Energy-efficient Sensors in Data Centers for Industrial Internet of Things (IIoT)

Full text

978-1-5090-6785-5/18/$31.00 © 2018 by IEEE
Energy-efficient Sensors in Data Centers for
Industrial Internet of Things (IIoT)
Williams Paul Nwadiugwu
Networked Systems Lab., School of Electronics Engineering
Kumoh National Institute of Technology,
Gumi, Gyeongsangbuk, South Korea
williams.nwa@kumoh.ac.kr
Dong-Seong Kim, Senior Member, IEEE
Networked Systems Lab., School of Electronics Engineering
Kumoh National Institute of Technology
Gumi, Gyeongsangbuk, South Korea
dskim@kumoh.ac.kr
Abstract— The various works on the corresponding and the
associated need for a wireless industrial sensor network
connectivity, applicable in the Industrial Internet of things (IIoT)
domain have been exhaustive quite recently. This study has also
clearly shown how they are grouped differently to that which is
required by the consumer industries, with much emphasis on
energy consumption and control, throughput reliability, stability
of the sensors and their embedded security characteristics. To
accomplish a deserving Quality of Service (QoS) of IIoT to which
is expected, the data centers must have a guaranteed
performance as a backbone of computation resources. This paper
implements and investigates energy efficient sensors in the
computing resources of data center for Industrial Internet of
Things (IIoT). The simulation works investigates the resulting
throughput performance and energy consumption profiles and
control of the sensor nodes with respect to the three-tier network
architecture as one of candidate backbone in IIoT.
Keywords— Data center; industrial internet of things (IIoT);
Low power wireless sensor nodes; three-tier architecture Data
center network backbone.
I.
I
NTRODUCTION
The work by [1] referred data center platforms as those
enormous and precarious large system units or machines that
are not constraint by time [1]. This level of data center’s
criticality was informed by two major criteria. The growing
and consistent demand for data computing resources, data
operations and eventual data filing and racking [2] by
industries such as social media, internet network providers or
software platforms with enormous server requirements,
incidentally becomes is the first major criteria, while the
second major criteria is the numerous modularity of
independent system platforms and their ability to interconnect
with one another, even for the very short-timed applications
[1].
It will be quite conclusive to state here that the dominance
of numerous research studies and practical implementation
approach in the area of Industrial Internet of Things (IIoT)
have been necessitated by the constant demands for low power
sensor units and low system processors, the quest for Big data
analytic, and the ever-growing presence of augmented and
intelligent wireless network systems, to have the system
processors and network sensors to stay interconnected. These
interconnected technologies allow for easier information
retrieval and processing techniques whenever there is the
presence of low-powered and durable sensors. An often
thought out ideology of equipping the physical matter or thing
such as industrial appliances and machines, oil-rigs and
pipelines, avionics and the entire transportation technologies,
with low powered sensors is no-longer novel, since these
sensors are already major players in different capacities such
as the mission-critical sectors such as avionics, oil
explorations, traffic and transportation management control
systems, to a more value-added consumables like the health,
beauty and the manufacturing industries, within the last
decade.
It is rather historical to denote here the fact that all
industrial operations technologies have yielded an appreciable
standard for maintaining a more reliable and secured network
which cannot be said of the consumer technologies that yields
a lower standard when compared. These high standards are
now used as a yard-stick often for business-critical industrial
internet of things applications. The engineers are now able to
tell through various scientific computations whether these
sensors are safe, low-cost and deplorable even in often
difficult or environmentally challenging locations, which are
common scenarios for industrial applications [3].
As part of our contributions to promote viable
development of this area of research, we have:
• Channeled our thoughts into the investigation,
detailed and holistic analysis of a three-tier data
center system architecture as a candidate backbone,
in the context of energy-efficient sensor nodes for
IIoT.
• Implemented a prototype energy-efficient sensor
node which may facilitate and ultimately reduce the
energy cost and go on to implement a flexible data
center architecture.
• Analyzed the resulting throughput and energy effects
of the sensors nodes with respect to the three-tier
IIoT architecture backbone of a data center
computing resources.
The remainder of our works here were exhausted as follows:
A preliminary study and problem statement of the work is
presented in section II. Then a system model, with emphasis

978-1-5090-6785-5/18/$31.00 © 2018 by IEEE
on the data center network topology and the practical
implementation of our works were exhausted in section III,
while section IV focuses on the performance evaluation with
analysis of both the testbed assessments scheme and the
simulation methodology. A conclusion of the work was
however made in section V.
II. PRELIMINARIES
AND
PROBLEM
STATEMENT
There are quite a number of research works targeting the
energy aware of DC for IIoT such as Gonalves et al. The
author in [4] presented a software defined network guide to be
implemented in the industrial internet of things (IIoT)
environment with major focuses on how a better Quality of
service (QoS) can be guaranteed with the provision of a fault
tolerant scheme. Surprisingly, none of their work presented
any result of a reference topology in advance. Meanwhile,
Dhondge et al. in [5] proposed a heuristic and opportunistic
link selection algorithm, HOLA, which not only reduces the
overall energy consumption of the IoT network but also
balances it across the network. However, their work did not
also present in detail, the results of their approach. Also, since
the paper tries to focus on the data aggregation as well, it
however does not see any need to consider analyzing the
scheduling of the networked control system and its allowable
delay bound as expatiated by D.S Kim in [6]. Hence, this work
simulates in advance as shown in the simulation results using
more detailed parameters as backbones for data center.
When IIoT networks are compared with the traditional
network approach, a renewed preference for IIoT networks
becomes undoubtedly higher. In order to apply Low-powered
wireless sensor nodes technology on data centers in an IIoT
platform, practical network related implementation issues will
be faced [7]. These issues are not far-fetched and they include:
• Inability at forwarding a plane delay. Since software
defined IIoT operates a centralized and top to bottom
control system, with enormous amount of fresh data
being forwarded to the controllers at a frequent rate,
ultimately resulting at forwarding delays and in some
scenarios, huge packet losses.
• Inability to guarantee an often complex and unstable
software and system architecture. For example, in
order to achieve how networks are programmed,
allowing the system control applications to gain
unrestricted access for the entire system control
becomes paramount.
This approach did not consider the network protocols as
part of this study since, it is expected that a suitable network
protocol would have been in existence and operation, prior to
the implementation and deployment of the sensor nodes. Other
factors which could limit or hinder the inter-operability of the
various system components such as environmental, human or
machine errors and limitations were also not considered as
part of this work, since the components are expected be in a
perfect working conditions prior to installations and
deployment works. However, considerable amount of effort
was put at analyzing the few existing approaches in the
market, while a holistic effort was made at switching from the
predominantly traditional approach into a more conventional
one. Our focus lies on the implementation of low-powered and
energy harvesting sensor nodes to an already existing IIoT
platform.
III. SYSTEM
MODEL
A. Data Center Network Topology
The three main entities visible in Figure 1 are well
connected and interfaced using the data center as their
computational center and in the process, requires enormous
amount of energy for operations and to power the numerous
sensors in use, hence, the need for this paper to focus
ultimately on the throughput awareness resulting directly from
power complexities of nods in a typical data center during an
IIoT process. A network topology is a collection of data
centers, wide area networks (WAN), core networks, aggregate
networks, the access networks and the edge layer occupied by
several entities of IIoT, entities such as industrial area,
housing, hospital, public area, and college.
The Network topology of a data center as illustrated in the
structural diagram of Figure 2, refers to the flow of energy. It
further buttresses how Data Centers are meant to act or replace
the server spaces which dictates how the internal flow should
be done. Figure 3 shows an illustration of the power that is
consumed by various units of a typical data center. It is
evident that the cooling infrastructures consumed bulk of the
energy resources, while the servers, storage systems and all
others are next up in that order.
Fig. 1. Data center multi-tier (three-tier) network topology [8].

978-1-5090-6785-5/18/$31.00 © 2018 by IEEE
B. Practical Implementation
As a way of finding a lasting and viable solution to the
aforementioned problem statements, the following
constructive aspects are greatly considered by this work.
• Designing hybrid devices that supports both the
Industrial IoT and the traditional network
architectures. A description of whether the packets
are of industrial sensor network or that of the legacy
network is carried out by the hybrid device, which
can solve the interoperability problem between
Industrial IoT and the traditional or legacy networks.
• A distributed control pane which main
responsibilities are to distribute multi-level
computing tasks are adopted here since performance
overheads are prominent with unitary or single
controller platforms. A more reliable network is
achieved.
A holistic look at the point or location of sensor installation
and monitoring is highly essential in the field of IIoT. In as
much as numerous studies have been made of how little or no
wire is required in the branch of wireless communication
technology, the impracticability of this approach remains
rampant, as a result of cost and other factors such as location
of deployment, weather, nearness to data center, servers Etc.
In order to maintain the flexibility and drive down the
financial implications of the project’s installation and
monitoring, we have ensured that the introduced sensors have
a longevity ratio of nearly 5-years when they run on batteries.
A good case study here was the TSCH based wireless sensor
network products known as the linear technology mesh
approach which are cable of running on a 2 AA batter and
within the capacity of 50A. When a stable environmental
condition is provided, nodes can run without any impediment,
whilst its energy source is concurrently harvested. This is
evident in Figure 4.
IV. PERFORMANCE
EVALUATION
A. Assessment Testbed
The focus of this part presents an illustrative approach of the
relevance and importance of a low-power sensor-based IIoT in
the context of Industrial IoT. A prototype approach was
presented in figure 4 justifying the concept of a sensor-powered
Industrial Internet of Things (IIoT) implementations. This
approach incorporated multi-platforms such as industrial
robotics, cloud data centers, AGV, IWN, RFID readers etc.
After deployment, durability tests were conducted in order to
have the system compared with the existing or traditional
approaches. It was discovered that while the existing approach
lacked an ability to inter-communicate within its numerous
components, our approach was able to conveniently do so
without any visible or inherent difficulties, hence, solving the
issue of component autonomy with regards to decision-making.
A look at ways of delivering more intelligent services
through this approach was also considered since information
from the combination of the component units of the machine
such the conveyor unit, the industrial robotics units and the
workpieces can be intelligently gathered and processed in real-
time using the negotiation mechanism as observed by the
author in [11]. Other notable difficulties with the traditional
approach is its inability to support a customized component
unit. Table I and Table II illustrates the efficacies of both the
component units and the entire system unit. Whilst Table I
gives a view of the contents of the data center architecture
which are basically a collection of switches, servers, storage
devices and sensors with their corresponding values, Table II
attempts to justify the energy cumulative of both the servers
and the switches. It can be seen that the total number of
equipment that were used was increased while still maintaining
Fig. 3. Energy variations of the individual units of a typical data center system
[10] is shown. Evidence shows that the cooling infrastructure consumed more
of the energy resources than the any other component parts.
Fig. 4. Deployed energy-harvesting sensor devices such as the thermal-
harvested wireless temperature sensor from ABB [3] is displayed.
Fig. 2. The energy paths or routes of a data center [9] is displayed. A
look at how a typical data center controls the system’s internal energy
flow is captured in this figure.

978-1-5090-6785-5/18/$31.00 © 2018 by IEEE
a low energy consumption by the system, a justification that
the implementation of low-powered sensors for a typical data
center unit of an Industrial IoT is a viable approach.
TABLE I: Component Unit’s task of a Data Center. [12]
No.
Data Center Architecture Three-tier
1. Switches (Core) 8
2. Switches (Aggr.) 16
3. Swiches (Access) 64
4. Servers 1536
5. Users 1
6. Task MIPS 300000
7. Task Memory 1000000
8. Task Storage 300000
9. Task Size 8500
10. Task Output Size 250000
11. No. of Sensors 10
TABLE II: Component Unit’s Performance Assessment.
No.
Energy Profiles and Variations of
Deployed Servers and Switches Values
1. Average Load of Servers 0.3
2. Data Center Load 26.0
3. Total Tasks 348899
4. Average Task of Servers 227.1
5. Task Rejected by Data Center 0
6. Task Failed by Servers 0
7. Total System’s Energy 6314.9
8. Switch Energy (Core) 932.1
9. Switch Energy (Aggr.) 1375.7
10. Switch Energy (Access) 3541.0
11. No. of Sensors 10
B. Simulation Results and Discussion
The approach of this work attempts to configure the energy
spread and to bring down the amount of energy consumed by a
typical data center unit judging with the energy consumed by
the component units they are made of, using low-powered
sensors. To achieve these, an instantaneous dissipation of
power by the component units at time t was given as:



=
f(





)(1)
Instantaneous energy dissipation equation is explained thus;
• S
t
– is an illustration of the internal system state with
respect to time t. It is subcategorized as operating
system states, the physical system state and the
application system state.
• A
t
– is an illustration of the application’s total data
received with respect to time t. these ranges from the
application system’s data parameters to the system’s
data input request arrival rates.
• E
t
– is an illustration of the data center’s data
scheduling and execution approach [13] with respect
to time t. in the part, the amount of energy consumed
by the CPU and its frequency at task scheduling, the
power consumed by the server and other workload
parameters are evaluated here.
Since the network architecture of all industrial system
application remains the network candidate backbone for such
systems, which cuts across all spheres or domain such as the
Industrial Internet of Things (IIoT) and the smart technologies.
This approach modelled its schema in that form by denoting
the network architecture as its candidate backbone. However,
in this approach, rather than implementing the older or
traditional backbone network technologies such as the one-tier
Fig. 6. Throughput-aware characteristics of the energy consumption and
variations with respect to the signal to noise ratio of the deployed sensors.
Fig. 5. Illustrative graphs showing the accounted power of servers, the
energies for core switches, servers and the aggregate switches.

978-1-5090-6785-5/18/$31.00 © 2018 by IEEE
or the two-tier, we went a step ahead to consider the three-tier
network backbone as our candidate backbone, since the three-
tier network are beneficial in terms of energy consumption and
management, coupled with the growing nature and versatility
of IIoT entity as a result of internet.
The purpose of this work was well emphasized by the
illustrative graphs in both Figure 5 and Figure 6 with the use
of MATLAB 2015a simulation tool. An evaluation of how
much power was consumed by the servers in a typical data
center when the load server is predefined was ascertained in
Figure 5. It was observed that with a considerable and
consistent increase in the number of servers, a drop in the
power consumption ratios was evident. An observation was
also noted that the amount of energy that was consumed by the
datacenter system was relatively higher when the number of
servers used was kept at below 250. The energy consumption
rates began to drop when the number of servers was increased
above 250 units. The energy drop became sporadic when
number of used servers was put at around 450. A stable energy
variation was however witnessed when servers were
maintained at over 500 units. A typical industrial internet of
things environments implements over 500 unit of servers for
its operations. A look at the energy consumption variations of
both the core and the aggregate switches however showed a
consistently stable consumption rates with the rates put at over
500 Watts/h for the core switches and over 120 Watts/h for the
aggregating switches.
Figure 6 showed that the data center load attained a higher
throughput value at a predefined load server and signal-to-
noise SNR(dB) ratio. From the above graphical illustration, an
apparent increase in the number of sensor nodes will imply a
sequential increase of the data center network system’s
throughput. According to the figure, a throughput of 2.388
x10
4
was noted when the signal to noise ratio was kept at
below 1dB. A consistent increase was observed when the SNR
was increased to nearly 3dB signal to noise ratio. The system’s
throughput was however normalized and kept stable at 2.4
x10
4
when the sensor is increased further to a signal to noise
ratio of 3.5dB and beyond.
V. C
ONCLUSION
The work of this paper presents an implementation
methodology of energy-efficient wireless sensor nodes for
industrial internet of things (IIoT) with respect to the three-tier
IIoT architecture backbone of a data center. In order to obtain
and manage these factors, we have proposed and implemented
Low power wireless sensor nodes. We have carefully observed
and analyzed how the energy can be harvested and ultimately
reduced through the scheme. The importance of harvesting
energy using low-powered and wireless sensor devices are is
advantageous in these following ways:
• Analyzing and Mapping of a typical data center’s
power consumption across the entire Industrial IoT
platform.
• Higher throughput value at predefined load server
and signal-to-noise SNR(dB) ratio is attained.
The simulation result shows the validation of the proposed
experimental approach and goes ahead to demonstrate the
effectiveness of the scheme with regards to the real-time
factors that were considered and the promising strategies for
the scheme’s deployment in other system spheres other than
the three-tier architectural back bone of a data center.
Finally, enhancing the system’s flexibility to multi-interact
with its component parts and maintaining a low installation
cost of the system, capable of running wireless sensor network
that is powered for IIoT applications for more than five (5)
years remains our prime focus in the future study.
Acknowledgment
This research was supported by BK21+ Project of the
Kumoh National Institute of Technology and the MSIP
(Ministry of Science, ICT and Future Planning), South Korea,
and under the ITRC (Information Technology Research
Center) support program (IITP-2016-H8601-16-1011), and the
Global IT Talent support program (IITP-2017-0-01811),
supervised by the IITP (Institute for Information and
Communications Technology Promotion), together with the
National Research Foundation of Korea(NRF) through the
Advanced Research Project (NO. NRF2017R1A2B4009900).
Reference
[1] R. Buyya, C. Vecchiola, and S. Selvi, “Mastering Cloud Computing:
Foundations and Applications Programming,” Amsterdam, The
Netherlands: Elsevier, 2013.
[2] L. Krug, M. Shackleton, and F. Saffre, “Understanding the
environmental costs of fixed line networking,” in Proc. 5th Int. Conf.
Future e-Energy Syst., 2014, pp. 8795.
[3] Joy Weiss, and Ross Yu, “Wireless Sensor Networking for the Industrial
Internet of Things” Dust Networks Product Group, Linear Technology.
[4] P. Gonalves, J. Ferreira, P. Pedreiras, and D. Corujo, “Adapting sdn
datacenters to support cloud iiot applications,” in 2015 IEEE 20th
Conference on Emerging Technologies Factory Automation (ETFA),
Sept 2015, pp.14..
[5] K. Dhondge, R. Shorey, and J. Tew,, “Hola: Heuristic and opportunistic
link selection algorithm for energy efficiency in industrial internet of
things (iiot) systems,” in Conference on Communication Systems and
Networks (COMSNETS), Jan 2016, pp. 16.
[6] D.-S. Kim, Y.-S. Lee, W.-H. Kwon, “Maximum Allowable Delay
Bounds in Networked Control Systems,” Control Engineering Practice
(IF: 1.912. ISSN: 0967-0661), Vol. 11, Issue 11, pp. 1301-1313, 2003.
[7] L. Hu, M. Qiu, J. Song, M. S. Hossain, and A. Ghoneim, “Software
defined healthcare networks,” IEEE Wireless Commun., vol. 22, no. 6,
pp.6775, Dec. 2015.
[8] Cisco, “Data Center Multi-Tier Model Design,” Cisco Data Center
Infrastructure.
[9] Y. Kodama et al., “Power reduction scheme of fans in a blade system by
considering the imbalance of CPU temperatures,” in Proc. IEEE/ACM
Int. Conf. CPSCom GreenCom, Dec. 2010, pp. 8187.
[10] Info-Tech, “Top 10 energy-saving tips for a greener data center,” Info
Tech Research Group, London, ON, Canada, Apr. 2010
[11] S. Wang, J. Wan, D. Zhang, D. Li, and C. Zhang, “Towards the smart
factory for industrie 4.0: A self-organized multi-agent system assisted
with big data based feedback and coordination,” Elsevier computer
networks, Comput. Netw., vol. 101, pp. 158168, Jun. 2016.
[12] Nwadiugwu Williams PC, Tarigan I.D.F, Jae-Min Lee, and Dong-
Seong Kim, “Energy-aware Data Center for Industrial Internet of Things
(IIoT),” in KICS [The Korean Institute of Communications and
Information Sciences.]

978-1-5090-6785-5/18/$31.00 © 2018 by IEEE
[13] J. Shuja, K. Bilal, S. Madani, and S. Khan, “Data center energy efficient
resource scheduling,” Cluster Comput., vol. 17, no. 4, pp. 12651277,
Dec. 2014.