Sei sulla pagina 1di 13

Send Orders for Reprints to reprints@benthamscience.

net
Recent Advances in Electrical & Electronic Engineering, 2021, 14, 347-359 347

RESEARCH ARTICLE

Research on Fog Resource Scheduling based on Cloud-fog Collaboration


Technology in the Electric Internet of Things

Youchan Zhu1, Yingzi Wang1,* and Weixuan Liang1

1
School of Control and Computer Engineering, North China Electric Power University, Baoding, China

Abstract: Background: With the further development of the electric Internet of Things (eIoT), IoT
devices in the distributed network generate data with different frequencies and types.
Objective: Fog platform is located between the smart collected terminal and cloud platform, and the
resources of fog computing are limited, which affects the delay of service processing time and re-
sponse time.
Methods: In this paper, an algorithm of fog resource scheduling and load balancing is proposed.
ARTICLE HISTORY First, the fog devices divide the tasks into high or low priority. Then, the fog management nodes
cluster the fog nodes through the K-mean+ algorithm and implement the earliest deadline first
Received: June 17, 2020 dynamic (EDFD) task scheduling algorithm and De-REF neural network load balancing algo-
Revised: October 27, 2020
Accepted: November 02, 2020 rithm.
DOI: Results: We use tools to simulate the environment, and the results show that this method has strong
10.2174/2352096514999210104144312
advantages in -30% response time, -50% scheduling time, delay, -50% load balancing rate, and en-
ergy consumption, which provides a better guarantee for eIoT.
Conclusion: Resource scheduling is an important factor affecting system performance. This article
mainly addresses the needs of eIoT in terminal network communication delay, connection failure,
and resource shortage. A new method of resource scheduling and load balancing is proposed. The
evaluation was performed, and it proved that our proposed algorithm has better performance than
the previous method, which brings new opportunities for the realization of eIoT.

Keywords: Power grid Internet of things, fog computing, data management, data cleaning, date integration, data storage.

1. INTRODUCTION data on energy consumption, resource utilization, or the sur-


rounding environment. These data are often not effectively
With the development of Cloud Computing (CC), Inter-
used. If the relevant enterprises or individuals can access the
net of Things (IoT), Big Data Analysis (BDA), artificial in- data and store and analyze it, it can be valuable information
telligence, and Cyber Physical Systems (CPS), people, ma-
[4].
chines, objects, sensors, and various smart devices can be
interconnected anywhere at any time [1]. Driven by these Cloud computing is a convenient "pay-as-you-go" meth-
technologies, the smart grid must pursue a higher degree of od for accessing and using shared resources, which can ef-
intelligence. It must have low-cost and sensing capabilities, fectively solve the problem of data explosion [5]. Cloud
advanced computing, data analysis capabilities, and network computing has large potential social and economic benefits,
physical integration capabilities. The equipment includes and users can obtain high-quality services at a lower cost.
electrical equipment, Internet of Things equipment, network For example, Amazon, Google, Ali, and Microsoft [6] have
communication equipment, mobile interruption equipment, developed cloud computing platforms to provide services to
electric vehicles, etc. [2]. The integration of various ad- global users [7]. There are also cloud computing models for
vanced smart devices and power grids is an inevitable choice business development in the energy industry. Still, compared
for the eIoT [3]. These devices generate a large amount of to other application scenarios, electricity is composed of four
major sources: “source, network, and storage” [8], with di-
verse data types, large data throughput, and real-time data
*Address correspondence to this author at the School of Control and Com-
puter Engineering, North China Electric Power University, Baoding, China;
processing. It has high sexual requirements, and the power
E-mail: liangwx1995@163.com system is related to the national economy and people's liveli-
hood. It also has high requirements for safety and stability

2352-0965/21 $65.00+.00 © 2021 Bentham Science Publishers


348 Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 Zhu et al.

[9]. However, delayed response during data transmission, Load balancing is an important method to improve re-
expensive high bandwidth of massive data, and inflexibility source allocation and system performance. The best load
of the transmission network topology will all become con- balancing is finding a balance between latency and power
straints on application service, quality, and user experience. consumption in the fog environment. Deng [21] et al. pro-
Cloud computing is not a universal solution for all IoT appli- posed a method for solving load balancing in a centralized
cations. scenario with minimal power consumption and service laten-
cy. The paper did not mention the response delay and trans-
Fog computing has emerged according to the require-
mission delay of the data transmission process, and the solu-
ment. The use of idle processing capabilities of heterogene-
tion which the centralized platform in this article proposes
ous terminals and edge devices to provide services has rapid-
may cause the failure of the entire system. Pham [22] et al.
ly developed in the manufacturing, energy, and transporta-
tion industries [10]. Fog nodes can be integrated to provide proposed a task scheduling method based on cloud and fog
collaborative computing. Users use heuristic collaboration
efficient management solutions for end-users because of the
between fog nodes and cloud platforms for large-scale data
flexibility to perform data transmission at the network edge
processing. The results show that the algorithm achieves a
on the network topology, services for billions of connected
balance between the cost of cloud resources and the maxi-
devices, and performance of real-time data processing for
mum cut-off time but does not take into account the re-
various applications. Fog computing usually uses a three-
layer architecture, namely cloud layer, fog layer, and percep- sources owned by the fog nodes, the execution of the Per-
formance, and energy consumption [23]. Choudhari [24] et
tion layer. The cloud layer is on the top, whereas the fog
al. proposed a priority-based task scheduling algorithm.
layer and the perception layers are arranged downward [11].
When the number of tasks is stable, the algorithm can handle
Cloud computing and fog computing are distributed compu-
the load well and reduce response time and system perfor-
ting methods, but cloud computing adopts a more centralized
mance. However, when the number of tasks is dynamically
approach, while fog computing follows a decentralized ap-
proach [12]. The cloud computing model relies on a fixed adjusted, the effect is bad. Dang [25] et al. proposed a fog-
based area and cloud (FBRC) architecture for processing
data center and is managed by the data center administrator.
local requests. When additional resources are needed, they
In contrast, the fog computing mode is more flexible. The
are handled jointly by multiple regions. Manisha [26] and
device is controlled and managed by different owners or
others proposed a load balancing algorithm for monitoring
institutions. Any device with computing power, storage ca-
real-time status in fog computing. The algorithm gives the
pacity, and transmission capacity can act as a device [13].
Therefore, it is challenging to schedule, process, manage and task a time period, and the task is processed only during this
period, which improves resource utilization and meets the
ensure data security on these devices, and the limited re-
needs of users with lower complexity, but increases the pres-
sources of fog devices make high-speed computing more
sure on network communications.
difficult [14]. Load balancing and efficient resource alloca-
tion help increase computing power and minimize overall Existing researches have considered a resource and delay
time. task scheduling but have not finished the problem of user's
dynamic changing behaviors within the time limit and re-
The purpose of scheduling is to optimize resource alloca-
source allocation during large tasks. To alleviate the prob-
tion and program CPU usage time and achieve low latency,
lems in the above studies, this paper focuses on the fog re-
high response, high reliability, high security, low energy
source scheduling problems solved by deep learning related
consumption, and cost [15]. The task scheduler analyzes the
technologies.
running devices and decides which tasks to run next. [16 ].
Task scheduling directly affects the performance of the 2. MATERIALS AND METHODS
entire system. In cloud-fog collaboration, Rahbari et al. [17]
proposed a backpack scheduling algorithm. They use the 2.1. Application Scenario
ifogSim simulator to optimize the search for symbiotic or-
ganisms. From the results, the network utilization and The power distribution monitoring system is mainly
equipment life have been improved using this algorithm, but composed of the state detection and video monitoring of the
energy consumption and costs have also increased. Backpack power distribution network and plays a remote control and
scheduling is suitable for low-load, time-sensitive applica- telemetry role [27]. By collecting and displaying various
tions. Yin et al. [18] proposed a fog resource scheduling status and power parameters in real-time, the operating status
method for intelligent manufacturing scenarios. This solution of the power distribution process can be grasped in time and
adopts the principle of minimum allocation of fog resources, accurately, and faults can be found quickly [28]. There are
reduces task delay, and can process more tasks in a short hidden dangers during the operation of the power distribu-
time [19]. However, when the network traffic is congested tion monitoring system, with few detection types, long delay
or, each node has many tasks, and a long execution time, the time, poor short-term prediction, and slow response time.
load of the fog node is very large. It is, therefore, easy to Once the fog node is overloaded or a disaster occurs, it can-
cause system paralysis. Madni et al. [20] proposed different not be done for the task. The response time and delay of the
heuristics to optimize task scheduling in cloud environments, task will increase significantly with the increase in the num-
including “first come, first served (FCFS)”, “Min-execution ber of tasks. Task scheduling and load balancing can achieve
time (M-ET)”, “Min-completion time (M-CT)” algorithms. rapid response according to the scale of the situation, thereby
However, the difference in actual scenarios makes the cor- achieving efficient and stable operation of the smart grid
rect selection of task scheduling algorithms complicated. [29].
Research on fog Resource Scheduling Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 349

Fig. (1). describes the service application of the EIoT in the power distribution network under a fog environment.

The fog dispatch-based distribution network monitoring algorithm reduces energy consumption and delay and en-
system can process real-time response applications locally sures the stability of the distribution network.
through agile control, data storage, and edge computing
functions, analyze the fog network load in the distribution 2.2. Problem Statement
network in the region, reasonably arrange power consump- This paper proposes a fog resource scheduling algorithm,
tion, and strengthen countermeasures. The response of the which must be accepted and processed before the delay of
power distribution system is fast. Fig. (1) describes the ser- the task requirements, mainly to solve the problem of insuf-
vice application of the EIoT in the power distribution net- ficient and scattered fog resources. We assume a large num-
work under a fog environment. ber of terminal IoT acquisition devices, intelligent communi-
The fog network is deployed in the entire power distribu- cation devices (fog gateways), a set of fog servers (fog
tion process. It can be integrated with the power distribution nodes, fog management nodes), and cloud data centers. The
monitoring system, such as power line monitoring, power smart terminal devices can access the fog gateway through
distribution room monitoring, and power distribution auto- an optical fiber, Bluetooth, WiFi, and other networks to pro-
mation monitoring in power distribution monitoring. The vide basic data for user requests. The fog gateway uses dual
terminal sensing equipment or monitoring equipment [29] fuzzy logic algorithms to divide each user tasks into high or
can be set in different power distribution rooms as required. low level tasks. The fog management node divides the fog
The collected data is transmitted to the fog gateway for pri- node into multiple clusters to perform tasks at different lev-
oritization of data according to wired or wireless networks. els. If the fog node cannot complete the user request before
The fog node then calculates and processes the collected the deadline due to insufficient resources or a poor commu-
data. The fog management node plays the role of task sched- nication environment, the current fog node will wait for task
uling and load balancing. The local fog data center imple- scheduling after sending a warning to the fog management
ments unified storage of video data, environmental data, node. If other fog nodes cannot complete the task request,
power data, and alarm data to ensure pre-warning and event they will send a message to the cloud center for processing.
management, medium suppression, and post-mortem re- When a fog node is overloaded in the cluster, the overloaded
view. node automatically triggers a load balancing algorithm. Fig.
(2) depicts the workflow in a fog environment.
The distribution network monitoring system based on fog
scheduling has the advantages of the short delay, local load 2.3. Fog Resource Scheduling Algorithms
balancing, fast response time, data filtering, and prepro-
cessing. It is unnecessary to transfer each raw data to the In this section, we will introduce resource scheduling al-
cloud, reducing the transmission bandwidth and delay re- gorithms for fog computing. Table 1 describes the main al-
quired for uploading to the cloud [30]. Use of the scheduling gorithms used in this paper.
350 Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 Zhu et al.

Fig. (2). depicts the workflow in a fog environment. (A higher resolution/colour version of this figure is available in the electronic copy of the
article).

Table 1. Fog resource scheduling description.

Device Device task Algorithm

Sensing terminal Data collection -

Task Priority
Fog gateway Dual fuzzy logic control algorithm
Categories

Fog node
K-means + clustering algorithm
Fog Management Node clustering

Task scheduling EDFD

Fog node Load balancing De-RBF

Cloud Delayed assignments EDFD

2.3.1. Hierarchical Algorithm for Double fuzzy Logic Con- In the first fuzzy logic control (FLC1), two input parame-
trol Tasks ters are considered, namely the task size (TS) and the task
A large number of tasks is needed to be processed in the arrival time (TAT). The second fuzzy logic control consists
fog environment. In order to improve the efficiency of task of two input parameters: the minimum task execution time
processing, this article preprocesses the tasks before assign- (minTeT) and the maximum task execution time (maxTET).
ing them to the fog nodes and divides the tasks into high and The smaller the task size, the shorter the arrival time, and the
low levels. High-level tasks are generally real-time tasks, smaller the task execution time, the higher the priority. The
which have strong time sensitivity and strong delay limits. calculation task priority is assigned to the fog node calcula-
Even if the deadline of the task is exceeded, it may cause tion through a dedicated fog gateway based on the outputs of
disaster. User application requests are represented by tasks. FLC1 and FLC2. The task classification rules are shown in
High-level tasks wait for the scheduling first. The traditional Table 2 below.
fuzzy logic algorithm has various problems, such as many
calculation rules, high complexity, and long waiting time. 2.3.2. Fog Node Clustering based on K-means++ clustering
We propose a dual fuzzy logic controller algorithm that puts Algorithm
the task into Q1 and Q2 queues to reduce delay and the task
response time. The algorithm flow is shown in Fig. (3) be- Clustering algorithms are a class of unsupervised ma-
low: chine learning algorithms that divide a dataset containing
nodes into multiple classes, each with an initialized central
Research on fog Resource Scheduling Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 351

Fig. (3). Double fuzzy logic control task classification algorithm.

Table 2. Task classification rules.

Input Output

TS TAT O1

low low high


FLC1
low high high
high low low
high high low
Input Output

minTET maxTET O2

low low high


FLC2
low high high
high low low
high high low

node. According to the calculated distance, the nodes closest and f (bi =1...n ) , it can be calculated by Euclidean distance for-
to each center are divided into different classes. Through the mula, as follows:
iterative method, the value of the cluster center is continu-
ously updated until the convergence function converges. The n
2
k-means algorithm is one of the typical partition clustering || f (a ) − f(b)||= ∑ f (a ) − f (b )
i i

algorithms. It has the advantages of simplicity and speed.


i =1 (1)
The communication between fog management nodes and This paper uses K-means ++ clustering algorithm to clus-
fog nodes has high bandwidth, low energy consumption, ter fog nodes, as follows:
easy configuration, and convenient clustering. The fog man- Step 1: Randomly select a fog node (FN1) from the fog
agement node periodically updates the load and current us- network (FNet)
age of each fog node through the fog node's monitor and
Step 2: Select a new fog node FN1, choose !"#$%&, the
determines the fog clustering method based on geographical !!
location, task priority, deadline, and resources available. probability is ! , ! is the shortest distance from a
!"#$%&' !
Considering the distance between the two fog nodes f (ai =1...n ) point to the closest center
352 Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 Zhu et al.

Step 3: Until we have k fog node FN={FN1 FN2 ,…, 2.3.4. De-RBF load Balancing Algorithm
FNk}
After clustering, the current load of a fog node may be
Step 4: Perform standard k-means clustering from the much larger than the load of other nodes. High-utilization
clustering center FNi, standard k-means clustering for the set fog nodes often have large energy consumption and small
of i ∈ {1...k} . throughput. The main purpose of load balancing is to de-
crease latency, improve system efficiency, and reduce energy
Step 5: For each sample !! in the data set, calculate its consumption in fog computing. We adopt a differential evo-
distance to the k cluster centers and divide it into the class lution radial basis function neural network (De-RBF) dynam-
corresponding to the cluster center with the smallest dis- ic load balancing threshold strategy. The algorithm which
tance. uses a radial basis function (RBF) algorithm achieves the
Step 6: Recalculate its cluster center for each category workload of all fog nodes evenly distributed. RBF neural
! network is a single hidden layer feedforward neural network.
!! , !! = !"!! ! .
|!! | It can use any suitable smoothing function to approximate
Step 7: Repeat Step 5,6 until the position of the cluster the input data set with a suitable neural network and obtain a
center no longer changes. function reconstruction with bounded errors, which can
avoid local minima problems and speed up the —learning
2.3.3. Earliest Deadline first Dynamic (EDFD) task Sched- speed to meet the latest delay requirements. The differential
uling Algorithm evolution algorithm is used for optimization, to solve the
width of the radial basis function of the hidden layer, and to
The EIoT scenario has many task types and a large num- simply and effectively enhance the ability of prediction ap-
ber of tasks. The fixed priority scheduling algorithm is not plications and processing nodes.
efficient. The fog management node performs task schedul-
The De-RBF structure used in this paper is shown in Fig.
ing on clustered fog nodes through the EDFD algorithm. The
(4). Our proposed radial basis function neural network with
EDFD algorithm dynamically determines the task priority
differential evolution consists of three layers: the input layer,
based on the start and end time of the task; that is, the small-
the hidden layer, and the output layer. De-RBF allocates
er the task size and the earlier the start and end time, the
resources to user tasks accordingly by determining the cur-
higher its priority. During task execution, task priorities will
rent demand. We enter the algorithm into the fog nodes and
change. HL represents a high-level task, LL represents a
calculate the load situation of each fog node. If the load of a
low-level task. Algorithm 1 describes the dynamic schedul-
fog node in the cluster exceeds the set threshold, the algo-
ing of tasks for fog nodes. When two tasks have the same
rithm will be activated. The task distribution and workload
deadline, the fog node chooses one of them arbitrarily. When
sharing among multiple fog nodes can be guaranteed.
the deadline of the task exceeds the requirement, the task
will be processed by the cloud center.
Algorithm 1: EDFD-based task scheduling algorithm

1: Begin Input: List of fog nodes ( F → f i =1...n ), Current user


task (S)
2: For all f in F do
3: Initiate HL tasks from Fog Scheduler at first
4: If f i is available to process !"#$
5: Do task execution
6: Else
7: Repeat step 3 until finding the appropriate fog node
8: Do task execution for all HL tasks
9: End If
10: If f i is available to process !"## then Fig. (4). describes the structure of the De-RBF neural network.
11: Do task execution
12: Else
We monitor the current load of M fog nodes in the input
13: Repeat step 3 until finding the appropriate fog node layer. We calculate the weight value based on the current
14: Do task execution for all LL tasks workload of the fog nodes in the hidden layer. We output the
15: End If predicted load balancing results among M fog nodes in the
output layer. The current load can be calculated by the fol-
16: End lowing formula:
Output: Tasks are scheduled to fog nodes and running task S
execution. ∑ S =1
A(TS )
L( f n )=
T (2)
Research on fog Resource Scheduling Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 353

Table 3. Simulation environment.

Simulation Tools Description

CPU ARM,2core@700MHz

Topology Full connection

RAM 2GB
System Configuration
Operating system Windows 7

Language Java

IDE eclipse

Number 64(8*8)

Bandwidth 10 Mbps

Storage 1Gb
Fog node
Cost of resources 3.0

Memory cost 0.05

Storage cost 0.001

Number 64
ifogsim Device
MIPS 1000

Collected terminal devices configuration RAM 10Mb

Bandwidth 1Mbps

RAM capacity 10

Fog gateway delay 10ms

Fog data center delay 40ms

Cloud Center delay 200ms

Among them, A (TS) represents the size of the task, S Where, L is a load constant. L is different in each cluster,
represents the grand total number of tasks executing on each and load balancing is triggered separately
fog node, T represents running time. The following formula
represents the average load of the fog cluster. 3. RESULTS AND DISCUSSION
N
{∑ j =1[ RET j ( s ) + ET ( s )]} 3.1. Simulation Experiment
AveLi =
N (3) In this section, ifogSim environment simulation is used to
Among them, AveL represents the average load in the evaluate the fog resource scheduling algorithm proposed in
this paper under the EIoT, and compared with other fog
cluster i , N represents the number of fog nodes in the cluster scheduling algorithms. Table 3 shows the settings and con-
i, represents the remaining time of the fog node j cur-
RETj figuration of the simulation environment.
rently executing the task, and ET(S) represents the execu-
tion time of all tasks.
We set up 64 collection devices, 64 fog nodes (servers), 8
Due to the dynamic nature of the EDFD scheduling algo- fog management nodes (servers), and 8 fog gateways. The
rithm, the threshold trigger calculation in load balancing is as acquisition device can send tasks to the fog environment at
follows: the same time through the fog gateway. The data requested
by each task is between 10-20Mb, the waiting time needs 10
DLEThreshold = L * AveLi (4) to 100 milliseconds, and the scheduling period is 100 milli-
seconds. Although the fog node can temporarily store data,
354 Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 Zhu et al.

Fig. (5). simulates network topology. (A higher resolution/colour version of this figure is available in the electronic copy of the article).

we still set up a data center in the fog environment and the Main symbols and instructions are shown in Table 4 below:
cloud environment to store task data. Depending on the task,
the fog node can choose a suitable data center for storage as
shown in Fig. (5). Table 4. Main symbols and instructions.

3.1.1. Parameter Definition


Symbols Represents Symbols Represents
In this paper, task response time, task scheduling time,
the number of fog
delay, load rate, and energy consumption are used as evalua- TR request time N FNs
nodes
tion indicators of the algorithm. Task response time is the
time it takes for the fog management node to initiate a task TBP processing time CT current time
allocation request and receive a response, including request start time of the
time and processing time. The task scheduling time is when RT response time SST
simulation
the task fog management node starts to assign the task and
STE task start time d delay
the fog node receives the task. It is the most important index
to measure the algorithm. Delay is the duration after a task is S ET end time of task C EC current energy con-
assigned to a fog node. The equipment load rate estimates sumption
the overall load distribution by measuring the current task last energy value
ST scheduling time LUT
amount of each fog node. The more balanced the load rate, update time
the stronger the task concurrency processing. System energy the number of
consumption refers to the total energy consumed by the sys- N Ts Energy total energy
tasks
tem, including collection terminal equipment, fog nodes, and
network components. The relevant calculation formula is
3.2. Comparative Analysis
shown below:
The previous literature method descriptions are shown in
RT = TR + TBP (5) Table 5.
ST = STE + S ET (6) 3.3. Result and Analysis

L = NTs / N FNs (7) 3.3.1. Response Time Comparison


Response time is the amount of time a user request takes
d = CT - SST (8) before the response is complete. Usually, less response time
is spent in a dynamic environment. Fig. (6) shows the com-
Energy = C EC + (CT − LUT ) (9) parison of the response time of this method with the previous
Research on fog Resource Scheduling Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 355

Table 5. Previous literature method descriptions.

Author (Paper) Instructions Disadvantage

Pham [32] Scheduling of large-scale applications in a delay- Not suitable for large-scale applications
and others constrained environment Task execution takes a long time to schedule

Yang [33] Reduce overall energy consumption, reduce instability


High energy consumption of centralized controller
and others and service delays

Butam [34] Allocate more memory and CPU execution time for Do not consider dynamic arrival of new requests
and others mobile users Long response time for task execution

Ningning [35] Increased number of users leads to high response


Load balancing for graph partitioning among fog nodes
and others times

Su [36] Simple structure and high latency


Simple fog architecture based on NSGA-II algorithm
and others Not suitable for complex topologies

Does not consider the current fog node for resource


Xu[37] Resource Allocation and Management in Fog Environ-
allocation
and others ment
Cause high network congestion

Topcuoglu[38] A fast scheduling method based on heterogeneous earli-


est finish time (HEFT) is proposed, and the critical path Priority-based task scheduling is more efficient
and others on the processor is considered

Wang[39] A multi population genetic algorithm (MPGA) consider-


Requires long resource scheduling processing time
and others ing load balancing and task scheduling

Fig. (6). Response time comparison. (A higher resolution/colour version of this figure is available in the electronic copy of the article).
356 Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 Zhu et al.

Fig. (7). Scheduling time comparison. (A higher resolution/colour version of this figure is available in the electronic copy of the article).

method, including BLA [31] and graph partitioning [32], mized through load distribution across the network. Fig. (8)
simple scheduling [37], HEFT [38], and MPGA [39]. shows our proposed method and the load balancing ratio of
the fog environment using DRAM [12]. The method of dy-
The response time of the user request depends on the
congestion of the fog node and the load of the fog node. namic resource allocation is calculated based on the number
of tasks on the fog nodes. When the multi-task arrival time
Load imbalance and the increase in the number of user re-
increases, the load of a specific fog node increases with the
quests for fog nodes directly affects the performance of
number of tasks, resulting in overload. Our proposed method
response time. The BLA [31] and graph partitioning algo-
calculates the current usage of each node and uses a load
rithms do not consider the current load of fog nodes, result-
monitor at each fog node. If the current load of one of the
ing in high response time. Considering the number of re-
quests, each request consists of 5 tasks. Each request takes fog nodes exceeds the threshold ( ≤ L ), a message will be
sent to the management node, and subsequent user requests
2500 milliseconds and gradually increases as the number of
will be distributed to another nearby fog node. The results
requests increases. The method proposed in this paper re-
show that the average scheduling time is reduced by about
quires 1800 milliseconds, which is 30% less than graph
40%.
partitioning and 45% less than BLA. Simple scheduling,
HEFT, and MPGA algorithms have low requirements for
task scheduling, which requires more time for scheduling.

3.3.2. Scheduling Time Comparison


The task scheduling problem is to minimize the total
execution time of the task through an effective scheduling
algorithm. Fig. (7) shows the performance of our proposed
method with CMaS [29], simple scheduling [37], HEFT
[38], and MPGA [39]. CMaS is a cost-time-aware schedul-
ing algorithm based on task deadlines. It can draw a schedul-
ing time map for a fixed number of tasks, but it increases
significantly with the number of tasks. Our proposed method
provides better processing network capacity even for a large
amount of input data, and the results show that the schedul-
ing time is reduced by about 20% on average.

3.3.4. Load Balancing Ratio Comparison Fig. (8). Comparison of load balancing rates. (A higher resolu-
In a cloud and fog environment, the number of tasks per- tion/colour version of this figure is available in the electronic copy
formed on fog nodes is limited, and load balancing is mini- of the article).
Research on fog Resource Scheduling Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 357

Fig. (9). Delay time comparison. (A higher resolution/colour version of this figure is available in the electronic copy of the article).

3.3.5. Delay Comparison


Fog computing should reduce latency issues to improve
scalability and reliability. Fig. (9) shows the comparative
delay of our proposed method in relation to the number of
NSGA-II [33] tasks. Fig. (9) (1) shows the comparative de-
lay of our proposed method with the number of fog nodes of
NSGA-II [33].
NSGA-II is simpler in nature, but when the number of
tasks and task size increase, the latency is greatly increased.
For 5 tasks, the latency of the fog node is 2000 milliseconds,
and for 30 tasks, the latency of the fog node is 7000 milli-
seconds. Our proposed method is stable and can organize a
large amount of data. For 5 tasks, the latency is 1500 milli-
seconds. For 30 tasks, the latency of the fog node is 2000
milliseconds.
In the NSGA-II algorithm, although the fog node re-
Fig. (10). Energy consumption comparison. (A higher resolu-
sources are reasonably used, the current load and failure
tion/colour version of this figure is available in the electronic copy
monitoring of the fog nodes are not considered. This is the
of the article).
main cause of network failure, and it is clear from Fig. (9)
(2) that the performance is poor. The delay increases with the
number of fog nodes, but in our proposed method, the delay CONCLUSION
varies slightly with the number of fog nodes.
Resource scheduling is an important factor affecting sys-
3.3.6. Energy Consumption Comparison tem performance. This article mainly addresses the needs of
EIoT in terminal network communication delay, connection
To compare the energy consumption, we consider the to- failure, and resource shortage. This paper proposes a delay-
tal energy required for task scheduling. Fig. (10) shows a based resource scheduling algorithm under fog collaboration.
comparison of the energy consumption of our proposed First, the fog device divides the collected data (task) into a
method with DEBTS for different numbers of tasks [32]. The high or low level based on the task size, arrival time, shortest
DEBTS algorithm reduces the energy consumption to a large execution time, and longest completion time. Then we use a
extent, but when the task length performed by the fog node is clustering algorithm to cluster the fog nodes. We rely on
large, the energy consumption increases significantly. For 5 EDFD and De-REF algorithms for task scheduling and load
tasks, the energy consumption is about 15000KJ, and for 30 balancing. This is done by extending the ifogsim toolkit to
tasks, the fog node consumes about 60,000 KJ. The load of simulate the fog environment, and evaluating the perfor-
our proposed method is more balanced, and the energy con- mance of the proposed algorithm in terms of response time,
sumption is lower. For 5 tasks, the energy consumption is scheduling time, load balancing rate, delay and energy con-
about 13000KJ. For 30 tasks, the energy consumption of the sumption, etc. Finally, we simulate the results of different
fog node is about 20,000 KJ. performance indicators. The evaluation performed proved
Therefore, our proposed method is superior to previous that our proposed algorithm has better performance than the
methods in reducing response time, scheduling time, load previous method, which brings new opportunities for the
balancing rate, delay rate, and energy consumption. realization of EIoT.
358 Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 Zhu et al.

LIST OF ABBREVIATIONS ACKNOWLEDGEMENTS


BDA = Big Data Analysis The authors thank Dear Prof. Youchan Zhu, and Yingzi
Wang for their valuable supports, and we also thank review-
CC = Cloud Computing
ers for their valuable comments.
CMaS = Cost-time-Aware Scheduling
REFERENCES
CPS = Cyber Physical Systems
[1] Y. Liu, "Study of data integration architecture for widearea distrib-
De-RBF = Differential Evolution Radial Basis uted power quality of power grid", In: 2018 International Symposi-
Function Neural Network um in Sensing and Instrumentation in IoT Era (ISSI), Shanghai,
China, 2018, pp. 1-6.
EDFD = Earliest Deadline First Dynamic http://dx.doi.org/10.1109/ISSI.2018.8538098
[2] L. Jin, L. Haosong, X. Zhongping, W. Ting, W. Shuai, W. Yutong,
eIoT = Electric Internet of Things H. Dongliang, K. Chunting, W. Jia and S. Dan "Research on wide-
area distributed power quality data fusion technology of power
FBRC = Fog-based Area and Cloud grid", In: 2019 IEEE 4th International Conference on Cloud Com-
puting and Big Data Analysis (ICCCBDA), Chengdu, China, 2019.
FCFS = First Come, First Served http://dx.doi.org/10.1109/ICCCBDA.2019.8725668
[3] L. Sun, K. Zhou, X. Zhang, and S. Yang, "Outlier data treatment
FN = Fog Node methods toward smart grid applications", IEEE Access, vol. 6, pp.
FNet = Fog Network 39849-39859, 2018.
http://dx.doi.org/10.1109/ACCESS.2018.2852759
HEFT = Heterogeneous Earliest Finish Time [4] J. Dai, H. Song, G. Sheng, and X. Jiang, "Cleaning method for
status monitoring data of power equipment based on stacked de-
IoT = Internet of Things noising autoencoders", IEEE Access, vol. 5, pp. 22863-22870,
2017.
M-CT = Min-completion Time http://dx.doi.org/10.1109/ACCESS.2017.2740968
[5] L. Wang, and Q. Liang, "Representation learning and nature en-
M-ET = Min-execution Time coded fusion for heterogeneous sensor networks", IEEE Access,
vol. 7, pp. 39227-39235, 2019.
maxTET = The Maximum Task Execution Time http://dx.doi.org/10.1109/ACCESS.2019.2907256
[6] Y. Ma, Z. Guo, Y. Chen, and L. Zou, "Multi-sourced data storage
minTeT = The Minimum Task Execution Time and index construction for equipment condition assessment", In:
2014 International Conference on Computational Intelligence and
MPGA = Multi Population Genetic Algorithm Communication Networks, Bhopal, India, 2014, pp. 681-685.
http://dx.doi.org/10.1109/CICN.2014.150
NSGA-II = Non-dominated Sorting Genetic Algo- [7] B. Lu, and W. Song, "Research on heterogeneous data integration
rithm for Smart Grid", In: 2010 3rd International Conference on Comput-
er Science and Information Technology., Chengdu, China, 2010,
RBF = Radial Basis Function pp. 52-56.
[8] L. Zhang, Y. Xie, L. Xidao, and X. Zhang, "Multi-source heteroge-
TAT = Task Arrival Time neous data fusion", In: 2018 International Conference on Artificial
TS = Task Size Intelligence and Big Data (ICAIBD), Chengdu, India, 2018, pp. 47-
51.
http://dx.doi.org/10.1109/ICAIBD.2018.8396165
CURRENT AND FUTURE DEVELOPMENTS [9] P. Chen, J. Liu, X. Liu, R. Zheng, and Y. Pan, "Research on tiered
storage method for big data of virtual information based on cloud
In future work, the optimization algorithm will further computing", In: 2019 International Conference on Smart Grid and
reduce the delay and the overall dependency and will realize Electrical Automation (ICSGEA), Xiangtan, China, 2019, pp. 308-
cloud-fog collaboration in the EIoT, show its application 311.
http://dx.doi.org/10.1109/ICSGEA.2019.00077
value, and verify the practical application of the algorithm in [10] U. Das, and V. Namboodiri, "A quality-aware multi-level data
this paper. aggregation approach to manage smart grid AMI traffic", IEEE
Trans. Parallel Distrib. Syst., vol. 30, no. 2, pp. 245-256, 2019.
CONSENT FOR PUBLICATION http://dx.doi.org/10.1109/TPDS.2018.2865937
[11] K. Wang, "Network data management model based on Naïve Bayes
Not applicable. classifier and deep neural networks in heterogeneous wireless net-
works", Computers Electrical Engineering, vol. 75, pp. 135-145,
2019.
AVAILABILITY OF DATA AND MATERIALS [12] Z. Song, Y. Sun, J. Wan, and P. Liang, "Data quality management
for service-oriented manufacturing cyber-physical systems", Com-
Not applicable. put. Electric. Eng., vol. 64, pp. 34-44, 2017.
[13] J. Fu, Y. Liu, H. Chao, B.K. Bhargava, and Z. Zhang, "Secure data
FUNDING storage and searching for industrial IoT by integrating fog compu-
ting and cloud computing", IEEE Trans. Industr. Inform., vol. 14,
The Fundamental Research Funds for the Central Univer- no. 10, pp. 4519-4528, 2018.
sities(No. 2018MS076,No. 2020MS120). http://dx.doi.org/10.1109/TII.2018.2793350
[14] J.I. Guerrero, "Heterogeneous data source integration for smart grid
ecosystems based on metadata mining", Expert Syst. Appl., vol. 79,
CONFLICT OF INTEREST pp. 254-268, 2017.
http://dx.doi.org/10.1016/j.eswa.2017.03.007
The authors declare no conflict of interest, financial or [15] K. Jia, X.M. Ju, and H.B. Zhang, "Research on big data fusion
otherwise. method of smart grid in the environment of internet of things", In:
Research on fog Resource Scheduling Recent Advances in Electrical & Electronic Engineering, 2021, Vol. 14, No. 3 359

4th International Conference on Cloud Computing and Security [28] J. Liu, and Q. Zhang, "Offloading schemes in mobile edge compu-
(ICCCS), 2018. ting for ultra-reliable low latency communications", IEEE Access.,
http://dx.doi.org/10.1007/978-3-030-00018-9_56 vol. 6, pp. 12825-12837, 2018.
[16] M. Cosovic, A. Tsitsimelis, D. Vukobratovic, and J. Matamoros, [29] X. Pham, N.D. Man, N.D.T. Tri, N.Q. Thai, and E. Huh, "A Cost
"5G mobile cellular networks: Enabling distributed state estimation and performance effective approach for task scheduling based on
for smart grids", IEEE Commun. Mag., vol. 55, 2017, no. 10, pp. collaboration between cloud and Fog computing", Int. J. Distrib.
62-69. Sens. Netw., vol. 13, pp. 1-16, 2017.
[17] D. Rahbari, and M. Nickray, "Scheduling of Fog networks with http://dx.doi.org/10.1177/1550147717742073
optimized knapsack symbiotic organisms search", In: Proceeding [30] Y. Yang, S. Zhao, W. Zhang, Y. Chen, X. Luo, and J. Wang,
of the 21st Conference of Fruct. Association, 2018. "DEBTS: Delay energy balanced task scheduling in homogeneous
[18] J. Ma, "Resource management framework for virtual data center Fog networks", IEEE Internet of Things J., pp. 1-11, 2018.
embedding based on software defined networking", Comput. Elec- [31] S. Bitam, S. Zeadally, and A. Mellouk, "Fog computing job sched-
tric. Eng., vol. 60, pp. 76-89, 2017. uling optimization based on bees swarm", Enterprise Inf. Syst., pp.
[19] L.F. Bittencourt, J. Diaz-Montes, R. Buyya, F.R. Omer, and P. 373-397, 2017.
Manish, "Mobility-aware application scheduling in Fog compu- [32] S. Ningning, G. Chao, A. Xingshuo, and Z. Qiang, "Fog computing
ting", IEEE Cloud Comput., vol. 4, no. 2, pp. 26-35, 2017. dynamic load balancing mechanism based on graph repartitioning",
http://dx.doi.org/10.1109/MCC.2017.27 China Commun., pp. 156-164, 2016.
[20] S.H.H. Madni, M.S.A. Latiff, M. Abdullahi, S.M. Abdulhamid, and http://dx.doi.org/10.1109/CC.2016.7445510
M.J. Usman, "Performance comparison of heuristic algorithms for [33] Y. Su, F. Lin, and H. Xu, "Multi-objective optimization of resource
task scheduling in IaaS cloud computing environment", PLoS One, scheduling in Fog computing using an improved NSGA-II", Wirel.
vol. 12, no. 5, 2017. Pers. Commun., vol. 102, pp. 1369–1385, 2017.
http://dx.doi.org/10.1371/journal.pone.0176321 PMID: 28467505 [34] H. Gupta, A.V. Dastjerdi, S.K. Ghosh, and R. Buyya, "iFogSim: A
[21] Y. Gao, H. Guan, Z. Qi, T. Song, F. Huan, and L. Liu, "Service toolkit for modeling and simulation of resource management tech-
level agreement based energy-efficient resource management in niques in the internet of things, edge and Fog computing environ-
cloud data centers", Comput. Electric. Eng., vol. 40, no. 5, pp. ments", Software Practice Expertise, WILEY, vol. 47, pp. 1275-
1621-1633, 2014. 1296, 2017.
[22] X.Q. Pham, and E. Huh, “Towards task scheduling in a cloud-fog http://dx.doi.org/10.1002/spe.2509
computing system", In: 18th Asia-Pacific Network Operations and [35] H.F. Sheikh, I. Ahmad, and D. Fan, "An evolutionary technique for
Management Symposium (APNOMS), 2016. performance-energy-temperature optimized scheduling of parallel
[23] T. Choudhari, M. Moh, and T. Moh, "Prioritized task scheduling in tasks on multi-core processors", IEEE Transact. Parallel Distribut.
Fog computing", In: ACMSE Conference Transactions Proceed- Syst., vol. 27, no. 3, pp. 668-681, 2016.
ings, 2018. [36] H.F. Sheikh, I. Ahmad, Z. Wang, and S. Ranka, "An overview and
[24] T.D. Dang, and D. Hoang, "FBRC: Optimization of task scheduling classification of thermal-aware scheduling techniques for multi-
in Fog based region and cloud", IEEE Trust- core processing systems, sustainable computing", Inf. Syst., vol. 2,
com/BigDataSE/ICESS., 2017. no. 3, pp. 151-169, 2012.
[25] M.P. Shelke, A. Malhotra, and P. Mahalle, "A packet priority inti- [37] C.C. Kai, C.Y.Hao, Yang Chih-Chieh, and Lee Jenq-Kuen,
mation-based data transmission for congestion free traffic man- "Switching supports for stateful object remoting on network pro-
agement in wireless sensor networks", Comput. Electric. Eng., vol. cessors", J. Super Comput., vol. 40, no. 3, pp. 281-298, 2007.
64, pp. 248-261, 2017. [38] H. Topcuoglu, S. Hariri, and M-Y. Wu, "Performance effective and
[26] S. Sharma, and H. Saini, "A novel fourtier architecture for delay low-complexity task scheduling for heterogeneous computing",
aware scheduling and load balancing in fog environment", Sustain. IEEE Transactions on Parallel and Distributed Systems., vol. 13,
Comput.: Inform. Syst., vol. 24, pp. 100355, 2019. no. 3, pp. 260-274, 2002.
[27] M. Mukherjee, Q. Zhang, R. Matam, C. Mavromoustakis, Y. Lv, [39] T. Wang, Z. Liu, Y. Chen, and Y. Xu, "Load balancing task sched-
G. Mastorakis, L. Constandinos, and G. Yunrong, "Task data of- uling based on genetic algorithm in cloud computing", In: 2014
floading and resource allocation in Fog computing with multi-task IEEE 12th International Conference on Dependable, Autonomic
delay guarantee", IEEE Access., pp. 1-1, 2019. and Secure Computing., 2014.
http://dx.doi.org/10.1109/ACCESS.2019.2941741

DISCLAIMER: The above article has been published in Epub (ahead of print) on the basis of the materials provided by the author. The Edito-
rial Department reserves the right to make minor modifications for further improvement of the manuscript.

Potrebbero piacerti anche