum balance
between the different metrics. Also, based on the network operator preference, the number of edge
controllers to be placed will be provided. This research paper plays a great role in standardization of
softwarization into Industrial communication systems for next generation wireless sensor networks.
KEYWORDS
Fog Architecture, Submodularity, Software Defined Network, Controller Placement, Virtual Process
Function.
1. INTRODUCTION
Recently, for next generation wireless sensor networks, there are critical requirements such as
increasing the operational efficiency of industrial control process. In addition, improving the
effectiveness of operational activities is an increasing demand for various industries today.
Furthermore, more flexible operations should be supported with the reduction of capital
expenditure [1]. Thus, softwarization is a key enabler to achieve these facilities. More
specifically, Software Defined Network (SDN) is a promising architecture for next generation
wireless sensor networks [2]. Healthcare systems and the automotive industry are simple
examples of industrial communication systems in WSNs. Legacy industrial control process
includes real time data collection form sensors, after that data processing is performed in
hardware controllers and finally execution of control commands through actuators [3-5]. With
SDN, softwarization can embrace the control process in an entity. Thus, the hardware controllers
can be replaced by software instances [6]. There are some advantages of SDN. One of them is
that the control process function can be placed in a commodity server which can be flexibly
provisioned on demand [7]. In addition, they have a good ability of upgrading in amore simple
way than hardware controller.

18
With the introduction of Mobile Edge Computing (MEC), next generation wireless sensor
networks can distribute substantial amount of computing and storage resources to edge nodes at
the extremes of the network with low latency and high bandwidth [8]. The critical aspects for
adoption of MEC into next generation wireless sensor networks are the robustness of the wireless
communication links, the minimized delay of network management and the associated overhead
not only between controllers and data-plane nodes, but also, between the controllers themselves
(Inter-controller communication) [9]. One way to improve the reliability of next generation
wireless sensor networks is to place the virtual process function (VPF) of the controller on
another edge node in case of node or link failure [10]. However, this process is limited because of
the resource constraints of the edge nodes besides their limited power and storage resources [11].
In this paper, we address the problem of placing edge controllers in next generation wireless
sensor networks with the objective of minimizing the network management delay, overhead
control messages and invalid control paths which increases the robustness of the system. We
formulate an optimization problem for placing the virtual process function into the edge nodes
and taking into consideration the questions: How many controllers should be placed at the edge
of the network in close proximity to the end sensors and actuators. Also, where exactly should be
the controllers placed to improve the resilient of the system. The objective is to minimize the
capital expenditure of next generation wireless sensor networks, while increasing its flexibility.
2. RELATED WORK
There has been a signiﬁcant interest in cloud computing for industrial use cases, both concerning
potential application areas and security aspects [12], [13], but the focus of these works is on the
architectures and requirements, rather than resource management.
There are recent works on resource allocation in MEC for sensor networks [14]. The problem of
allocating visual sensors with correlated measurements to computing resources to maximize
system capacity is introduced in [8]. In [15], the authors propose the problem of allocating health
sensors to health cloud servers to maximize system utility. The authors in [16] modelled the
individual Virtual Network Function (VNF) placement problem as a generalized assignment
problem where the controllers assigned to different nodes to minimize the total assignment cost.
The placement of chains of VNF is modelled in [17] as an Integer Linear Problem (ILP). The
problem of controller placement to maximize the resilience under link or node failure is
introduced as ILP and numerical results are provided in [18].
To the best of our knowledge, our work is the first work to handle the edge controller placement
problem as an optimization problem for industrial use case. The optimization problem considers
resilience constraints besides the delay of network management and the associated overhead to
achieve the optimal placement of the controller in SDN industrial communication system for next
generation wireless sensor networks. This research work considers minimizing the delay and
control overhead which plays a great role in minimizing the operating expenses. This work also
reduces the down time of the network by maximizing the reliability performance of WSNs. The
novelty of our work appears by applying submodularity conditions to the optimization
performance of next generation wireless sensor networks.

19
3. SYSTEM MODEL
Figure 1. The communication infrastructure consists of BSs, MEC nodes and sensorsand backhaul network.
We consider a system that consists of a set of B Base stations (BSs). A subset N node are
equipped with computational and storage resources and serves as edge nodes. We denote by K a
subset of N that can host control process function such that K  N .The b
ase stations are
interconnected by a backhaul network. Sensors and Internet of Things (IoT) devices
communicate wirelessly within the base station. The failure of an edge node or the wireless
communication link result in failure scenario. This failure leads to that the edge node becomes
unsuitable to host the virtual process function (VPF) [19]. We assume that the network operator
is able to estimate the occurrence probability of each failure scenario. We denote by πl the failure
occurrence probability. It is assumed that Generalized pareto distribution is used to estimate such
failure occurrence probability. Generalized pareto distribution is used because it provides
Independent Identical Distributed (IID) probability for failure scenarios.
Each pair of controllers are communicated with each other, for synchronization purposes, using
messages. These messages are exchanged at a constant rate besides other exchanged at a rate that
depends on the controller load. When considering SDN enabled edge network, the interaction
between a controller and data-plane nodes takes place through Openflow protocol. Heartbeat
messages are periodically transferred between the controller and data plane nodes, besides static
request/reply messages. Thus, as the network scales up, the overhead grows. For synchronization
purposes between a cluster of controllers, there is a significant overhead that should be
considered in modelling the edge controller placement problem for industrial use cases.
4. PROBLEM FORMULATION
In this section, we formulate the problem of edge controller placement as an optimization
problem. We consider a network of a diverse set of N  B . This subset supports edge nodes
which can host Virtual Process Function (VPF) an act as an edge controller. Not all edge nodes
are active all the time, thus, we assume that K  N active per time. This results in that K edge
nodes generate PacketIN messages with arrival rate  packets/time. The probability of the
generated traffic can be estimated by . Thus, the expected value of the
generated traffic is where vk {0,1}is a decision binary optimization
variable to denote whether the edge node is available or not. This availability is justified by
enough storage capacity to save virtual process image and enough power for execution.
One of the important metrics to be considered in edge controller placement for industrial use
cases is the reliability of the communication system. The network may fail due to the wireless
link goes down or the component failure. We consider basically a set of L failure scenarios based

20
zk
zk
on the valid control paths, where the control paths are the logical links between the controllers or
between the edge nodes and the controllers L  {1, 2, , L}. We denote by l the failure
probability of each link by using generalized pareto distribution as it produces Independent
Identically Distributed (IID) scenarios. Let us denote by hnk {0,1} a binary optimization
variable that denotes whether a control path between edge node n and a controller at k exists or
not. Hence, the expected value of invalid control paths
We consider a binary optimization variable y nk {0,1}to denote whether node n is assigned to a
controller at node k ( y nk = 1) or not ( y nk = 0 ). Hence, the assignment policy can be expressed
as
The assignment of an edge node to a controller induces a cost. This cost increases with the
topological distance. We denote dnk  0 is the delay (in millisecond) when node n is assigned
to a controller at node k. The total assignment cost is expressed as follows
The controllers exchange also, messages between them for synchronization purposes. There are
two types of inter-controller associated overhead. One type of them is that messages are being
exchanged at a constant rate. The other type are messages exchanged at a rate depends on the
controller traffic load. w Const
 o denote messages exchanged at constant rate while
w dep
 o denotes controllers load dependent messages. These messages exchanged
between controllers placed at node z and node k. We consider also x k
as a binary
optimization variable to denote whether a controller placed at node k ( x k 1) or not ( x
k 0). Thus, the placement policy
The total overhead synchronization costs between a pair of controllers at node j and node k is
given as follows
It should be considered that each node n should be assigned to only one controller such that

21
Also, the controller can be placed at node k if and only if the sum of computing resource
requirement does not exceed the generated traffic from node k.
In addition, the controller should be placed at node k, so that node n can be assigned to it, such
that
Based on the network operator preferences, the network operator can perform trade-off
betweenthe different metrics. A weight value   0 is used to balance the optimization
problem.
By increasing  , more priority is given to the synchronization overhead cost. The edge
controller placement problem can be expressed as follows
The above optimization problem is a challenge problem as it contains discrete variables with
objective function with quadratic and cubic terms. This problem is NP-Hard problem, as the edge
controller placement depends on the topological distance between the controllers.
5. OPTIMIZATION ALGORITHMS
We begin by showing that for a given controller placement x, the optimal assignment
policy y can be found. Denote x k is a controller placed at node k. The set of all possible
locations for the controller, also called the ground set G
A subset of elements X  G corresponds to a controller placement policy X. Let X n the binary
representation of the set of controllers positions such that X  (x 1, x 2 ,..., x k ) , the objective
function f can be expressed as a set function such that
For a given controller placement at X, the optimal assignment is y. This leads us to a definition
of a set of functions. These functions called submodular function [20].
Definition: let G is the ground set, a set function is said to be submodular
function
if and only if, for two subsets A , BG and A  B and every element that, i G B , it holds
that,

22
which means adding an element to a smaller set, resulting in that the respective gain expands.
This is called the marginal value [21]. The marginal value increases with the smaller set
function [22].
We will show in the appendix that the objective function f(x) under certain conditions on the cost
value is a submodular function As shown from the appendix that the total balanced cost function
is submodular, we use two optimization algorithms to balance the different metrics and to
minimize the overall cost. One algorithm is a Greedy algorithm. The algorithm proceeds in K
iterations which corresponds to arbitrary order r1 ,r2, …..,rk of the ground set G. At each
iteration, two solutions are maintained A and B. Also, the reliability of valid control paths is
checked. Initially, A is assigned to 0 and B is assigned to G. At Kth iteration, the algorithm is
either adds rk to A or removes it from B. This decision is done randomly and greedily based on
the marginal value of each of the two options. The two solutions coincide and A=B after Kth
iterations. Thus, the algorithm returns the optimal placement correspond to the minimum cost
value. With that algorithm, we get a solution to the edge controller placement problem with an
approximation that w dep is constant and identical for all cases The other algorithm is the
Connectivity Ranking Algorithm (CRA) for comparison purposes of the obtained results [23].
The connectivity ranking algorithm sort the nodes in descending order according to their
connectivity. The idea is to repeatedly place a new controller at one of the k edge nodes and
calculate the total cost Jtot until the next cost is higher than the current one. Then, we choose the
current placement as the optimal one.

23
6. EVALUATION RESULTS
In this section, we provide the results by running the proposed algorithms using real wireless
network topology on Matlab software. We use MANIAC mobile ad hoc network in [24]. The
network contains 14 nodes which allows us to execute the proposed algorithms in a reasonable
time. We define the delay cost as the aggregate delay of the links of the valid shortest path. We
set the delay of each link with average value 12.2 msec. We set the average probability of failure
is 0.15 and the probability of estimated traffic is 0.65. We set w w const
 0.2*hops and w dep

0.6*hops, where hops indicate the number of hops of the shortest path between the respective
nodes.
Figure 2 shows the relationship between the weight value  and the total balanced cost Jtot. As
shown form the figure, the greedy algorithm outperforms connectivity ranking algorithm (CRA)
along the values of  . Furthermore, the performance of Greedy algorithm exceeds connectivity
ranking algorithm at larger value ofwhere more priority is given to the delay and reliability than

24
overhead synchronization cost. Figure 3 shows the number of controllers needed to be placed to
optimize the controller placement problem. For lower values of  where minimum delay is
preferred, the number of edge controllers is high to minimize the delay cost between data-plane
nodes and edge controllers. While the number of edge controllers goes down when the
synchronization overhead cost is preferred with large values of  , so that inter
-controller
communication is reduced.
Figure 2. The total balanced cost against Beta range
Figure 3. The number of controllers against Beta range
7. LIMITATIONS AND FUTURE WORK
The aforementioned work did not take into consideration the shape of traffic overheads. In the
real world, different nodes generate diverse types of traffic that must be considered effectively.
Also, the configured network has to support SDN configuration to be able to apply the
optimization formula and submodularity conditions. In addition, the delay of each link isassumed
to be constant which is not the case in the real time for heterogeneous wireless sensor networks.
In the future, we plan to investigate different delays for disparate links. Additional mechanisms
of forming a controller will be analysed. We will consider the interference and the congestion of
the links in placing the edge controller in next generation WSNs.
8. CONCLUSIONS
In this paper, we addressed the edge controller placement problem for next generation wireless
sensor networks. The problem was formulated as an optimization problem which is submodular
function. We used Greedy algorithm and Connectivity Ranking Algorithm (CRA) for findingthe
optimal solution for the balanced problem. The evaluation results were shown and provided that
Greedy algorithm outperforms Connectivity Ranking Algorithm (CRA) along the value of weight
value β. This paper provides an interesting optimization formula for industrial use cases that
helps significantly in standardization of next generation wireless sensor networks.

25
REFERENCES
[1] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, “Edge Computing: Vision and Challenges”, IEEE
Internet of Things Journal, vol. 3, no. 5, 2016.
[2] D. Kreutz, F. Ramos, P. Esteves Verissimo, C. Esteve Rothenberg, S. Azodolmolky, S. Uhlig,
“Software-Defined Networking: A Comprehensive Survey”, in Proc. of the IEEE, vol. 103, no. 1, pp.
14-76, 2015.
[3] M. Peng, S. Yan, K. Zhang, C. Wang, “Fog Computing based Radio Access Networks: Issues and
Challenges”, IEEE Network, vol. 30, no. 4, pp. 46-53, 2016.
[4] I. S. Acharyya and A. Al-Anbuky, “Towards wireless sensor network softwarization,” in Proc. of
IEEE Conference on Network Softwarization, 2016, pp. 378–383.
[5] C. Alippi, R. Fantacci, D. Marabissi, and M. Roveri, “A cloud to the ground: The new frontier of
intelligent and autonomous networks of things,” IEEE Comm. Mag., vol. 54, no. 12, pp. 14–20, 2016.
[6] A. Singh “A survey and classification of controller placement problem in SDN” International Journal
of Network Management, 2018.
[7] H. Xu, Z. Yu, C. Qian, X. Li, Z. Liu, “Minimizing Flow Statistics Collection Cost of SDN Using
Wildcard Requests”, in Proc. IEEE Infocom, 2017.
[8] E. Emil, G. Dn, and F. Viktoria, “Radio and computational resource management for fog computing
enabled wireless camera netoworks,” in Proc. of IEEE GlobeCoWorkshop on Internet of Everything,
2016, pp.1–6.
[9] A. Muqaddas, A. Bianco, P. Giaccone, G. Maier, “Inter-controller Traffic in ONOS Clusters for SDN
Networks”, in Proc. IEEE ICC, 2016.
[10] Zhao, P., & Dán, G. “Resilient placement of virtual process control functions in mobile edge clouds”
in IFIP networking conference and workshops, 2017.
[11] R. Cohen, L. Lewin-Eytan, J. S. Naor, and D. Raz, “Near optimal placement of virtual network
functions,” in Proc. of IEEE INFOCOM, 2015, pp. 1346–1354.
[12] M. Yigit, V. C. Gungor, and S. Baktir, “Cloud computing for smart grid applications,” Computer
Networks, vol. 70, pp. 312–329, 2014.
[13] G. Dn, R. B. Bobba, G. Gross, and R. H. Campbell, “Cloud computing for the power grid: from
service composition to assured clouds,” in Proc. of Usenix Workshop on Hot Topics in Cloud
Computing (HotCloud), 2013.
[14] A. Sallahi, M. St-Hilaire, “Optimal Model for the Controller Placement Problem in Software Defined
Networks”, IEEE Communications Letters, vol. 19, no.1, pp. 30-33, 2015.
[15] S. Das, S. Misra, M. Khatua, and J. J. Rodrigues, “Mapping of sensor nodes with servers in a mobile
health-cloud environment,” in Proc. Of IEEE International Conference on e-Health Networking,
Applications & Services (Healthcom), 2013, pp. 481–485.
[16] M. Ghaznavi, A. Khan, N. Shahriar, K. Alsubhi, R. Ahmed, and R. Boutaba, “Elastic virtual network
function placement,” in Proc. Of IEEE CloudNet, 2015, pp. 255–260.
[17] H. Moens and F. De Turck, “VNF-P: A model for efficient placement of virtualized network
functions,” in Proc. of IEEE International Conference on Network and Service Management
(CNSM), 2014, pp. 418–423.
[18] M. Xia, M. Shirazipour, Y. Zhang, H. Green, and A. Takacs, “Network function placement for NFV
chaining in packet/optical datacenters,” Journal of Lightwave Technology, vol. 33, no. 8, pp. 1565–
1570, 2015.
[19] A. Hmaity, M. Savi, F. Musumeci, M. Tornatore, and A. Pattavina, “Virtual network function
placement for resilient service chain provisioning,” in Proc. of IEEE International Workshop on
Resilient Networks Design and Modeling (RNDM), 2016, pp. 245–252.
[20] U. Feige, V. Mirrokni, J. Vondrak, “Maximizing Non-monotone Submodular Functions”, in Proc.
IEEE FOCS, 2011.
[21] N. Buchbinder, M. Feldman, J. Naor, R. Schwartz, “A Tight Linear Time (1/2)- Approximation for
Unconstrained Submodular Maximization”, in Proc. IEEE FOCS, 2012.
[22] S.O. Gharan, J. Vondrak, “Submodular Maximization by Simulated Annealing”, in Proc.
ACM/SIAM SODA, 2011.
[23] Z. Su, M. Hamdi “MDCP: Measurement-aware Distributed Controller Placement for Software
Defined Networks” IEEE 21st International Conference on Parallel and Distributed Systems, 2015.

26
[24] A. Hilal, J.N. Chattha, V. Srivastava, M.S. Thompson, A.B. MacKenzie, L.A. DaSilva, P. Saraswati,
CRAWDAD dataset vt/maniac: http://crawdad.org/vt/maniac/20081101,
https://doi.org/10.15783/C78W2V
APPENDIX
Based on that the weighted sum of submodular functions is also submodular function.
Here, Js _Const and Js _dep denote the constant and the load dependent synchronization
costs,respectively. JR
denote the expected value of invalid control paths as the reliability metric.
For a given node k K , load cost. Ja denote the total assignment cost that includes
Let us consider two placement sets A and B where A  B  G , where G is the ground set that
contains all possible values of controller placement. We add an element x k G B to both
placement sets.
1) For the function Js _Const and the placement set A, the marginal value of adding a
controller at node m, x m
If we replace A with placement set B, the marginal value decrease. Hence, the above
function is submodular function.
2) For the invalid control paths function, by adding invalid control path between node
n and m to the function, the marginal value decrease with placement set B.
Hence, the function is submodular.
3) For the assignment function, with the placement set A, where node n is assigned to
controller at node m the marginal value is

27
Thus, the marginal value decrease if we replace A with the placement set B.
4) For the function Js _dep , with placement set A, the marginal value of adding a controller
at node m, x m is w dep
which is independent of the assignment policy with the assumption of that
its value is constant. So, the function is also submodular.

EDGE CONTROLLER PLACEMENT FOR NEXT GENERATION WIRELESS SENSOR NETWORKS

More Related Content

Similar to EDGE CONTROLLER PLACEMENT FOR NEXT GENERATION WIRELESS SENSOR NETWORKS (20)

Recently uploaded (20)

EDGE CONTROLLER PLACEMENT FOR NEXT GENERATION WIRELESS SENSOR NETWORKS