This document proposes a campus edge computing network based on IoT street lighting nodes. It aims to address the issue of inadequate network resources on campus from the increasing number of IoT devices and data streams. The system employs street lights as decentralized edge computing nodes that connect IoT devices, collect and process sensor data, and communicate with the campus cloud platform. The cloud platform uses neural network algorithms to predict network resource requirements, analyze the workload of each service, and efficiently allocate resources across the campus network to maintain quality of service as the number of IoTs grows. Experimental results showed the approach reduces cloud loading and can dynamically adjust resource distribution for balanced performance.
2. CHANG AND LAI: CAMPUS EDGE COMPUTING NETWORK BASED ON IOT STREET LIGHTING NODES 165
In Section V, the final part, the conclusion, and future work
suggestions are offered.
II. RELATED WORKS
A. Smart IoT Campus
In 2005, the definition and scope of IoT components included
people, things, software platform, networking capabilities, and
information in the report of the International Telecommuni-
cation Union. With the development of IoT technology, there
are more and more sensors for sensing speed, temperature,
humidity, pressure, position, level, energy consumption, radi-
ation, vibration sensors, and so on. IoT is gradually applied to
life, work, and industry with sensing information, gathering, and
analysis technology. Different sensors are used for different pur-
poses, such as smart aviation, energy management, environmen-
tal protection, and biological sciences; smart campus is also one
of the essential developments [8]–[10]. Sensors and wearable
technology connect students, teachers, and classes about teach-
ing, traffic, energy, and safety on campus. The search shows
that virus propagation model reflects viral infections and could
serve to enhance the security [11]. Moreover, Stefano Bracco
et al. proposed an energy management system to present in the
control room of the smart polygeneration micro-grid with an
adequate model of all the components [12]. This paradigm of
IoT defines a specific model for a campus to enable new in-
teractions between objects, including smart classrooms, smart
parking, and smart learning; a key technology for realizing a
smart campus is virtualization technology. Total cloud com-
puting needs to support the diversification and popularization
of networks, as well as to improve the utilization of resources
with more and more application services on the smart campus
[13]–[15]. The network exchange is divided into NFC and
wide area network technology, with most commonly used
IEEE802.15.4 formulated by the IEEE for low-power, low-rate,
and short-distance transmission characteristics. For transmis-
sion and exchange of information between different networks,
a gateway is also needed to coordinate the format of the data in
each domain. The gateway usually establishes service-oriented
architecture to provide different service applications [16], [17].
B. Smart Street Lights Applications
Because of the existing infrastructure of the essential city and
its spread over a wide range of areas, the smart lights become the
most critical task for building a smart city combined commu-
nications, sensing, and intelligence-related technologies. The
smart light pole will aim at constructing the wisdom sensing
extension spectrum of smart cities, connecting the entire city
through various light poles as network nodes. In addition to
the introduction of street lamp monitoring services, and further
building a micro-climate system to collect weather information
around each light pole node, to make the information of all re-
gions of the city more transparent and detailed for the smart city
and smart campus [18]–[20]. Some research uses low-power
ZigBee mesh network to provide maximum energy efficiency in
response to adaptive traffic on the road [21], [22]. The system
combines the additional benefits of LED technology, such as
the use of a wireless control system for traffic-based dimming
to conserve energy, at a savings of 80% over previously installed
metal halide bulbs [23]. Some research proposed that solar street
lights consume minimal energy from the city network by using
solar energy to charge the battery during the day. Also, with
the development of transmission control protocol and internet
protocol (TCP/IP) communications over centralized software in
the cloud via GSM, high-efficiency LED lighting is managed
and monitored [24]. With the development of the IoT, addi-
tional sensors are gradually installed on street lights to provide
smart city services [25]. The smart streetlight also becomes a
large-scale network service computing node because it com-
bines many sensing services.
C. Workload Prediction
Workload prediction is a way of predicting the distribution
of follow-on resources based on job characteristics, such as
time, bandwidth, and user behaviors. Different algorithms with
a variety of approaches have been proposed for many years.
Cloud environment allocation is one of the studies on workload
perdition [26], [27]. It tries to predict the cloud computing node
request and manage the total resources. Some algorithms focus
on energy consumption for maximum resources and minimum
consumption to ensure that services are satisfied with service
level agreements. However, the time delay neural network
(TDNN) is work for workload prediction in the cloud envi-
ronment [28], [29]. It applies the neural network to compute the
delay time and change the weight of neural layers. The input
consists of the vector
y(n) =
m
j=1
wj φ
p
i=0
wj (i)x (n − 1) + bj
+ b0 (1)
The neural network computes the next value as output based
on the first few values; however, it needs a large amount of con-
tinuous data to achieve this good prediction over a long time.
The feature of TDNN is unsuitable for predicting the workload
for the new environment. The regression model involves an al-
gorithm of relationship computing for a dependent variable or
independent variables. Regression model includes changes in
independent variables while the other independent variables re-
main unchanged. The non-linear regression formula is shown as
yi = a0 + aixi + aix2
i + · · · aixm
i + δi (2)
where a is the weight coefficient of the regression model and
δ is the error of prediction in the formula. The m value depends
on the degree of freedom computing model. The algorithm
which predicts data distribution is applied for imbalance issues
on the cloud server. This strategy first cluster all workload
data to some groups and then determine whether to place
them in different virtualizations based on capacity or focus
characteristics, e.g., computing time, quality of service, and
energy consumptions. The allocation method also detects the
under-loaded server and calculates that its total workload under
the maximum required to load and redistributes the scheduling
probabilities. The statistic based load balance (SLB) [30] helps
virtual machines (VM) to make host selections for resource
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.
3. 166 IEEE SYSTEMS JOURNAL, VOL. 14, NO. 1, MARCH 2020
imbalanced. SLB estimates the resource requirement of each
VM with historical data included computing and performance.
The workload perdition of SLB is defined as the formula
Li (VMcpu) = E (li (VMcpu)) =
n
j=1 lij (VMcpu)
n
(3)
where Li is the value of predicted loading at time i, li is the
loading value, and lij is the loading at time i on the jth day.
D. Cloud Resource Allocation
Cloud resource allocation is the main focus specifically on the
scheduling of services or resource creation within the cloud. The
study proposes a planning model that addresses the scheduling
of various cloud services and the integration of related ser-
vices to accommodate the complexities of cloud and in-service
changes [31], [32]. Some research performance aspects justify
the dynamic cloud plan case. Moreover, apply it to business
development. Some research also presented a method to de-
tect the amount of VM requests for dynamic configuration and
adjustable application services. It provides performance and
workload information to meet the quality of service require-
ments and decide whether to reject the request [33]. The trust
degree mentioned above is based on the number of failures since
the VM returned to the resource pool. The monitor will instantly
collect the credibility of the cloud slots. Moreover, it will set
up the record table and save the credibility of the entire cloud
resource. The scheduling algorithms mostly consider effective-
ness and immediacy.
III. PROPOSED SMART IOT NETWORK SYSTEM
This study mainly presents a decentralized smart IoT cam-
pus platform that includes decentralized nodes and IoT campus
cloud to improve the imbalance of network resources caused by
the rapid proliferation of IoT devices. The smart campus plat-
form establishes decentralized IoT nodes with street lights to
connect campus IOT devices and related information and com-
munication campus cloud for data reports. The cloud platform
is responsible for collecting relevant IoT information and link-
ing the existing school-related service to the retrieval service
data model. It establishes resource prediction and allocation of
related virtual environment to ensure that each service achieves
the requisite quality with the neural algorithm based on each ser-
vice and the IoT device data. The overall architecture is shown
in Fig. 1.
A. Distributed Campus Light Nodes
Considering the overall campus is often more extensive; this
study combines campus street lights as decentralized IoT com-
puting nodes to quickly establish the overall campus IoT layout
construction environment. The overall structure of street lights
is shown in Fig. 2. It monitors street lighting electricity infor-
mation with the street lighting controller and sends the data to
the central control center.
Through the processing of the measurement data, the cloud
platform provides street lamp status management, real-time
diagnosis, and power-saving services. Also, this study aims
Fig. 1. Smart IoT campus edge computing network.
Fig. 2. Structure of stree tlights.
to plug some IoT devices into the light poles to allow the system
to capture the status of micro-weather stations, camera video
transmission, plus emergency notification status, in addition to
WiFi devices providing WiFi Internet access service, and event
information push with digital signage. Overall, relevant IoT in-
formation collocation analysis, statistics, and rule judgments
are converted into effective results, extending the wisdom of
campus applications. The specific wisdom of the campus IoT
objectives are as follows.
1) Street Lights Management: The original street light man-
agement system can only wait for the passers-by to report events
through the Internet, telephone, SMS, and manually input in-
formation. Through street light sensors, 24/7 street lamp mon-
itoring services are offered, and upon finding abnormal states,
take the initiative to inform management regarding street light
perception services.
2) Micro-Weather Station Information: The environment
module collects information, including CO, CO2, PM2.5, wind
speed, temperature, humidity, ultraviolet, and noise. After the
system gets the environmental information, it analyzes and
sets relevant rules to judge whether it is an event notification.
Also, users can obtain weather bureau information for the next
12 hours or 1 week from the display board.
3) Emergency Notification: If a particular event triggered the
emergency button or the image detection suggested abnormal
behavior, it will be immediately reported to the management
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.
4. CHANG AND LAI: CAMPUS EDGE COMPUTING NETWORK BASED ON IOT STREET LIGHTING NODES 167
center so that relevant personnel will be sent to the scene to
assist.
4) IP Cam: In addition to security monitoring, users can
take pictures and edit the images themselves, and then send the
photo or image to e-mail or use Bluetooth and NFC mode to
mobile devices.
5) Wireless WiFi: WiFi gateway provides the WiFi signals
to enable the public to connect to the Internet for use and to
obtain mobile network usage for prediction future.
6) Digital Signage: The signage usually not only uses the
information to push and broadcast but also uses dynamic or static
rendering to link to the presentation of weather information; if
people use the image, it will be linked when editing is achieved,
constituting a triple play.
B. IoT Broker of Edge Computing Nodes
Today, IoT services are flooded with a variety of messag-
ing protocols and message exchange formats. The IoT devices
are usually commutations with different protocols. Beside, con-
strained application protocol (CoAP) and message queuing
telemetry transport (MQTT) are commonly designed for the
IoT, but they have different architects, e.g., NAT Issue, trans-
port, and message. The IoT Broker was designed for the edge
computing of the different protocols and transports the com-
ments from device to device to avoid the server loading. If there
is a high degree of heterogeneity among services, messages can-
not be aggregated for effective message exchange, and the IoT
services can only operate independently [34]–[36]. In the cur-
rent IoT environment, the most commonly adopted messaging
agreement is the MQTT and CoAP. The IoT Broker is not only
for CoAP and MQTT heterogeneous messaging agreements and
content exchange format translation intermediaries but also for
the MQTT protocol nodes connected to the MQTT Broker [37].
Also, it can connect with the smart campus cloud to receive
messages from the network core. Through the translator mech-
anism designed in this research, the communication between
the MQTT Node and the CoAP sensor is achieved, thereby
implementing bidirectional conversion between the CoAP and
MQTT protocols; it consists of two major components: Mes-
sage Broker and Manager. Message Broker was built as an
open source project with JavaScript language and the Man-
ager component was written in Java language. As shown in Fig.
3, Message Broker serves as a clearinghouse for Message Ne-
gotiation, which includes CoAP Server components that allow
CoAP Node endpoints to connect and receive MQTT resource
messages. The MQTT Broker component allows MQTT nodes
to connect and exchange MQTT resources, or to obtain CoAP
Sensor resources. The topic router component can record IoT
Broker containing the resource name. Message cache compo-
nents will be retained for each resource name of the last message
for temporary storage. The manager is used to obtain resources
messages on the CoAP Sensor side of the environment. The
user datagram protocol (UDP) Socket component is used to
receive the Broker Request sent by the controller and request
information from the external resource included in the Broker
Request. The UDP Socket component sends the information
to the CoAP client for observation and obtains the external
resource.
Fig. 3. IoT Broker module.
Fig. 4. Model of resource manager.
C. Resource Manager of IoT System
The workload is defined as the amount of processing slot used
when the smart campus service works in the cloud computing
environment. This study focuses on the number of applications
working and the total resource allocation in the cloud comput-
ing environment [38]–[40]. Besides, a recurrent neural network
model is proposed based on the prediction of total workload,
including IoTs data processing, mobile resource, and virtual
class for the cloud servers. The proposed system architecture is
shown in Fig. 4 that consists of clients, resource master, predic-
tor, servers, and four VMs which included four cores and 8G
memory in every VM.
The resource manager provides the smart campus services
with a resource allocation algorithm to avoid resource imbal-
ance. The total steps of resource management are presented as
follows.
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.
5. 168 IEEE SYSTEMS JOURNAL, VOL. 14, NO. 1, MARCH 2020
1) Job Request: When the new services start, the job will
be split into many processes in the smart campus cloud
environment. The workload information is composed of
these processes for the request.
2) Available server’s selection: The manager will choose the
available servers and create virtual resources based on the
requirements for the job.
3) Workload Predictor: Resource manager sends the history
and class information of services to the predictor. The
predictor uses a neural network algorithm to compute how
much resources are needed for the service requests.
4) Prediction Results Report: The predictor calculates the
workload information via previous workload information
for the following time and reports the result. The resource
manager will allocate suitable resources according to the
results and remaining resources in the smart campus en-
vironment.
5) Remaining Resource Return: When smart cloud services
complete service requests, they will send resource and
workload information.
D. Resource Prediction Model
The prediction of the neural network model consists of three
layers: an input layer, a hidden layer, and an output layer. The
input layer receives the workload information from X1 to Xa at
different time steps of time sequence in the neural network. At
time t, the vector of the input layer is shown as
x(t) = [x(t), x (t − 1) , . . . , x (t − p)]T
(4)
where p is the number of selected delay line memory. The output
is fed back as the input k at the next time step. It means that the
neural network will remember the workload information. The
vector selects the number of p and decides the memory period
of the neural network to remember the past workloads.
For the hidden layer of the neural network, the output of
single neuron j is shown as formulae (5) and (6)
yj (t) = f
i∈A∪B
wji (t − 1) ki (t − 1) + bj
(5)
where f function is the activation layer with the neuron j; wji is
the weight of layers; and bj is the offset
ki(t) =
xi(t), if i ∈ A
yi(t), if i ∈ B
(6)
where ki(t) is the real input xi(t) if i ∈ A and is the output yi(t)
if i ∈ B. The output of the hidden layer is computed with m
neurons
y(t) =
m
j=1
wj yj . (7)
The output vector y(t) is transformed from the input vector
shown as
y (t) =
m
j=1
wj f
i∈A∪B
wji (t − 1) ki (t − 1) + bj
+ b0
(8)
Fig. 5. Overall distribution of the map.
The learning algorithm defines the activation function for the
neural network. Moreover, the error function is defined by output
neuron Sa (t) of the hidden state neurons. The error function E(t)
is defined as
E(t) =
1
2
a
a=1
(da (t) − Sa (t))2
(9)
where da (t) is the response value in the neuron. Moreover, the
Sa (t) is shown as follows:
state : Sa (t) 1 − β, if accepted
state : Sa (t) β, if rejected
(10)
where β is the tolerance value of the neuron network. There are
different cases to present different results. The first one is that
the network reject a negative string when Sa (t) β; the other
is that the network accept a positive string when Sa (t) 1–β.
The steepest descent method calculates the updated weight
Δ wkj = −α
∂E(t)
∂wkj (t)
= α (da (t) − Sa (t)) ·
∂Sa (t)
∂wkj (t)
(11)
where α is the learning rate.
IV. IMPLEMENTATION AND EXPERIENCES
A. Smart IoT Edge Computing Nodes
This study builds 176 IoT computing nodes with campus
street lights to connect the entire campus to IoT information
network. A total of 30 points in the above establishment were
selected as IoT sensing devices. The overall distribution of the
map is shown in Fig. 5.
The campus manager will quickly understand the overall situ-
ation around the campus, including power consumption, camera
video, and sensor information with smart campus webpage, as
shown in Fig. 6. The smart IoT service is also built as the
smart energy; various classrooms and street lighting power con-
sumption information combined with the campus class table to
determine whether the event did not turn OFF lights. Smart light-
ing presents that the lighting is dynamically adjusted according
to the intersection of the camera, the flow of traffic, and human
perception. Also, the event can also be based on the relevant
alarm threshold to determine whether there are abnormal events,
to quickly inform the campus managers.
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.
6. CHANG AND LAI: CAMPUS EDGE COMPUTING NETWORK BASED ON IOT STREET LIGHTING NODES 169
Fig. 6. Smart IoT application.
Fig. 7. Experiment result with regression and TDNN.
B. Experiment of Network Perditions
The total experimental data are collected from the IoTs de-
vice, mobile node, and campus system in the public environ-
ment. There are 3224 users and 164 142 jobs executed on the
smart campus environment, with 12 neurons and four delay
times in the hidden layer of the neural network for the predic-
tion model. The simulation result is shown in Fig. 7. There are
three parts of the algorithm for analyzing the learning effect.
First, the experimental result shows that the proposed method
Fig. 8. Enlargement of experiment results.
Fig. 9. MSE of workload prediction.
is similar to the regression method and TDNN before 600 time
slots. At that period, the data are insufficient to calculate more
accurate prediction results. Moreover, it shows a better result
than the previous data between 600 and 1600 because the pre-
dicted number is close to the real workload number. The last
one is better than the regression method after 1600 time slots.
The model presents that a predicted line is closed to the real line
even though there is some error.
Fig. 8 shows that there are some errors in both the proposed
method and the regression function for workload predictions, but
the proposed method shows the better effect than the regression
model. The proposed method is similar to the real workload,
but the regression cannot show the rapid changes in the data.
Therefore, in this study, the neural learning method is more
accurate than the regression method. In this study, mean square
error (MSE) was used to calculate the prediction results. The
MSE is calculated as follows:
MSE =
1
N
N
i=1
(p − t)2
(12)
where p is the predicted workload value; t is the total workload
value; and N is the number of data.
The MSE is given in Fig. 9 for the TDNN, regression, and
proposed method. The result shows that the proposed method is
better than the TDNN and regression methods. A further com-
parison with TDNN shows that the proposed method is better
than TDNN. Moreover, TDNN needs amounts of input data as
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.
7. 170 IEEE SYSTEMS JOURNAL, VOL. 14, NO. 1, MARCH 2020
training data, so the TDNN is unsuitable even in the rapidly
changing IoT environment. The regression algorithm, which
usually needs long computing time or stable system analysis
of the linear model, may be better than our algorithm for long
time series over a month or a year, but the computing time is an
essential factor for the prediction problem. The proposed algo-
rithm is more suitable for workloads over a short period, and to
ascertain the trend of the real workload.
V. CONCLUSION
In considering the situation of the wisdom of IoT devices, the
resource imbalance and quick response time were challenges
for smart IoT services. This study proposes the architecture of a
smart IoT platform which built the streetlight as the computing
nodes and prediction model for the workload of smart campus
environments. The results show that the proposed predictions
help the cloud manager to avoid some resource imbalance when
there are large amounts of request requirements from smart cam-
pus services. The experimental results show that the proposed
model achieves better predictions over a short period than the
regression method and TDNN. Hence, the cloud provider can
connect IoT devices and apply the resource allocation manage-
ment more efficiently with the proposed platform. In the future,
we will target a variety of different applications to establish
exclusive application patterns to help improve the prediction
accuracy.
REFERENCES
[1] A. Sheth, “Internet of things to smart IoT through semantic, cognitive,
and perceptual computing,” IEEE Intell. Syst., vol. 31, no. 2, pp. 108–112,
Mar./Apr. 2016.
[2] R. N. Calheiros, R. Ranjan, and R. Buyya, “Virtual machine provisioning
based on analytical performance and QoS in cloud computing environ-
ments,” in Proc. Int. Conf. Parallel Process., Taipei, Taiwan, Sep. 2011,
pp. 295–304.
[3] S. Ward and M. Gittens, “Building useful smart campus applications using
a retired cell phone repurposing model,” in Proc. 3rd Int. Conf. Elect.
Biomed. Eng., Clean Energy Green Comput., Beirut, Lebanon, Apr. 2018,
pp. 43–48.
[4] A. Zhamanov, Z. Sakhiyeva, R. Suliyev, and Z. Kaldykulova, “IoT smart
campus review and implementation of IoT applications into education
process of university,” in Proc. 13th Int. Conf. Electron., Comput. Comput.,
Abuja, Nigeria, Nov. 2017, pp. 1–4.
[5] H. Yan and H. Hu, “A study on association algorithm of smart campus
mining platform based on big data,” in Proc. Int. Conf. Intell. Transp., Big
Data Smart City, Changsha, China, Dec. 2016, pp. 172–175.
[6] V. C. Emeakaroha, P. Healy, K. Fatema, and J. P. Morrison, “Analysis of
data interchange formats for interoperable and efficient data communica-
tion in clouds,” in Proc. IEEE/ACM 6th Int. Conf. Utility Cloud Comput.,
Dresden, Germany, Dec. 2013, pp. 393–398.
[7] Y.-C. Chang, R.-S. Chang, and F.-W. Chuang, “A predictive method for
workload forecasting in the cloud environment,” in Advanced Technolo-
gies, Embedded and Multimedia for Human-Centric Computing. Berlin,
Germany: Springer, 2014, ch. 65, pp. 577–585.
[8] Y.-B. Lin, L.-K. Chen, M.-Z. Shieh, Y.-W. Lin, and T.-H. Yen, “Cam-
pusTalk: IoT devices and their interesting features on campus applica-
tions,” IEEE Access, vol. 6, pp. 26036–26046, May 2018.
[9] L. Di, “The internet-of-things framework for campus-vehicle early warn-
ing,” in Proc. IEEE Symp. Robot. Appl., Kuala Lumpur, Malaysia, Jun.
2012, pp. 544–547.
[10] T. Ueda and Y. Ikeda, “Socio-economics and educational case study with
cost-effective IOT campus by the use of wearable, tablet, cloud and open
E-learning services,” in Proc. ITU Kaleidoscope: Challenges Data-Driven
Soc., Nanjing, China, Nov. 2017, pp. 1–8.
[11] L. Wang, C. Yao, Y. Yang, and X. Yu, “Research on a dynamic virus prop-
agation model to improve smart campus security,” IEEE Access, vol. 6,
pp. 20663–20672, 2018.
[12] S. Bracco, M. Brignone, F. Delfino, and R. Procopio, “An energy man-
agement system for the Savona campus smart polygeneration microgrid,”
IEEE Syst. J., vol. 11, no. 3, pp. 1799–1809, Sep. 2017.
[13] W. Li, J. Tordsson, and E. Elmroth, “Modeling for dynamic cloud schedul-
ing via migration of virtual machines,” in Proc. 3rd Int. Conf. Cloud
Comput. Technol. Sci., Athens, Greece, Nov. 2011, pp. 163–171.
[14] C.-F. Lai, M. Chen, J.-S. Pan, C.-H. You, and H.-C. Chao, “A collaborative
computing framework of cloud network and WBSN applied to fall detec-
tion and 3-D motion reconstruction,” IEEE J. Biomed. Health Informat.,
vol. 18, no. 2, pp. 457–466, Jan. 2014.
[15] B. Hirsch and J. W. P. Ng, “Education beyond the cloud: Anytime-
anywhere learning in a smart campus environment,” in Proc. Int. Conf.
Internet Technol. Secured Trans., Abu Dhabi, United Arab Emirates, Dec.
2011, pp. 718–723.
[16] J. Chen et al., “Exploiting ICN for realizing service-oriented com-
munication in IoT,” IEEE Commun. Mag., vol. 54, no. 12, pp. 24–30,
Dec. 2016.
[17] B. Cheng, D. Zhu, S. Zhao, and J. Chen, “Situation-aware IoT service
coordination using the event-driven SOA paradigm,” IEEE Trans. Netw.
Serv. Manage., vol. 13, no. 2, pp. 349–361, Mar. 2016.
[18] A. Gharaibeh et al., “Smart cities: A survey on data management, security,
and enabling technologies,” IEEE Commun. Surveys Tut., vol. 19, no. 4,
pp. 2456–2501, Oct.–Dec. 2017.
[19] P. T. Daely, H. T. Reda, G. B. Satrya, J. W. Kim, and S. Y. Shin, “De-
sign of smart LED streetlight system for smart city with web-based
management system,” IEEE Sensors J., vol. 17, no. 18, pp. 6100–6110,
Jul. 2017.
[20] M. Shahidehpour, C. Bartucci, N. Patel, T. Hulsebosch, P. Burgess, and
N. Buch, “Streetlights are getting smarter: Integrating an intelligent com-
munications and control system to the current infrastructure,” IEEE Power
Energy Mag., vol. 13, no. 3, pp. 67–80. Apr. 2015.
[21] F. Leccese, “Remote-control system of high efficiency and intelligent
street lighting using a ZigBee network of devices and sensors,” IEEE
Trans. Power Del., vol. 28, no. 1, pp. 21–28, Dec. 2013.
[22] B. C. Mishra, A. S. Panda, N. K. Rout, and S. K. Mohapatra, “A novel effi-
cient design of intelligent street lighting monitoring system using ZigBee
network of devices and sensors on embedded internet technology,” in Proc.
Int. Conf. Inf. Technol., Bhubaneswar, India, Dec. 2015, pp. 200–205.
[23] G. Shahzad, H. Yang, A. W. Ahmad, and C. Lee, “Energy-efficient intel-
ligent street lighting system using traffic-adaptive control,” IEEE Sensors
J., vol. 16, no. 13, pp. 5397–5405, Apr. 2016.
[24] B. Kul, “IoT-GSM-based high-efficiency LED street light control sys-
tem (IoT-SLCS),” in Proc. 26th Int. Scientific Conf. Electron., Sozopol,
Bulgaria, Sep. 2017, pp. 1–5.
[25] G. B. Satrya, H. T. Reda, J. W. Kim, P. T. Daely, S. Y. Shin, and
S. Chae, “IoT and public weather data based monitoring control soft-
ware development for variable color temperature LED street lights,” Int.
J. Adv. Sci., Eng. Inf. Technol., vol. 7, no. 2, pp. 366–372, 2017.
[26] R. N. Calheiros, E. Masoumi, R. Ranjan, and R. Buyya, “Work-
load prediction using ARIMA model and its impact on cloud applica-
tions’ QoS,” IEEE Trans. Cloud Comput., vol. 3, no. 4, pp. 449–458,
Aug. 2014.
[27] C.-K. Tham and B. Cao, “Stochastic programming methods for workload
assignment in an ad hoc mobile cloud,” IEEE Trans. Mobile Comput.,
vol. 17, no. 7, pp. 1709–1722, Oct. 2017.
[28] Md. T. Imam, S. F. Miskhat, R. M. Rahman, and M. A. Amin, “Neural
network and regression based processor load prediction for efficient scal-
ing of grid and cloud resources,” in Proc. 14th Int. Conf. Comput. Inf.
Technol., Dec. 2011, pp. 333–338.
[29] J. M. Tirando, D. Higuero, F. Isaila, and J. Carretero, “Predictive data
grouping and placement for cloud-based elastic server infrastructures,” in
Proc. IEEE/ACM Int. Conf. Cluster, Cloud Grid Comput., 2011, pp. 285–
294.
[30] K. Kaur, S. Kaur, and V. Gupta, “Flow statistics based load balancing in
OpenFlow,” in Proc. Int. Conf. Adv. Comput., Commun. Informat., Jaipur,
India, Sep. 2016, pp. 378–381.
[31] M. Dakshayini and H. S. Guruprasad, “An optimal model for priority
based service scheduling policy for cloud computing environment,” Int. J.
Comput. Appl., vol. 32, no. 9, pp. 23–29, Oct. 2011.
[32] Z. Zhang, H. Wang, L. Xiao, and L. Ruan, “A statistical based resource
allocation scheme in cloud,” in Proc. Int. Conf. Cloud Serv. Comput.,
2011, pp. 266–273.
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.
8. CHANG AND LAI: CAMPUS EDGE COMPUTING NETWORK BASED ON IOT STREET LIGHTING NODES 171
[33] G. Wu, W. Bao, X. Zhu, W. Xiao, and J. Wang, “Optimal dynamic reserved
bandwidth allocation for cloud-integrated cyber-physical systems,” IEEE
Access, vol. 5, pp. 26224–26236, Nov. 2017.
[34] D. Abramson, R. Buyya, and J. Giddy, “A computational economy for
grid computing and its implementation in the Nimrod-G resource broker,”
Future Gener. Comput. Syst., vol. 18, no. 8, pp. 1061–1074, 2002.
[35] X. Li, D. Li, J. Wan, C. Liu, and M. Imran, “Adaptive transmission opti-
mization in SDN-based industrial internet of things with edge computing,”
IEEE Internet Things J., vol. 5, no. 3, pp. 1351–1360, Jun. 2018.
[36] J. Liu, J. Wan, B. Zeng, Q. Wang, H. Song, and M. Qiu, “A scalable and
quick-response software defined vehicular network assisted by mobile-
edge computing,” IEEE Commun. Mag., vol. 55, no. 7, pp. 94–100, Jul.
2017.
[37] S. Tabatabai, I. Mohammed, A. Al-Fuqaha, and M. A. Salahuddin, “Man-
aging a cluster of IoT brokers in support of smart city applications,” in
Proc. IEEE 28th Annu. Int. Symp. Pers., Indoor, Mobile Radio Commun.,
Montreal, QC, Canada, Oct. 2017, pp. 1–6.
[38] Y. Tu, Y. Lin, J. Wang, and J.-U. Kim, “Semi-supervised learning with gen-
erative adversarial networks on digital signal modulation classification,”
Comput. Mater. Continua, vol. 55, no. 2, pp. 243–254, May 2018.
[39] J. Wang, Y. Cao, B. Li, H.-J. Kim, and S. Lee, “Particle swarm optimization
based clustering algorithm with mobile sink for WSNs,” Future Gener.
Comput. Syst., vol. 76, pp. 452–457, Nov. 2017.
[40] J. Wang, J. Cao, S. Ji, and J. H. Park, “Energy efficient cluster-
based dynamic routes adjustment approach for wireless sensor networks
with mobile sinks,” J. Supercomput., vol. 73, no. 7, pp. 3277–3290,
Jul. 2017.
Yao-Chung Chang (M’03) received the Ph.D. de-
gree from National Dong Hwa University, Hualien,
Taiwan, in 2006.
He is an Associate Professor and a Chair with
the Department of Computer Science and Infor-
mation Engineering, National Taitung University,
Taitung, Taiwan. His primary research interests
include intelligent communication System, IoT, and
cloud computing.
Dr. Chang is a recipient of the subsidization
program in universities for encouraging exceptional
talent, Ministry of Science and Technology, Taiwan
Ying-Hsun Lai (M’15) received the Ph.D. de-
gree from National Cheng Kung University, Tainan,
Taiwan, in 2013.
He is an Assistant Professor with the Department
of Computer Science and Information Engineering,
National Taitung University, Taitung, Taiwan. His
research interests include embedded systems, IoTs
applications, cloud computing, and artificial intelli-
gence algorithm.
Authorized licensed use limited to: ESCUELA POLITECNICA DEL LITORAL (ESPOL). Downloaded on January 08,2022 at 21:38:37 UTC from IEEE Xplore. Restrictions apply.