1 s2.0 S2452414X2300119X Main

Journal of Industrial Information Integration 38 (2024) 100546
Contents lists available at ScienceDirect
Journal of Industrial Information Integration

journal homepage: www.sciencedirect.com/journal/journal-of-industrial-information-integration
Industry 4.0 and industrial workflow scheduling: A survey

Kaya Souaïbou Hawaou a, *, Vivient Corneille Kamla b, Sonia Yassa c, Olivier Romain c,
Jean Etienne Ndamlabin Mboula d, Laurent Bitjoka e
a
Department Of Electrical Engineering and Industrial Automation, ENSAI, University of Ngaoundere, Ngaoundere, Cameroon
b
Department Of Mathematics and Computer Science, ENSAI, University of Ngaoundere, Ngaoundere, Cameroon
c
ETIS Laboratory, CNRS UMR8051, University of CY Cergy, Paris, France
d
Inria Nancy-Grand Est, CAMUS team – ICube Pôle API, France
e
Department of Electrical Engineering, Energetics and Automatics, ENSAI, University of Ngaoundere, Ngaoundere, Cameroon
A R T I C L E I N F O A B S T R A C T
Keywords: The Internet of Things (IoT) or Web 3.0 is a growing technology that allows devices to communicate with each
Industry 4.0 other as well as with the cloud. The Industrial Internet of Things (IIoT), or Industry 4.0, is the industrial
Workflows scheduling application of IoT. Industry 4.0 transforms production equipment into IoTs, which in turn transmit production-
Cloud computing
related data or their operational status to other connected devices or servers for rapid storage or processing. The
Edge computing
Fog computing
results of the processing are used in real-time for the control and supervision of the production system. With the
time constraints related to the availability of the processing results, companies have more and more computing
servers. These IoT application computing servers are the fog. While processing IoT application data, fog servers
require cloud intervention to offload tasks to free them up and provide data storage. Edge computing joins the
game when connected objects are themselves computing resources. One of the main challenges of these tech
nologies is the scheduling of dependent/independent tasks. Scheduling IoT workflows in the fog to optimize
quality of service (QoS) metrics often requires offloading some tasks to the cloud in order to meet customer
satisfaction and service level agreement (SLA) constraints. Therefore, many algorithms have been developed to
schedule workflows in the cloud-fog environment. To organize this research, we present a literature review on
scheduling workflows in cloud, fog, and edge environments. We outline a taxonomy that systematically classifies
existing approaches. Finally, we identify challenges for future work.
1. Introduction understanding of the manufacturing process while enabling efficient and

sustainable production [4]. One of the main reasons for IIoT adoption by
Recently, the Internet of Things (IoT) has been one of the most manufacturers, utilities, agricultural producers, and healthcare pro
popular technologies improving the quality of life while promoting viders is to increase productivity and efficiency through smart, remote
interaction between humans and objects [1]. IoT, dealing with many management [4]. By enabling monitoring down to the square foot
connection-based internet devices has become a predominant topic in through a dense sensor deployment, precision agriculture powered by
network applications [2,3]. The Internet of things paves the way for IIoT can help farmers better measure agricultural variables like soil
ubiquitous connected infrastructures supporting innovative services and nutrients, fertilizer used, seeds planted, soil water, and temperature of
promising greater flexibility and efficiency [4].The main goal of IoT is to stored produce, nearly doubling productivity [6,7,8,4]. Finding the ideal
provide various applications and services such as those for condition demand and carrying out specific tasks in robust computing
manufacturing, transportation, medical treatment, transfer instrument, ecosystems, like fog and cloud computing, while also providing simple
energy management, health care, and industrialization [5]. It benefits tasks for smart things to succeed, can help determine the limitations of
are not only interesting for customer applications but also for the in contemporary smart things, such as storage capacity, battery life,
dustrial domain [4]. As a subset of IoT, Industrial IoT (IIoT) covers the network devices, home appliances, and treatment capability [5].
areas of machine-to-machine (M2M) and industrial communication Cloud computing shows its limits, especially for industrial applica
technologies with automation applications, provides a better tions in real or near-real time [9]. Fog computing, a freshly developed
* Corresponding author.
E-mail address: [email protected] (K. Souaïbou Hawaou).
https://doi.org/10.1016/j.jii.2023.100546
Available online 20 December 2023

2452-414X/© 2023 Elsevier Inc. All rights reserved.
K. Souaïbou Hawaou et al. Journal of Industrial Information Integration 38 (2024) 100546
distributed computing environment, promotes IoT ecosystems [10]. Fog various services in real-time, which can greatly reduce the delay of
computing, which distributes intelligence devices based on exchange workflow processing and ensure the quality of service of real-time ap
able users, is an updated form of cloud computing [11,12]. Due to the plications. The cloud layer, responsible for handling non real-time and
vast scale, heterogeneity, and high latency seen in some cloud ecosys intensive-computation tasks, comprises cloud storage, cloud data center,
tems, traditional cloud-based IoT systems face difficulties [13]. Fog and cloud computing tool and provides remote services to the intelligent
computing is a technology set up to support the cloud in its operation factory. In this architecture, the further away the computing environ
and more precisely in the operation of IoT. The use of Fog computing in ment, the higher its computing capacity, the shorter the execution time,
the industrial context could notably improve the performance of several but the higher the latency. However, the closer the computing envi
applications such as maintenance, inspection, and logistics [9]. With Fog ronment, the lower its computing capacity and latency. The main
Computing, data is transmitted from the collection point to a gateway challenges of distributed computing environments include service
for processing and then back to the edge. It acts as a bridge, bringing placement, resource discovery, service migration, load balancing, task
together Cloud and Edge, collects data at the edge of the network and scheduling, resource allocation, energy efficiency, and quality of service
processes it in real-time, reduces bandwidth usage, and latency, and [1,19]. Among these resource management issues, resource and task
increases security through encryption at the source [14]. Edge scheduling challenges are important because of their effect on
Computing is any type of computing program that provides low latency improving overall system performance [19]. IoT and IIoT applications
closer to requests [15]. In cloud computing, processing, control, and require several heterogeneous resources for their execution, for which
storage data are sent to a centralized cloud, whereas in fog computing, sufficient resources must be allocated [20,19]. Efficient and practical
network management, storage, and computation are balanced between task scheduling in distributed computing environments with available
the local and the central levels [16]. Edge Computing reduces the resources can improve the efficiency and accuracy of the task execution
bandwidth used and the latency of information processing, because data process and maximize the utilization of resources such as CPU, memory,
streams are processed in real-time [17]. Distributed computing envi bandwidth, and minimize delay, cost, and energy consumption [19].
ronments must be constructed based on the computation capabilities
and locality in hierarchical architectures [18] as shown in Fig. 1 which
consist of 3 layers of computing: Infrastructure layer, Fog layer, and, 1.1. Motivation and research queries
Cloud layer. In the Infrastructure layer (IIoT/Edge layer), significant
IIoT devices such as smart terminals, smart grid sensors devices are Today’s industries face volatile raw material prices, fluctuating
employed. This layer promotes request for the accessing of certain de consumer demand and increased global competition. Customers expect
mands from Cloud datacenter and Fog nodes. The Fog layer includes higher quality and value than ever before, as well as more customization
devices with limited computing capability such as computers, routers, options, including the ability to configure selections and buy online.
servers. This layer can sense the requests of IIoT devices and provide Therefore, industries need to be aware of industry trends in order to
secure their position in the market.
Fig. 1. 3-tier fog-cloud architecture.
2
To do this, industries need to be at the forefront of technology (In 1.3. Paper organization
dustry 4.0) and therefore need a high-performance computing envi
ronment to analyze their data and determine effective strategies for The paper is organized as follows. Section 2 presents Industry 4.0, its
planning their activities or workflows on company resources so as to objectives, enabling technologies, and challenges. The next section
ensure consumer satisfaction while maximizing company profit. These discusses workflow scheduling and its taxonomy. Section 4, shows
computing environments can either be housed inside the company (Edge related work based on workflow scheduling and task scheduling. Sec
computing or other ressources) or can be housed outside the company tion 5 looks at the analytical discussion based on the work mentioned in
(Fog or Cloud computing) to augment local computing resources, in our paper. Section 6 presents the research queries and inferences. Ac
which case it can be called Fog-Cloud or Edge-Cloud computing. cording to identified research gaps, Section 7 discusses challenges for
The analysis of data or the determination of effective strategies for future research direction, and we finish with a conclusion in Section 8.
planning the company’s activities on the above-mentioned computing Fig. 2 shows the organization of the paper.
environments is an activity that brings additional costs to the company
and thus increases the cost of the services or products offered. These 2. Industry 4.0
costs can be measured in time, money and energy. The minimisation of
these different costs is an NP-hard problem [21] according to the liter 2.1. Industrial revolution
ature and its resolution is a real challenge for both companies and
researchers. Industrial revolution is a challenge faced by many companies and
Many algorithms for scheduling workflows in distributed computing organizations, progress is still climbing since the first industrial revo
environments are proposed in the literature by researchers [21,22,23, lution in the 1800s. From industry 1.0 to Industry 5.0, we have observed
24,25,26,27,5,28,29,30,31]. These different algorithms optimize qual multiple changes.
ity of service metrics such as makespan, energy, latency, bandwidth, The first industrial revolution started in 1780 at the end of the 18th
budget or cost for the execution of the workflow. century [32]. With the introduction of the steam engine, manual pro
There arises a need to provide a literature review based on the work duction techniques were replaced by mechanical ones [33]. This revo
of scheduling tasks of several types (industrial, scientific, etc.) in Cloud, lution is known as Industry 1.0 or mechanization.
Fog, and Edge environments. Most work on IIoT task scheduling in The internal combustion engine, as well as the development of the
Cloud-Fog environments deals with independent tasks, but, to our telegraph and telephone, led to the beginning of Industry 2.0, also
knowledge, there is almost no work on dependent IIoT tasks. We are known as electrification, in the late 19th century [33]. The technological
going to list some works on task scheduling in Cloud-Fog environments advances across all industries included the use of new materials, and
taking into account the type of tasks submitted while noting the chal energy sources such as coal, steam engine, petroleum, and electricity
lenges of these works. [32]. This revolution is a period of economic growth, because, as ma
Research questions about task scheduling in Cloud, Fog and Edge chines replace workers, the productivity of companies is increasing as
environments can be expressed as follow: well as the unemployment rate [34].
The third industrial revolution, also known as digital revolution or
• RQ1: What types of tasks/workflows are most scheduled in the automation was born in the second half of the twentieth century thanks
Cloud, Fog, and Edge computing environments? to the invention and use of computers and electronics, and the discovery
• RQ2: What are the most used QoS metrics in scheduling problems in of nuclear energy [33].
Cloud, Fog, and Edge computing environments? The first three revolutions were powered by disruptive new tech
• RQ3: what is the most common type of algorithm categorization used nology, they transformed the way society is living and worked,
to solve scheduling problems? improved the productivity and efficiency of companies, and revolu
• RQ4: Is the collaboration of Cloud, Fog and Edge computing envi tionized how goods were produced and how work was organized [35].
ronment suitable in the context of tasks/workflows scheduling? The significant growth of technologies start developing in industry 4.0
• RQ5: What is the most requested computing environment for tasks/ by implementing automation and data exchange in manufacturing
workflows scheduling? technologies.
• RQ6: What are the most widely used simulation tools for evaluating Industry 4.0 also known as digitization, is considered to be
tasks/workflows scheduling algorithms in Cloud, fog, and edge technology-driven. It aims to replace humans with machines and make
computing environments? manufacturing systems, warehouses and logistics smart [35].
• RQ7: What kind of data are used in the evaluation of algorithms in It is a use of advanced technologies such as artificial intelligence, IoT,
scheduling problem? robots, 3D printing, cloud servers, and machine learning [34].
• RQ8: What are the challenges of scheduling tasks/workflows in the By replacing the human heart with a robotics one, companies have
Cloud, Fog, and Edge computing environment? realized that technology alone cannot guarantee the functioning of a
structure, it needs human assistance to create significant value streams
The objective of this article is to answer all these questions by [33]. This observation gave birth to the 5th industrial revolution.
providing a state-of-the-art of works on tasks/workflows scheduling in Industry 5.0 or personalization is the upcoming technology of the
Cloud, Fog and Edge environments. previous generation designed for efficient and intelligent machines [34].
Does the birth of industry 5.0 mean that industry 4.0 is over? The
1.2. Research highlights and contributions fourth revolution is not over yet and will continue to have an impact on
many businesses. Our work will be oriented towards Industry 4.0, as it
The main contribution of this work are: aims at automation, smart manufacturing process improvement and
productivity/production optimization. It deals with the concepts of IIoT,
• Review of the concepts of industry 4.0 and workflow scheduling fog, smart production, and machine-to-machine communication (M2M).
• Classification and comprehensive review of existing works focusing Fig. 3 resume the industrial revolution, from industry 1.0 to industry
on scheduling algorithm, optimization metrics, publication year, 5.0.
simulation tool, computing environment, and type of job
• Presentation of an analytical discussion of the work done 2.2. Concepts
• Identification of the research challenges for tasks/workflows sched
uling in cloud-fog computing environment for future research work Industry 4.0 was first introduced in 2011 from a project in the high-
3
Fig. 2. Flow of the article.
Fig. 3. Timeline of industrial revolutions [32].
tech strategy of the German government, which promotes the comput conventional industry [40]. Industry 4.0 forces us to step aside and let
erization of manufacturing [36]. It is the intelligent networking of ma the machines work.
chines and processes for industry with the help of information and In the Industry 4.0 era, production systems, in the form of Cyber
communication technology [37]. Hatem Azzam [38] in his LinkedIn Physical Production System (CPPS), can make intelligent decisions
social network page defines Industry 4.0 as a further developmental through real-time communication and cooperation between
stage in the organization and management of the entire value chain “manufacturing things” [41], enabling flexible production of
process involved in the manufacturing industry. According to [39], In high-quality personalized products at mass efficiency [42,43,39].
dustry 4.0 is the intelligent networking of machines and processes for
the industry based on Cyber Physical System (CPS) – a technology that
achieves intelligent control using embedded networked systems. The 2.3. Main pillars or enabling technologies
German Chancellor Angela Merkel, said that Industry 4.0 is the
comprehensive transformation of the whole sphere of industrial pro Sensing, integration, and communication are the three major pro
duction through the merging of digital technology and the internet with cesses in an IIoT application. Network topologies and protocols are
employed for communication, while RFID tags are used for sensing [44].
4
The enabling technologies that are particularly significant for Industry delicate and complex tasks because they are outfitted with cutting-edge
4.0 adapted on [40] and [45] are listed below. These technologies relate software, artificial intelligence, sensors, and machine vision. They can
to the four main characteristics of Industry 4.0: Data, Interconnectivity, also identify, process, and respond to information they pick up from
Integration and Innovation. their surroundings.
2.3.1. IoT and related technologies 2.3.7. Augmented reality

This technology creates the independence of connected objects from They enhance the physical world with digitally generated visual or
humans. Most of the physical objects in Industry 4.0, such as devices, sensory information. They enable a variety of services, such as part se
robots, and products, use the key technologies of the Internet of lection in a warehouse, providing workers with realtime information to
Everything, Radio Frequency Identification (RFID), Machine to Machine improve decision-making and work procedures [46]. This idea is
(M2M), Industrial Internet of Thing (IIoT). It is an opportunity for In fundamental to Industry 4.0. Workers can view real-time IoT data,
dustry 4.0 because it can process information in real-time, locate a digital parts, maintenance or assembly instructions, training materials,
product at any stage of production, reduce operating costs by elimi and more using smart glasses or mobile devices with an augmented re
nating problems on the communication network, waste management, ality (AR) system when they are looking at a physical object, such as a
supply chain, production process optimization, and many others [34]. piece of machinery or a product.
2.3.2. Cloud computing 2.3.8. Additive manufacturing

It is a collection of vast quantities of data, and their analysis and Also known as 3D printing, is a process that takes a digital repre
exploitation, immediately on the factory floor. It allows large-scale sentation in 3D and produces the corresponding physical object layer-
processing, inexpensive and flexible for computing and storage [46]. by-layer [46]. Products can be stored as design files into virtual in
Cloud computing technology is offering efficient innovation and eco ventories and printed on demand, reducing transportation costs. Origi
nomics of the scale, it uses the internet to access, store and manage data nally used for rapid prototyping, 3D printing today has a wide range of
on remote servers [34]. The industrial cloud is a computing environment uses, including distributed manufacturing and mass customization. 3D
that runs and provides an enabling environment for industrial applica printing is applied to test the parts of a new product. Companies can now
tions [47,48,49,34]. The cloud houses the data that powers Industry 4.0 produce tools, fixtures, and manufacturing aids in-house within days
technologies, and the cyber-physical systems at its center use the cloud using 3D printing, increasing customer satisfaction
to communicate and work together.
2.3.9. Cybersecurity technologies
2.3.3. Big data Technologies to defend against cyber attacks on systems, networks,
It acquires a very large amount of data from various sources (IoT programs, devices, and data. No business, particularly industrial com
devices, factory equipment, ERP and CRM systems, weather and traffic panies, should pursue digital transformation without taking cyber se
apps), and these data must be processed quickly. Their complete curity into account at every stage and decision. Using a Zero Trust
collection and evaluation help support real-time decision-making. architecture and cutting-edge technologies like blockchain and machine
Massive data analyze and separate what is important from what is not, learning, companies can automate threat detection, prevention, and
helping to draw conclusions and support more effective knowledge response, and, reduce the risk of production delay and minimize the risk
transfer to achieve business objectives [46]. The information resulting of data breaches within their network.
from real-time data analysis improves decision-making and automation All of these smart devices are linked over the Internet and are ready
in supply chain management areas such as logistics, supply chain to enable Industry 4.0 applications [44].
planning, manufacturing, and procurement. Big data is used in opti
mizing warehouse operation; predicting demand (adjusting and opti 2.4. Applications
mizing its product portfolio); proactive upkeep (supply chain
management, non-conformity analysis, and security) Numerous IIoT projects have been implemented recently in a variety
of industries, including agriculture, manufacturing, healthcare, trans
2.3.4. Artificial intelligence (AI) portation, and building automation [44]. Since Industry 4.0 applications
AI uses computers to simulate natural intelligence in order to inter have these qualities—intelligence, dependability, security, sustainabil
pret external data and act intelligently with little or no human inter ity, and privacy—are referred to as smart [44]. Leading uses for industry
vention. Applications of AI technologies considerably improve the 4.0 according to [51] include:
manufacturing industry’s capabilities since they operate across a range
of business functions and levels, from staffing to product design, 2.4.1. Smart city
increasing efficiency, product quality, and worker welfare [50]. It allows The development of technology is leading to the development of
robots to manage more computational tasks and make independent many fields at the same time. To develop a future smart city, many
decisions based on environmental data in real-time for Industry 4.0 [50]. domains need to be smart, such as smart home, smart office, smart
institution, smart healthcare centers, smart agriculture and smart
2.3.5. Digital twins transportation. All these domains have different IIoT applications with
They can be used to test, simulate and optimize the organization and, different requirements. The development policy of smart city is based on
for example, reduce the time between a picked order and its shipment six factors: smart economy, smart mobility, smart environment, smart
[46]. It is a virtual simulation of a real-world process, product, and population, smart life and smart governance [52,44].
system based on IoT sensor data. In order to ensure they are making the
proper investments, firms can test new technologies and concepts using 2.4.2. Smart product
simulator software in a safe, virtual environment. Digital twins are used The incorporation of sensors and microchips makes Industry 4.0
in information transparency, technical assistance, and decentralized items intelligent. The Industry 4.0 design must incorporate current
decision. manufacturing processes to allow for human-product interaction [44].
Cars, watches, washing powder, and so on, are set to become “smart” as
2.3.6. Industrial robotics their makers attach sensors to their packaging that can detect when the
It is a technology that is prepared to perform production tasks and product is being used and can communicate with smartphones when
then run smoothly without human intervention. These robots can handle scanned, they are eliciting the question of invasion of privacy and,
5
consequently, personal safety [52] companies’ equipment will support proprietary or nationally recognized
data and communication protocols; as a result, competition and com
2.4.3. Smart manufacturing merce will suffer, and costs will rise. Contrarily, independent and widely
Smart industry includes distributed automated systems and robotics, accepted international standard communication protocols, data formats,
integrated with sensors, actuators, microchips, autonomous systems and and interfaces can guarantee interoperability between various industries
controllers. To monitor, manage and control such an environment, in and nations, promote the widespread adoption of Industry 4.0 tech
dustrial processes require ultra-reliable, low-latency communication nologies, and guarantee open markets for producers and goods around
[44]. These applications are characterized by latency, power con the world [40].
sumption and workload parameters. In the smart factory manufacturing
process, machines will have the ability to improve process through 2.5.5. Employment and skills development
automation and self-optimizing decision making [52]. Unemployment rates are likely to rise as low-skilled employees are
Industry 4.0 implementations have been shown in numerous demos, likely to become unemployed, and those who are able to evolve with
testbeds, and use cases over recent years, the majority of which take the technology will see autonomy, less drudgery and more interesting work.
shape of smart factories or parts of smart factories. Examples include the Employers need creative staff with decision-making and computer skills
Japanese Robot Revolution and Industrial IoT effort [53,39], the French [40]. Although numerous programs have been started to promote the
initiative Industrie du futur [54,39], and the network of business and development of ”digital skills,” young people may not be interested in
research organizations called SmartFactory [42,55,39]. digitizing the workplace [40].
2.5. Challenges 2.5.6. Latency

The latency of a network is the sum of transmission, processing,
The critical requirements of Industry 4.0 applications present many propagation, and queuing times. IIoT applications demand real-time
unresolved issues in capacity, real-time connectivity, remote mainte streaming rather than batch processing, and real-time communication,
nance, and communication network topology Across multiple industries and are time-sensitive [44]. These decentralized solutions work best
(smart agriculture, smart grid, smart cities, smart gadgets, and health with Fog computing because it provides low latency, dependable con
care) [44]. The challenge of Industry 4.0 included a high data rate for nectivity, and mobility [44]. Latency is influenced by a number of fac
various applications [56,34]. tors, including network architecture, resource allocation, node power
consumption, and node storage capacity [44]. It is essential to minimize
2.5.1. Resources and energy efficiency any delays in order to achieve low latency.
The efficient operation of heterogeneous smart objects depends on
the availability of memory and powerful computing. Due to the scaling 2.5.7. Handling data growth
problems with smart objects enabling IIoT applications, the network Businesses will encounter more data that is being produced at a
complexity is growing day by day [44]. According to their needs, various quicker rate and provided in more formats as more businesses rely on AI
smart gadgets assisting an IIoT application will consume a tremendous usage. AI algorithms need to be simpler to understand to sift through
amount of energy on various scales. Every new IIoT application has the these enormous amounts of data. These algorithms also need to be
difficulty of ensuring network QoS while minimizing energy consump capable of combining data that may be of various sorts and timeframes
tion of smart IoT devices, fog nodes, and clouds [44]. To reduce energy [57].
and resource consumption, it is required to review the manufacturing
processes and the design of machines and industrial plants [39]. 2.5.8. Security
In a smart factory, where various machines and devices are linked to
2.5.2. Data ownership and security one or more networks, flaws in any one of those devices could leave the
With so much data flowing through the network, companies need to system open to intrusion. Companies must prepare for both enterprise
know who owns the industrial data in circulation and ensure that their system vulnerabilities and machine-level operational vulnerabilities in
data will not be used by their employees and/or competitors without order to address this problem. Due to the fact that many rely on their
their knowledge. Smart services will rely on the data generated by smart technology and solution providers to identify vulnerabilities, businesses
devices during their manufacture and use [40]. For instance, auto are not completely equipped to handle these security threats [57].
makers are hesitant to disclose data generated by their vehicles for fear The main focus of Industry 4.0 is the optimization of intelligent,
that digital rivals may limit their earnings [40]. The implementation of adaptable factories, supply chains, and distribution models where ma
an AI algorithm requires data for training and testing. chines record and transmit more data to human operators and other
machines via machine-to-machine interactions. To increase efficiency,
2.5.3. Legal issues tracking, capacity utilization, manufacturing quality, and industry
Industry 4.0 is the era of digitization and autonomy of smart devices, development, smart devices and processes must constantly exchange
many questions arise on the legal front. Among these issues, we have the information about failures, components, inventories, various demands,
intellectual property of a product, the supervision of employees, and the and various orders. Industrialization is related to environmental pollu
manufacture of a defective product by a robot [40]. For example, how tion [35]. Adoption of industry 4.0 is required to follow industrial laws
should the courts decide who in the network is at fault if a product and regulations that can help smart machine’s autonomy. Therefore, it is
produced by an autonomous manufacturing system is unsafe or defec important to create smart manufacturing with a focus on sustainability
tive? The promotion of innovation through the protection of intellectual and reducing greenhouse gas emissions [35].
property and the dissemination of knowledge, both of which are sources
of future advancement, have been urged to coexist in harmony by the 3. Workflows scheduling
French Conseil d’analyse economique [40].
Any program or application can be divided into several interdepen
2.5.4. Standards dent tasks which exchange a huge amount of data between them, this is
Standards must ensure the exchange of data between machines, commonly known as workflow application [58,21]. Workflows sched
systems and software within a production line as products enter the uling is one of the important issues in Cloud, Fog and Edge Computing
smart factory along the supply chain by integrating robots into the which considers Quality of Service (QoS) requirements such as deadline,
manufacturing process [40]. Only one company’s or group of cost and energy consumption. The workflows scheduling problem is
6
characterised by the fact that it features a group of complex and con Pegasus [71] is a workflow management system based mainly on
flicting objectives [21]. data-intensive scientific applications. It has developed some data man
agement algorithms in the grid environment and uses the RLS system as
3.1. Definitions and types of workflows data management at run-time. Data are moved asynchronously in order
to reduce runtime latency and dynamically delete to reduce storage
Hollingsworth [58] defines workflows as the computerised facilita usage [71]. Its features make it unique (portability in time and space,
tion or automation of a business process, in whole or part. A workflow is data reuse, and automated data management).
the execution of several interdependent tasks under a permanent de Kepler [71] is a scientific workflow management system. It is an
pendency constraint, where a task can only start its execution if the open source Java framework in the grid environment. The difference
execution of its parents are completed [25]. Scheduling problem being a between business and scientific workflows is controlflow orientation
relatively old one, and refers to the action of allocating resources to the and dataflow orientation. It has its own actor-oriented data modeling
processing of tasks. To do this, many approaches have been proposed to method for handling big data in the grid environment. The grid actors,
solve it in different platforms [59,60,61,62,63,26,29,30,64,31]. Ac called FileFetcher and FileStager respectively, use GridFTP to retrieve
cording to Yassa et al., [21] scheduling a workflow application is the files from, or move files to, remote locations on the grid. It adopt the SRB
process of assigning every task of the workflow to a given resource for its system for runtime data management [71].
execution. Workflow scheduling is one of the most important issues in Taverna [71] is a scientific workflow system for bioinformatics that
distributed computing environments, which requires the allocation of provides a new process definition language, Sculf, modeling the appli
VMs to corresponding workflow tasks based on different functional and cation data in a dataflow. It considers workflow as a graph of processors,
non-functional requirements [65]. Workflow scheduling can be transferring a set of data inputs into a set of data outputs [71].
single-objective or multi-objective. Askalon [71] is a workflow system designed for scheduling. It ag
Two broad categories use the notion of workflow, we have the gregates the computing overhead and data transfer overhead together to
experimental protocols (scientific workflows) and the treatment chains get a value ‘weight’, and, makes no distinction between the computing
practiced in commercial, financial and pharmaceutical fields (business resource and data host. It has its own process definition language called
processes) [63]. AGWL [71].
A simple workflow is represented by a Directed Acyclic Graph (DAG)
which consists of vertices and edges. Each edge is directed from one 3.3. Optimization metrics
vertex to another, so that there is no loop in the process.
Ludäscher et al., [66] define a scientific workflow as the description The amount of data to be transferred, the optimization of conflicting
of a process for accomplishing a scientific objective, usually expressed in Quality of Service (QoS) metrics while taking into account user and
terms of tasks and their dependencies. Scientific workflow tasks are provider requirements, and the respect of SLAs make scheduling a
computational steps for scientific simulations or data analysis steps, difficult problem to solve. Workflow scheduling in cloud, fog, and edge
their common elements or stages are acquisition, integration, reduction, computing is an NP-hard problem because it’s seen as a combinatorial
visualization, and publication of scientific data [66]. They can be problem, and is impossible to find the optimal global solution in a
organized (at design time) and orchestrated (at runtime) according to polynomial way [63]. It is difficult and time-consuming to find the
dataflow and possibly other dependencies as specified by the workflow optimal solution with exact algorithms, reason why algorithms from the
designer, and, can be designed visually, using block diagrams, or metaheuristics, machine learning family are used to find satisfactory
textually using a domain-specific language [66]. We have several spe solutions in a reasonable time. Scheduling workflow in Cloud, Fog, and
cific scientific workflow applications for various research fields, applied Edge computing environments means allocating the best possible set of
on the different computational layers. Some workflow applications may resources for the workflows considering various objectives such as
be better executed at the cloud layer, others at the fog layer. Some ap deadlines, costs, energy, and QoS [27]. The objectives of workflow
plications benefit from cloud-fog-edge collaboration to achieve better scheduling is to optimize the job execution process for maximum
performance. On the cloud and fog layer, we find Montage [67], in as resource usage of distributed computing environments, and, the aim of
tronomy, used to build large mosaics of sky images; Cybershake [67], in multi-objective optimization is to provide an optimal solution for a set of
seismology, used to calculate probabilistic seismic risk curves for objectives that can be contradictory [19]. The optimized metrics in the
geographical sites in the Southern California region; Epigenomics [67], workflow scheduling problem are presented below:
in bioinformatics, automates the execution of various genome
sequencing operations; SIPHT [67] performs an extensive search for • Makespan: It is a significant objective that indicates the overall time
small untranslated RNAs that regulate several processes such as secre required to finish a complete workflow [72,19]. The minimization of
tion or virulence in bacteria, Defog [68], in industry, is used to evaluate makespan leads to fast execution of the applications and can help in
industrial workflows, while the Bosch Group dataset [69] is used for measuring the quality and efficiency of task/workflow scheduling
industrial manufacturing in production plants. On the fog layer, Work [19]. It can be computed by using the following equation [19]:
flow Recognition dataset [70] is used for tracking and recognition by
Makespan = FT − ST (1)
video surveillance of the production chain in a complex industrial
environment.
The execution of a workflow is managed by a workflow management
system [63]. Where FT denotes the Finish time of the last task and ST is the
starting time of the first task.
3.2. Workflows management systems
• Execution time: It is the time taken by a system to execute the task
A workflows Management System is a system that completely de [19]. The waiting time is not included in CPU or execution time [73,
fines, manages and executes “workflows” through the execution of 19]. It can be expressed like:
software whose order of execution is driven by a computer representa
tion of the workflow logic [58]. It must be robust against performance ExecutionTime = FTi − STi (2)
variations and tolerant against failure. The execution of the applications
is automated through workflow technologies [71]. Some of the popular
and successful Workflows Management System are listed below:
7
Where FTi is the finish time of the taski and STi is the starting time of randomly and are unknown beforehand. The tasks to schedule arrive
the taski. randomly and the VMs are created and destroyed randomly, ac
cording to the submitted tasks.
• Cost: Financial costs include processing, communication, network • Mobility: Mobility is a key element to improve user experience on
usage, and energy consumption cost [19]. The client needs to pay the fog applications, it poses several challenges, including managing
provider. It can be expressed by using the following equation [63]: user preferences and time and distance constraints [80,19]. Alloca
tion and scheduling in this environment are challenges to meet user
cost = Cex + Ctr (3)
demands based on user preferences and QoS requirements for
real-time applications [81,19].
• Security: The security problem is essentially a problem of trust
When Cex is the execution cost and Ctr is the transfer cost. where the main issue is the transparency of transport, storage and
processing of data in these environments, it must be guaranteed,
• Energy consumption: The cloud-fog environment must be efficient whether on the network or in the Datacenter where they will be
for users and profitable for providers. Providers are under pressure to stored. Because of its mobility, heterogeneity, and extensive
reduce energy consumption and carbon dioxide. All devices consume geographic spread, the fog environment is less safe and more prone
energy even in an idle state, and this consumption increases when to security breaches than cloud computing [78]. This issue impacts
the device is used [19]. the privacy and confidentiality of user data that travels between
• Latency: Latency is one of the most important metrics to measure the distributed computing environments and can be accessed by un
performance of any task scheduling algorithm, it is also known as wanted users.
delay or response time [19]. The total latency is the sum of trans
mission latency and computational latency with as transmission la Workflow scheduling is a challenging problem due to resource lim
tency, the communication delay to transfer data between resources itations, heterogeneity, unpredictable arrival rate, and the sheer volume
and as computational latency, the necessary time to processed task of tasks that need to be completed. To solve this issue, numerous
[74,19]. research efforts have been made to develop effective scheduling tech
• Reliability awareness: is the probability that the task can be suc niques that maximize the use of resources and various performance
cessfully completed within the users quality of service constraints, indicators.
with or without resource and/or task failures [75].
• Throughput: Is the total number of tasks performed per unit time in 3.5. Taxonomy
a system [76,19]. It can be computed as follow:
Taxonomy is a scheme of classification and categorization of things
Throughput = NumberOfTask / Makespan (4) into groups or types. Some taxonomies of workflow scheduling, grouped
by scheduling algorithms and scheduling schemes are presented below:
• Resource utilization: Maximizing the use of resources such as VMs 3.5.1. Scheduling algorithms
is important, it is necessary to keep the resources as busy as possible We present a taxonomy of scheduling algorithms based on existing
in order not to waste money, as they are paid for but not used [76]. methods that consider different optimization metrics; we group the
Application failure can be caused by poor resource utilization different algorithms into mathematical programming, heuristic, meta
resulting in decreased system performance [19]. Workflow sched heuristic, hyper heuristic, hybrid heuristic, and machine learning shown
uling algorithms should efficiently allocate resources like processor, in Fig. 4.
network bandwidth, and memory to the tasks.
• Quality of Service: Metrics such as security, user experience, and • Mathematical programming or mathematical optimization
quality of experience are not quantifiable and affect user experience (MP): Mathematical programming is a particularly helpful tool for
[77,19]. Better quality of service results in the satisfaction of several tackling complex problems, such as those that can be characterized
users of the system at the same time [19]. as an objective function with a set of mathematical constraints, that
simplifies complex decision-making [82]. It is a discipline within
3.4. Challenges in workflows scheduling applied mathematics that, in essence, allows you to make the optimal
choice from a set of alternatives [83]. One of the most common ap
The challenges faced by workflows scheduling in cloud-fog envi plications of mathematical programming is to solve various planning
ronment can be grouped into: and scheduling problems by finding optimal ways of using limited
resources in a reasonable time to achieve a goal using input values.
• Heterogeneous fog applications: Numerous Internet of Things To do this, it uses a number of different techniques, including Integer
(IoT) devices and sensors are available, each with unique commu Linear Programming (ILP) and Mixed Integer Linear Programming
nication radio, sensor, computing, storage, and other capabilities (MILP) [82,83].
[78]. Among fog applications, we have those that are fault-tolerant - ILP: It is a technique in which one or more decision variables are
and others that are real-time. Real-time applications have low la constrained to take integer or binary values [19], the objective and
tency requirements, therefore, need to be executed on priority [19]. each constraint are a linear function [82,19]. Research paper that
Resource allocation becomes an NP-hard problem when one wants to used this method are: [84,85]
satisfy many user’s dynamic requirements in real-time with limited - MILP: It aims to find the minimum or maximum value of a linear
resources capacity [79,19]. Choosing the right resources and man function on a space with linear constraints with certain decision
aging and coordinating such heterogeneous IoT device networks will variables which are either integers or continuous numbers [86].
prove to be very difficult [78]. MILP are widely used in industry because they can be solved in a
• Stochastic environment: The Fog-Cloud environment is stochastic reasonable time using efficient algorithms. Some researchers used
in many approaches. Allocating the limited resources of fog nodes to this method to solve scheduling problem: [87,88].
IoT workloads according to their computational needs is a challenge
[19]. As an illustration, the rate at which the tasks arrive, how long MLP problems are much harder to solve than ILP problems [82].
they takes, and the amount of processing required all change However, MP shows its limits when input values are unknown (as MP
8
Fig. 4. Taxonomy of scheduling algorithms based on [19].
does not take into account uncertainties and changes), when there are designing methods for learning parameter settings, the basic idea is
too many limitations, when the problem size is large, and when the to develop a general method that operates on small moves and
problem has non-linear effects. learns which motions should be applied at each step of the solution
process [95]. It is Hyper heuristic method.
• Heuristic: The term heuristics derives from the ancient Greek heu • Hyper heuristic: The term hyper heuristic was defined in the early
riskeinˆand means ”to find” [63]. It is a concept used in combinatorial 2000s as a heuristic algorithm chosen instead of many from a set of
optimization, graph theory, complexity theory of algorithms, artifi candidates heuristics, but the idea of designing high-level heuristic
cial intelligence, etc. Heuristic algorithms, powerful and flexible methodologies goes back to the early 1960s [96]. Hyper heuristic
methods of solving NP-hard problems, examine the whole environ techniques have been employed to develop approaches that are more
ment to return the best solution with the objective of finding the best general than optimization search methods and traditional techniques
and most acceptable answers according to the problem requirements [97]. It is an automated search procedure that improves the gener
[89]. Several researchers have adopted heuristic methods as alization of search techniques to solve difficult computer search
problem-solving strategies, such as Ijaz et al., [28] that proposed problems [96,19]. it is used in order to develop a system for solving a
EM-MOO, Rahbari and Nickray [60] that developped KnapSOS, class of problems instead of solving a particular domain of a problem
Stavrinides and Karatza [90] that proposed a Partial Computation [19]. Kabizardeh et al., [98] and Rahbari [99] used hyper heuristic
method. Heuristics are specific methods that do not guarantee exact algorithm for solving scheduling problem. Although hyper heuristic
solutions and deal with single or bi-objective problems. These algo algorithms can solve complex real-world problems, they are difficult
rithms are fast, but they are insufficient for distributed systems with to implement [19].
a large-scale dynamic environment [91], they are greedy. Meta • Hybrid heuristic: A hybrid heuristic algorithm combine two or
heuristic strategies are effective methods for solving NP-hard opti more heuristic or metaheuristic algorithm by choosing the advan
mization problems with high efficiency. These algorithms can be tages of each of them and compensate the limits at each step. It gives
expressed as Heuristic + Randomization [91]. better results than a single heuristic [100]. Hybrid algorithms are a
• Metaheuristic: In the last few years, meta-heuristic algorithms have combination of heuristics and metaheuristics, or heuri
gained popularity and are commonly used for solving complex stic/metaheuristic algorithms, or multi-agent system/ (meta)heu
computational problems [92,19]. Meta-heuristic algorithms have ristic. The multi-agent approach uses the installation of a software
proven to be very efficient in solving scheduling problems [92]. agent on a set of nodes of the infrastructure to be managed, the
Researchers use them to find optimal or near-optimal solutions for scheduling work being distributed among the agents [101,102,103,
allocating tasks in a distributed computing environment [93,19]. It 59]. It was implemented in order to decentralize the scheduling of
can be categorized as [19]: workflows [59].Some researchers used hybrid strategy for solving
- Bio-inspired (BI)-based meta-heuristics: These algorithms scheduling problems [104,26,29,30,31,64].
mimic the evolutionary process of biological organisms observed in • Machine Learning (ML): Machine learning is a technique that is
nature to solve optimization problems. As an example of bio- gaining popularity because it can produce results without human
inspired algorithms, we have GA [63], Tabu Search (TS) [93], SA intervention by integrating mathematics, statistics, and machines to
[93]. learn from raw data [105]. As a subsection of ML algortihm, we have
- Swarm intelligence (SI)-based meta-heuristics: Swarm optimi Deep Reinforcement Learning (DRL). DRL is the result of the fusion
zation algorithms are inspired by the collective behavior of swarms of deep learning and reinforcement learning that emerged as an
and the computation of their evolution. PSO, Bee Life Algorithm active area of research, it has proven its ability to solve decision
(BLA), and ACO are some examples of SI-based algorithms [94,19]. making problems where the magnitude of the mapping is too large,
However, using existing algorithms to solve new problems or new the deep neural network (DNN) on the other hand helps to approx
instances of a similar problem is not easy, thus, in many cases, it is imate functions, if used as such (DNN), a DRL model can be created
necessary to build new algorithms to solve particular instances [19]. [106] and [107] used DRL algorithm to solve scheduling
[95]. Recently, researchers have tried to automate the process of problem.
9
3.5.2. Scheduling scheme problem in distributed Computing environments are presented below:
The large-scale distributed system includes many subscribed re
sources to perform tasks faster, stably, accurately, and available [108]. 4.1.1. Integer linear programming (ILP)
The growth in the number of resource providers increases the growth in Scheduling workflows in multi-cloud environments reduces resource
system heterogeneity, which is a positive factor, while the increase in operating costs for customers and resource limitations imposed by ser
decisionmaking complexity has a negative effect on the decision-making vice providers. To contribute to this solution, Mohammadi et al., [84] in
process [109]. The scheduling scheme in the different layers of 2018 analyzed the optimal cost treatment as a function of variations in
computation based in Fig. 5 is presented below. execution time and scientific workflow size in multi-cloud environ
ments, studied the sensitivity of optimal cost to variations in delay and
• Cloud computing: Task and Workflow scheduling problem are workflow size, and, proposed an integer linear programming - delay
aimed to allocated an appropriated VM to each task taking into ac constraint (ILP-DC) model to schedule these workflows. The proposed
count the user-specified constraints (deadline, costs), Cloud Service mathematical model was implemented in CPLEX solvers, with the
Providers (CSP) specified constraints (power usage, temperature, characteristics of the multi-cloud machines used drawn from IaaS pro
SLA violation), functional and non functional requirements of ap viders Amazon EC2, GoGrid, BlueLock and StormCloud. Simulations
plications [110]. were carried out in the workflow management systems Epigenome,
- Task scheduling: It can provide benefits to the users by reducing LIGO, Montage and CyberShake, whose characteristics were taken from
execution cost, mitigating makespan, decreasing SLA violation and the Pegasus workflow generator. Simulation results compared with
meeting user-specified deadline; and the CSP benefits are a follows: coarse-grained mixed integer programming (MIP-CG), IP-FC and
reducing power consumption, scalability, fault tolerance, improve multi-cloud partial critical paths with preprocessing (MCPCPP) show
ressource utilization [110]. that the ILP-DC approach is better than these approaches in terms of
- Workflow scheduling: In addition of before-mentioned con minimizing the cost of delay-constrained scientific workflows in a
straints, it consider dependency among tasks, data transfer costs, multi-cloud environment.
security considerations regarding the data transfer, and so on; it In 2021, Movahedi et al., [85] proposed an Opposition-based Chaotic
can provide the same benefits with task scheduling, some workflow Whale Optimization Algorithm (OppoCWOA), which is an improved
scheduling schemes can specify the order of tasks, and even in version of the Whale Optimization Algorithm (WOA), with a view to
heterogeneous clouds, the type of VM can be determined [110]. solving the task scheduling problem in Fog-Cloud environments
modeled in a timely manner. This scheduling problem is formulated as a
However, it is worth mentioning that task and workflow scheduling mathematical integer linear programming (ILP) optimization model
approaches may pursue different objectives in each kind of cloud envi taking into account time and energy consumption in the Fog. The so
ronment [110]. lution was simulated under the Pycharm IDE and developed in Python.
Results compared with WOA, Artificial Bee Colony (ABC), Particle
• Fog Computing: A fog node receives a lot of task that need to be Swarm Optimization (PSO), and Genetic Algorithm (GA) show the ef
scheduled before their deadline. With the rise of the Internet of ficiency of the proposed solution in terms of time, energy, and
Things, scheduling tasks and workflows in a Fog environment has time-energy balance. However, the simulation testbed has not yet been
become an important research topic. Task and workflow scheduling fully pre-validated.
depends on many criteria based on the user requirements, including
satisfying QoS. 4.1.2. Mixed integer linear programming (MILP)
• Task scheduling: Task scheduling aims to assign a set of tasks to fog To minimize the cost of mapping resources and networks between a
nodes to satisfy user QoS requirements while minimizing makespan fixed IoT set and the constructed fog infrastructure, Martinez et al., [87]
and task transmission time [111,112]. in 2020 proposed an optimal formulation of fog infrastructure design
- Workflow scheduling: Workflow scheduling aims to assign and sizing using the Mixed Integer Linear Programming (fog-DC-MILP)
computing resources with different processing abilities to the tasks method. This approach, simulated in CPLEX solver, and compared with
of workflow application, which can minimize the makespan and the greedy algorithm fog-DC-Greedy and matching-based model
cost [113,112]. Fog-DC-Match, showed its effectiveness in optimizing latency and cost.
Santos et al., [88], in 2021, proposed a Mixed Integer Linear Pro
4. Related survey gramming (MILP) formulation for the IoT service allocation problem on
different Low-Power Wide Area Networks (LPWAN) that takes into ac
This section presents the obtained results from an in-depth study of count the concept of Service Function Chain (SFC) in the Cloud and
the research literature based on the classification of scheduling methods. several optimization objectives (latency minimization, minimization of
Table I, gives a multidimensional classification of these existing results. compute node utilization, minimization of service migrations,…).
Different objectives are taken into account at each iteration of the
model, and each iteration refines the previous solution by improving
4.1. Mathematical programming (MP)
another objective function. This model approach is generic, takes into
account many aspects of Cloud and wireless computing, and can be
Research work using MP methods to solve task/workflow scheduling
Fig. 5. Taxonomy of scheduling scheme in the layers of computing.
10
Table I
List of related works.
Authors Year Categorization Algorithms Metrics Tools Types of jobs Computing
environment
Mohammadi et al., 2018 MP ILP-DC Cost CPLEX Scientific Cloud

[84] Workflow
Movahedi et al., [85] 2021 MP OppoCWOA Time, Energy Pycharm Task Cloud-Fog
Martinez et al., [87] 2020 MP fog-DCMILP Latency, Cost CPLEX Task Cloud-Fog
Santos et al., [88] 2021 MP MILP Latency, Energy, User Request, Active CPLEX Task Cloud-Fog
Nodes, Service Migration, Transfer Time,
Active Gateways
Mora et al., [114] 2015 Heuristic ICS Deadline, Bandwith Not Defined Task Cloud-Edge
Mao et al., [115] 2017 Heuristic SOJTOSTPAA Latency, Energy Not Defined Flow shop Edge
Rahbari and Nickray 2017 Heuristic KnapSOS Energy, Cost, Sensor Lifetime, Network iFogsim Scientific Fog
[60] Utilization Workflow
Toniolli [61] 2020 Heuristic Adaptive PCH Makespan, Cost Hardware Scientific Cloud-Fog
Environment Workflow
Ijaz et al., [28] 2021 Heuristic EM-MOO Makespan Energy Matlab Scientific Cloud-Fog
Workflow
Stavrinides et 2021 Heuristic PC Deadline, Error Propagation Platform Developped Scientific Fog
Karatza [90] by theAuthors Workflow
Bentabet et al., [62] 2022 Heuristic BW-AWA Makespan, Energy, Bandwidth iFogSim Scientific Fog
Workflow
Ndamlabin et al., 2022 Heuristic SMWSO and Energy, Cost, Cloudsim Scientific Cloud
[25] SMWSH Deadline Workflow
Guan et al., [116] 2007 Metaheuristic SA Makespan, Increasing yield Phone equiped with Scientific Web server
GPS Workflow
Yassa et al., (a) [21] 2013 Metaheuristic GA* Makespan, Cost, Availability, Reliability, Cloudsim Scientific Cloud
Budget, Deadline Workflow
De Maio and 2020 Metaheuristic MOWO Execution time, Cost, Reliability SLEIPNIR Scientific Fog
Kimovski [117] Workflow
Wen et al., [118] 2021 Metaheuristic GREENGA Energy Cloudsim Industrial Cloud
Task
Ding et al., [119] 2018 Metaheuristic CTFS Cost, Execution Time Not Defined Scientific Fog
Workflow
Xu et al., [113] 2019 Metaheuristic IPSO Makespan, Cost Matlab Scientific Cloud-Fog
Workflow
Javaheri et al., [27] 2021 Metaheuristic DO–HHO Makespan, Offloaded, Deadline iFogsim Scientific Fog
Workflow
Singh et al., [120] 2021 Metaheuristic CSA Makespan, Cost Cloudsim Task Cloud
Gunduzalp et al., 2022 Metaheuristic AChOA Makespan cloudsim Task Cloud
[91]
Kabirzadeh et al., 2017 Hyper HH (GA, PSO, Makespan, Energy, Cost, Network Uti- iFogsim Scientific Fog
[98] Heuristic ACO, SA) lization Workflow
Rahbari [99] 2022 Hyper HHS (GA, PSO, Latency, Energy iFogsim Task Fog
Heuristic ACO, SA)
Xu et al., [121] 2019 Hybrid LBP-ACS Makespan, Energy Cloudsim Task Cloud-Fog
Heuristic
Zenyu yin et al., [24] 2022 Hybrid HMA Makespan, Energy Platform developped Industrial Cloud-Fog
Heuristic by the authors Task
Yassa et al., (b) [22] 2013 Hybrid DVFS-MODPSO Makespan, Cost, Energy Cloudsim Scientific Cloud
Heuristic Workflow
Cui et al., [122] 2017 Hybrid GA-CACO Makespan, Cost Hardware Task Cloud
Heuristic environment
Wang and Li [23] 2019 Hybrid HH (IPSO & Task Processing Time, Energy, Smart MATLAB Industrial Fog
Heuristic IACO) Manufacturing Performance Measures Task
Subramoney and 2020 Hybrid GA-PSO Makespan, Cost, FogWork flowSim Scientific Cloud-Fog
Nyirenda (a) heuristic Energy Workflow
[104]
Hassan et al., [26] 2020 Hybrid SSPSO (PSO & Makespan, Power Consumption, Task iFogsim Scientific Cloud-Fog
Heuristic SSA) Offloaded, Missed Deadline, Effects of Workflow
DDoS attacks
Abd Elaziz et al., (a) 2021 Heuristic AOAM Makespan, Eergy Matlab Scientific Fog
[5] Workflow
Abd Elaziz et al., (b) 2021 Hybrid AEOSSA Makespan Matlab Task Cloud-Fog
[123] Heuristic
Aggarwal et al., 2022 Hybrid SAFFOA Response Time, Delay Matlab Scientific Cloud
[124] Heuristic Workflow
Subramoney and 2022 Hybrid MS-PSO Makespan, Cost, Energy, Load Balancing FogWork flowsim Scientific Cloud-Fog
Nyirenda (b) Heuristic Workflow
[125]
Mokni et al., (a) [29] 2022 Hybrid MAS-GA Execution Time, Cost, Makespan WorkflowSim Scientific Cloud-Fog
Heuristic Workfow
Mokni et al., (b) [30] 2022 Hybrid Fuzzy-cone Deadline, Cost JADE Scientific Cloud-Fog
Heuristic Workfow
Narayana et al., [64] 2022 Hybrid QEESMF Round Trip Time, Computational Load per Not Defined Scientific Fog
Heuristic Makespan, Request Reception Time Workflow
(continued on next page)
11
Table I (continued )
Authors Year Categorization Algorithms Metrics Tools Types of jobs Computing
environment
Mokni et al., (c) [31] 2023 Hybrid MAS Makespan, Cost WorkflowSim Scientific Cloud-Fog
Heuristic Workfow
Sheng et al., [106] 2021 ML DRL, Task Execution Order, Task Allocation Pytorch Task Edge
REINFORCE,
FCN
Moon et al., [107] 2021 ML DQN Makespan, Cost Pytorch Job Shop Cloud-Edge
applied to a wide range of IoT use cases. The model is validated by them and taking into account latency. The second phase concerns the
simulating the solution under the CPLEX solver, choosing several smart DeadlineAware stepwise Frequency Scaling (DAFS) approach for the
city use cases. reduction of the energy consumption. The evaluation of the proposed
algorithm using synthetic and real applications shows that it achieves a
4.2. Heuristic better trade-off between energy consumption and makespan compared
to popular alternatives. This algorithm was compared to existing ap
Mora et al., [114] in 2015, proposed an imprecise computation proaches (Energy Makespan in Mobile Cloud Computing, Minimal
scheduling method for embedded system workloads, which combines Optimistic Processing Time, Predict Earliest Finish Time) and the result
imprecise computation strategies with Cloud computing paradigms, to reduces energy consumption by up to 50 % with minimal impact on
provide flexible implementation frameworks for embedded or mobile execution time. The simulation environment was developed in MATLAB
devices. This algorithm determines the most appropriate scheduling by the authors.
model between the local device and the Cloud, determines Stavrinides and Karatza [90] in 2021 have proposed an algorithm for
network-induced delay, and predicts bandwidth and delay costs asso scheduling multiple realtime IoT workflows in a heterogeneous Fog
ciated with remote communication and processing in the Cloud. The environment using partial computations with endto-end error propa
main idea of this method, which involves adjusting the number of tasks gation (two-phase partial computation scheduling technique - PC). Ac
executed according to the time available and the computing power of cording to the authors, this is the first paper in the literature to solve a
the system, is to take into account the relative priority of each task when scheduling problem using partial computations to deal with propagation
deciding on the order in which the workload is executed. The simulation errors between immediate child tasks and between all tasks in the
is a proposed scenario of a SmartDrive application for autonomous workflow, as well as their impact on the processing time of each affected
decision-making aimed at increasing user safety and convenience. This task. Workflows were generated using their own random DAG gener
application is part of the smart city concept and the construction of ator, and a synthetic workload was used in place of the available traces,
vehicle networks in which several vehicles communicate with each in order to obtain unbiased results not limited to any particular type of
other and with the city’s infrastructure to exchange information. The workload. The solution, simulated on a platform developed by the au
result demonstrates the simplicity of the proposed model and its ease of thors in C++, and, compared to a basic heuristic (which does not use
use in designing scheduling tasks. partial computation), shows that PC is resistant to the effects of input
Mao et al., [115] proposed in 2017 a Sub-Optimal Joint Task Off error propagation across workflow tasks, and can deliver a lower delay
loading Scheduling, and Transmit Power Allocation algorithm (SOJ overrun rate than the basic policy, in exchange for an average result
TOTPAA) for Mobile Edge Computing systems, with multiple accuracy of 2.43 %.
independent tasks based on flow shop scheduling and convex optimi Bentabet et al., [62] in an effort to ensure the smooth operation of the
zation. The proposed solution aims at optimizing task offloading Fog in terms of data storage, access and processing of IoT devices
scheduling and transmit power allocation in order to reduce latency and requiring high bandwidth and low latency, propose a
energy. The simulation result shows that the task offload scheduling is bandwidth-sensitive Cloud-Fog workflow scheduling algorithm that
more critical when the radio and computing resources available in the schedules tasks by priority in the different allocated resources while
MEC systems are relatively balanced. The simulation studies the impact minimizing the makespan and without increasing energy consumption
of optimal task offload scheduling on delay performance. (bandwidth-aware workflow allocation or BW-AWA). To achieve this
Rahbari and Nickray [60] propose a method for scheduling and goal, they used a list algorithm based on the Cloud/Fog workflow
allocating tasks to resources (VMs) KnapSOS (Knapsack Symbiotic Or scheduling strategy that sorts tasks by priority (computational cost,
ganisms Search), a knapsack optimization algorithm based on sched amount of encapsulated data, ranking of the predecessor task) and al
uling and symbiotic organism search. The objective of this work is the locates them to resources based on the node with optimal bandwidth.
reduction of energy consumption, cost, sensor lifetime and optimization The solution has been simulated under iFogsim and the simulation result
of network utilization. The simulation is done under iFogsim and the is compared with FCFS, HEFT, CPOS (Critical Path on a processor) and
result compared with FCFS and Knapsack show that KnapSOS is per Popularity algorithms.
forming well. Ndamlabin et al., [25] proposed in 2022 two algorithms for sched
Toniolli [61] in 2020 proposed an adaptation of the Path-Clustering uling workflows in a cloud environment with the aim of optimizing
Heuristic algorithm for scheduling multiple workflow applications in a energy consumption, processing time of submitted workflows and cost.
Cloud-Fog environment in order to search the tradeoff between make The first Structure-based Multi-objective Workflow Scheduling with an
span and price. The simulation was done under a Hardware Platform Optimal instance type (SMWSO) algorithm introduces a new approach
and the result reveal that the adaptation of PCH is better than HEFT, to determining which VM instance is appropriate for scheduling work
CCMAS and CCSH. flows. The second algorithm named Structure-Based MultiObjective
To address the challenge of resource allocation in a cloud-fog envi Workflow Scheduling with Heterogeneous Instance Types (SMWSH)
ronment due to complex and strict QoS constraints, Ijaz et al., [28] highlights the use of VMs for a scheduling problem in a heterogeneous
implemented a scheduling algorithm, Energy Makespan Multi Objective medium. The simulation result compared with the Reliability and En
Optimization (EM-MOO) that optimizes the energy consumption and ergy Efficient Workflow schedule (REEWS) algorithm shows that the
makespan of workflows. This algorithm works in two phases, the first proposed algorithms provide better energy efficiency in 80 % of work
phase concerns the modeling of the problem as a multi-objective opti flow/workload scenarios and save more than 50 % of global energy. The
mization problem by allocating tasks to resources capable of processing simulation was made under Cloudsim.
12
4.3. Meta heuristic environment with the goal of finding a tradeoff between cost and
Makespan. The PSO algorithm solves the process of mapping tasks to
4.3.1. Bio inspired based computing resources. The simulation was done in Matlab and the result
Guan et al., [116], designed an agricultural work scheduling algo shows that the proposed approach reduces the overall execution time of
rithm based on petri net and simulated annealing (SA) algorithm for the workflows compared to the classical PSO, but the execution costs of
agricultural production enterprises with the aim of reducing farmers’ both approaches are close to each other.
work time and increasing yield. The data needed for the study, from Javaheri et al., [27], address the problem of scheduling workflows in
agricultural processes and the evolution of uncertainties, are collected a multi-Fog environment, taking into account the availability of VMs
and recorded by a phone equipped with a GPS (Global Positioning and the ability of providers to process requests. To do so, they imple
System) and having a stable internet connection. The Petri net used to mented the Discrete Opposition- Harris Hawk Optimization (DO–HHO)
model discrete event systems, describes the process of agricultural work algorithm which is a discrete opposition-based version of the HHO al
(graphically and mathematically), and makes a simulation of the acti gorithm for scheduling tasks in the Fog environment; they combined
vation of agriculture on the allocation of resources (labor, machinery, with this solution the Hidden Markov Model (HMM) that predicts the
etc.); it facilitates the exploitation of agricultural land and resources. availability of VMs and the ability of providers to process requests. This
The SA is used to obtain high quality results, close to the optimal solu HHM is combined with the Viterbi algorithm and the Baum Welch al
tion for the scheduling of agricultural work. The work consists of map gorithm. The simulation is done under iFogsim and the result is
ping the farmland being exploited, planning the orders of passage on this compared with 4 algorithms cited in the literature with 5 scientific
land generated by a priority list, recording the farm data in a web server workflows. And 3 other algorithms on which they applied the HMM with
with a phone equipped with a GPS and a browser. The simulation result 5 scientific workflows. The QoS metrics are the makespan, the number of
showed that the petri net is applicable to model and simulate agricul workflows having exceeded the deadline, and, the number of workflows
tural workflows; and the SA algorithm compared with a conventional offloaded to the Cloud.
local search algorithm shows that SA has a great superiority in pro In 2021, Singh et al., [120] make a comparison of several meta
ducing a high quality solution for agricultural work planning. heuristic algorithms for scheduling heterogeneous tasks in a Cloud
Yassa et al., (a) [21] have developed an improved genetic algorithm environment. The different algorithms they conducted their studies with
GA* based on viral infection, to solve the multi-objective scheduling are Particle Swarm Optimization, Ant Colony Optimization, Genetic
problem of workflows in a Cloud environment. This solution optimizes Algorithm, Artificial Bee Colony, Crow Search Algorithm and Penguin
several QoS metrics (makespan, cost, reliability and availability) and Swarm Optimization. These algorithms have been simulated in Cloud
constraints on these metrics (deadline and budget) and also strong sim and the Crow Search algorithm is better than the others in terms of
constraints that force the execution of certain tasks on certain virtual Makespan and Cost optimization.
machines. The simulation was done under Cloudsim and data from Gunduzalp et al., [91] use for the first time an adapted version of the
Amazon Elastic Compute Cloud (EC2) and London e-Science Center Adaptive chimp optimization algorithm (AChOA) for task scheduling in
workflows were used to make a comparison with other metaheuristic the Cloud environment aiming at optimizing the makespan, as one of the
approaches. main problems affecting the Cloud performance is task scheduling. This
De Maoi and Kimovski [117] propose a multi-objective workflows algorithm checks the availability of resources before launching the
optimization (MOWO) algorithm in the fog based on the NSGA-II met scheduling process, several functions are taken into account in the
aheuristic aiming at optimizing the following QoS metrics: execution calculation of the AChOA algorithm in order to improve its efficiency
time, cost and reliability. The simulation was done under SLEIPNIR and performance. The simulation was done under Cloudsim and the
(Spark-enabled mobiLe Edge offloadIng Platform moNte-carlo sImu simulation result compared to other methods shows that AChOA per
latoR), a Monte carlo simulation framework for scheduling workflows in forms much better.
the Fog. The simulation result was evaluated in two different ways:
running the solution in the cloud Vs running in a hybrid cloud/fog 4.4. Hyper heuristic
environment; comparing MOWO with HEFT, MAXREL and MINCOST
algorithms. The hybrid cloud/fog environment is favorable for workflow Kabirzadeh et al., [98] propose a hyper heuristic for scheduling
scheduling and MOWO performs well compared to HEFT, MINCOST and workflows in Fog computing composed of GA, PSO, ACO and SA algo
MAXREL. rithms with the goal of optimizing the makespan, reducing energy
Wen et al., [118] in 2021 proposed a multi-objective industrial consumption, reducing cost and network utilization. For each workflow,
workflow scheduling algorithm (GreenGA) that aims to minimize the only one metaheuristic (the best one) is used for scheduling, SA is used
cloud service provider’s revenue, and also minimize the use of for allocating workflows to available resources. The simulation was
non-renewable energy by using the actual electricity prices with the done under iFogsim and the results compared to ACO, PSO, GA and SA
availability of green energy in different cloud datacenters. This algo algorithms show that HH is better than these metaheuristics.
rithm is an adaptation of the GA metaheuristic that performs a combi Rahbari in 2022 [99], made an analysis of heuristics and
nation of renewable energy and energy cost. The simulation was done meta-heuristics based methods, he proposed a hyper heuristics based
under cloudsim and the performance of the latter was evaluated using a scheduling method for resource allocation to fog task taking into ac
real industrial workload, the results show a better efficiency of the count latency and energy consumption. The heuristics that composed
proposed approach. HHS are GA, PSO, ACO and SA, the simulation is done under iFogsim.
The performance of HHS is evaluated with FCFS, concurrent, DP,
4.3.2. Swarm intelligence based NSGA-III, MaOPSO, PSO, ACO, GA and SA through the total execution
Ding et al., [119] present a cost-effective scheduling strategy for time, energy consumption and application loop delay. The results show
multi-workflow with time constraints (CTSF) combined with PSO and that HHS is favorable compared to these algorithms of the literature.
Min-Min algorithm to schedule a multiple workflow in a Fog computing
system. This approach aims to optimize the cost and execution time of 4.5. Hybrid heuristic
the tasks. The experimental results show that CTSF strategy can achieve
the best overall performance in terms of cost and time compared with 4.5.1. Heuristic / metaheuristic
other strategies such as FCFS-P and Max-P. In 2019, Xu et al., [121] establish a task scheduling algorithm in a
Xu et al., [113], in 2019, implemented an improved Particle Swarm Cloud-Fog environment, with the aim of solving the task scheduling
Optimization algorihm (IPSO) to schedule workflows in a Cloud-Fog problem with the priority constraints of IoT applications, taking into
13
account energy consumption and energy reduction while meeting the Hassan et al., [26] proposed a method for scheduling IoT workflows
execution time of these tasks. The proposed algorithm is a hybrid heu by taking into account the effects of DDoS attacks on the scheduling
ristic resulting from the combination of the laxity-based priority algo process. They propose a hybridization of particle swarm optimization
rithm (LBP) and the ant colony system algorithm: A laxity and ant (PSO) and salp swarm algorithm (SSA) for scheduling multi-fog work
colony system algorithm (LBP-ACS). The LBP algorithm manages the flows, SSPSO algorithm. They use two hidden Markov models on each
delay sensitivity of tasks and the ACS algorithm minimizes the total fog nodes: the first one averages the bandwidth and the second one
energy consumption. The result of the simulation performed under averages the available VMs for scheduling. SSPSO optimizes makespan
cloudsim shows that LBP-ACS reduces the energy consumption related and power consumption, reduces the number of tasks offloaded to the
to the processing of all the tasks submitted to it, ensures a reasonable cloud, reduces the number of missed deadlines and reduces the effect of
scheduling time, reduces the scheduling failure rate of the tasks asso DDoS attacks on submitted workflows. The simulation is done under
ciated with mixed deadlines compared to HEFT, GfE, DEACO iFogsim and the result compared with MOWO and Hybrid-EDF shows
algorithms. that SSPSO is better than these algorithms in terms of missed deadlines,
In [24], the authors addressed the multi-objective task scheduling offloading tasks to the cloud, and DDoS attacks on VMs.
problem of intelligent production plants in Cloud/Fog computing. They In [5], considering the problem of service continuity, QoS compli
propose an algorithm based on the priority of tasks that minimize ance in the Fog environment and the problem of network traffic and cost
makespan and energy consumption. They propose an improved algo increase caused by the large data transmission from IoT devices, the
rithm composed of ACO (IACO) combined with MBO (IMBO) (HMA or authors concerned with energy consumption, propose an intelligent IoT
Hybrid MBO-ACO). To build the priority list, they use the Merge Sort application workflow scheduling method based on the improved arith
algorithm to select the tasks with high priority and rank them in metic optimization algorithm (AOA) called AOAM to solve the IoT
descending order. The algorithm is simulated under a platform devel workflow scheduling problem in the Fog environment. This model sol
oped in C++ and the result of the simulation is compared with other ves the task scheduling problem in Fog while maximizing the QoS and
algorithms in the literature (FCFS, cloud services and IMBO). makespan measure. The proposed solution improves the search capa
bility of AOA by using the efficient and powerful search operators of
4.5.2. Metaheuristic / metaheuristic MPA (Marine Predators Algorithm) to address the diversity of used so
Yassa et al., (b) [22] proposed in 2013 an DVFS-Multi-Objective lutions and local optimal problems. The main advantage of using MPA is
Discrete Particle Swarm Optimization (DVFS-MODPSO) based on the improvement of the local search capability of AOA which avoids the
DVFS (Dynamic Voltage and Frequency Scaling) and hybrid PSO algo attraction to local points and increases the convergence speed. The
rithm. It is a multi-objective workflow scheduling algorithm in the Cloud simulation was done in MATLAB and the obtained results are compared
aimed at optimizing Makespan, cost and Energy consumption. They with other state of the art methods (AOA, MRFO [Manta Ray Foraging
simulate the proposed approach with a synthetic workflow application Optimization], MPA, CHOA [Chimp optimisation Algorithm], SSA [Salp
and two common workflow structures: parallel and hybrid structures. Swarm Algorithm] and AEOSSA) using Friedman’s test and taking into
The result was compared with a HEFT algorithm implemented by them account the NASA and HPC2N workloads; it showed that AOAM is
and prove that the proposed approach is better then the HEFT heuristic. promising and solved task scheduling efficiently compared to the other
Cui et al., [122] in 2017 worked on the research and optimization of comparative methods while minimizing makespan, energy consumption
cloud services scheduling algorithms by proposing a Task Scheduling and fitness.
System (TSS) model and implementing a hybrid GACACO (Genetic With the aim of increasing the performance and productivity of IoT
Algorithm-Chaos Ant Colony Optimization) metaheuristic which is a systems, Abd Elaziz et al., [123] in 2021, developed an alternative task
combination of Genetic Algorithm and Ant Colony Optimization algo scheduling technique for IoT applications in a cloudfog environment
rithm subdivided into two modules. In the user-centered module, the based on modified ecosystem-based artificial optimization (AEO) com
processing time of each task conforms to the general distribution, in the bined with the operators of the salp swarm algorithm: AEOSSA. The
scheduling module the metrics taken into account are the makespan and proposed approach aims at optimizing the Makespan. The performance
the cost. The simulation is done in a hardware environment and the of the latter is evaluated using different synthetic and real-world data
result shows that GA-CACO is better than GA and CAO. sets of different sizes. The simulation result performed under Matlab
Wang and Li [23] proposed a task scheduling optimization method shows that AEOSSA is better than SSA, PSO, HHO, FA and AEO in terms
based on a hybrid heuristic (Improved PSO and Improved ACO) of fog of Makespan and troughput.
computing in a smart factory (concerning the production line) while To address the challenge of scheduling workflows in a cloud envi
minimizing task processing time, energy consumption, and improving ronment by taking user priority into account, Aggarwal et al., [124]
smart manufacturing performance metrics, such as production effi implemented a self-adaptive fruit fly optimization algorithm (SAFFOA)
ciency, product quality, and equipment utilization rate. This algorithm with advanced encryption algorithm (AES), which reduces the response
solves the problem of end devices with limited computational resources time and execution delay of the user’s request. The proposed approach
and high power consumption, and makes the scheme feasible for has been implemented and simulated in Matlab and the simulation
real-time and efficient end device processing tasks. The authors use a result compared with FCFS, PSO, ABC, GA, FFOA, DE and ETC algo
4-layer architecture (application, cloud, fog, end device layers) for rithms shows its effectiveness.
modeling the problem. The simulation is done under MATLAB, the re According to Subramoney and Nyirenda (b) [125], the major chal
sults are evaluated with IPSO, IACO and RR, the evaluation metrics are lenge in distributed computing environments is generating optimal so
processing time (execution time), power consumption and reliability. lutions by reducing execution time, cost and energy consumption, as
The solution proposed by the authors is better than the algorithms of the well as ensuring proper load balancing. They have implemented a Multi
literature. SwarmParticle Swarm Optimization algorithm to improve the sched
Subramoney and Nyirenda (a) [104] performed a comparative uling of scientific workflows in CloudFog environments by optimizing
evaluation of four populationbased optimization algorithms (PSO, GA, execution time, cost, power consumption and load balancing. The Fog
DE and GA-PSO) for scheduling workflows in the Fog. This work pro WorkflowSim simulator is used in the process of evaluating objectives
vides the foundation for the weighted sum objective function for the that serve as performance measures. The MS-PSO approach compared to
workflow scheduling problem and develops this function based on three PSO, GA, DE, and GA-PSO algorithms shows that it is more stable and
objectives: makespan, cost, and energy. The results of simulation done reliable. The application workflows used are Montage, Cybershake,
with FogWorkflowSim show that the hybrid combination of the GA-PSO Epigenomics, LIGO and SIPHT.
algorithm is slightly better than the standard algorithms.
14
4.5.3. Multi agent system (MAS)/ (Meta)heuristic Moon et al., [107] conducted a study based on multi-access edge
In 2022, Mokni et al., (a) [29] proposed an approach based on the computing (MEC) with the goal of cooperating manufacturing processes.
cooperation of multi-agent systems with the GA algorithm to schedule They applied Deep Q-Network (DQN) which is a learning model in the
IoT workflows in a hybrid Cloud-Fog environment. This approach aims shop floor scheduling process. They conducted scheduling studies in the
to optimize task execution time, cost and makespan. The MAS approach edge computing ecosystem of their manufacturing processes without the
minimizes execution time and cost, and, the GA approach based on help of cloud computing data centers because the latter present sensitive
deadline and budget constraints optimizes makespan, cost, reliability issues for the manufacturing process (security issues and the commu
and availability. The Workflowsim simulator was used to evaluate the nication delay). Therefore, they proposed a method to perform inde
performance of the solution with Five Benchmarks (epigenomics, pendent scheduling at the network edge through cooperative scheduling
inspiral, montage, cybershake, sipht), the result show that MAS-GA among edge devices in a multiple access edge computing framework.
applied to a hybrid Cloud-Fog environment performs better than The proposed framework was evaluated, analyzed, and compared to
MAS-GA applied to a Cloud or Fog environment. The metrics for eval existing frameworks in terms of solutions and services. It was compared
uating the solution are response time, execution time and makespan. In with traditional DQN, SPT and LPT on one hand, and Cloud, Cloud-Edge
order to solve the problem of scheduling workflows with conflicting on the other.
constraints in a hybrid Fog-Cloud environment, Mokni et al., (b) [30]
developped the fuzzy-cone algorithm which is a multi-agent system- 5. Analytical discussion
based workflow scheduling algorithm with negotiation offers taken into
account with fuzzy constraints. The work consists in optimizing the In this section, we present our analysis of the different previous
deadline and the budget while satisfying the constraints of the users and works mentioned above. The analysis is based on the classification of the
the service providers with two types of agents: contract agents for the algorithms, the simulation environment, the computing environment,
execution of the users’ workflows and provider agents in charge of the metrics used, and the year of publication of these works.
maximizing the profit of the fog-cloud providers. These agents collab Fig. 6 exhibits the percentage of work focused on scheduling tech
orate with each other and negotiate for the placement of workflows in niques by different researchers. It shows that 60 % is addressed on
the Cloud or fog environments, the QoS metrics to be optimized and the Scientific workflow scheduling while 27 % focus on resolving task
SLA compliance. The solution has been simulated under JADE, and the scheduling and 7 % resolve Industrial Task scheduling, 3 % focus on Job
result, compared to GA, PSO and MAS-GA algorithms show that shop and Flow shop scheduling.
Fuzzy-Cone is better than these algorithms in the literature. Fig. 7 indicates the works which have considered each objective. It
Narayana et al., [64] proposed an efficient scheduling model that indicates that the makespan is the factor which have been focused by
optimizes the quality of service and energy in the scheduling process of several scheduling schemes in 21 %. Among the other metrics, cost has
fog and edge nodes. It is a supervised learning approach named ”Qual been considered by 18 %, energy consumption by 17 % Deadline by 6 %,
ity-aware Energy Efficient Scheduling Model for Fog computing execution time by 4 %, latency by 4 %, reliability by 2 %, and the other
comprised IoT networks”. The statistical measures of moving averages metrics by 28 %.
and Heiken-Ashi models were used as the basis for the proposed Fig. 8 expose the category of algorithms used in scheduling problem.
scheduling model. To achieve optimal scheduling, parameters such as It indicates that Hybrid heuristic is the most used category with a rate of
round-trip time, computational load per makespan, and request recep 35 %, followed by metaheuristic with 23 %, Heuristic 22 %, MP 10 %,
tion time were taken into account. The statistics of the quality measures ML and hyper heuristic 5 %.
obtained from the experimental study indicate that the proposed model Fig. 9 expose the computing environment used by researchers to
significantly reduces these parameters compared to EBSA (Energy solve scheduling problem. Hybrid Cloud-fog is considered in 37 % of
Balancing Scheduling Algorithm), JODPS (Joint optimization of data studies, followed by Fog with a rate of 28 %, Cloud with 22 %, Edge with
placement and scheduling) and HMCOT Heuristic min-conflicts opti 5 %, Cloud-edge with 5 %, and Web server with 3 %.
mizing technique. Fig. 10, shows simulation tools used by researcher to solve sched
In 2023, Mokni et al., (c) [31] develop an approach that merges uling problem. Cloudsim is used in 18 %, Matlab 15 %, ifogsim 15 %,
partitioning, sequencing and scheduling of workflows with the objective Not defined 10 %, Pytorch 5 %, FogWorkflowSim 5 %, WorkflowSim 5
of optimizing the makespan and the cost while taking into account the %, CPLEX 5 %, Hardware environment 5 %, Platform developped by
conflicting constraints of users and suppliers. This algorithm is based on authors 5 %, SLEIPNIR 3 %, Phone 3 %, Jade 3 %, Pycharm 3 %.
multiagent systems and a fuzzy inference system. The solution imple Fig. 11 shows the different years of publication of related works cited
mented under workflowsim and the result compared with EM-MOO and in the literature. We note that the range of these years is between 2007
MAS-GA shows that the proposed approach is more efficient than these and 2023. 2 % of these works are done in 2007, 3 % in 2015 and 2023, 5
algorithms. % in 2013 and 2018, 8 % in 2019, 10 % in 2017, 13 % in 2020, 23 % in
2022 and 28 % in 2021.
4.6. Machine learning
In this paper [106], the problem of scheduling computationally

intensive tasks at Egde Computing is addressed. Multiple tasks are
scheduled on VMs to maximize the long-term task satisfaction degree
(LTSD). The problem is formulated as a Markov decision process. To
solve this task scheduling problem, the authors use Deep Reinforce
Learning (DRL) to schedule time (execution order) and allocate re
sources to tasks (assignment of tasks to specific VMs); the REINFORCE
algorithm for task scheduling and the Fully Connected Neural Network
(FCN) to extract task features. Task execution order and task allocation
to resources are jointly optimized. The simulation results were obtained
using Python3 run with Pytorch. DRL is implemented to solve the MDP
(Markov Decision Process) problem and is compared to greedy-FCFS and
greedy-SJF with respect to the average degree of satisfaction and the
success rate of tasks. DRL is better than these two algorithms. Fig. 6. Type of jobs.
15
Fig. 7. Metrics used.
Fig. 10. Simulation tools.
Fig. 8. Categorization.
Fig. 11. Publications per year.
focused by several scheduling schemes in 25 %.
• RQ3: What is the most common type of algorithm categorization

used to solve a scheduling problem?
According to Fig. 8, the most common category of algorithm used to

solve scheduling problems in Cloud,Fog and Edge computing environ
Fig. 9. Computing environments. ments is the Hybrid heuristics algorithm.
6. Research queries and inferences • RQ4: Is the collaboration of Cloud, Fog, and Edge computing envi
ronments suitable in the context of tasks/workflows scheduling?
In this section, we provide the following answers to the research
questions raised in Section 1.1. As shown as in Fig. 9, we can see that there is work that has been
done in hybrid Cloud-Fog and Cloud-Edge environments.
• RQ1: What types of tasks/workflows are most scheduled in the
Cloud, Fog, and Edge computing environment? • RQ5: What is the most requested computing environment for tasks/
workflows scheduling?
The different types of tasks scheduled in the Cloud, Fog, and Edge
computing environ ment are scientific, industrial, job shop, and flow As shown in Fig. 9, the most computing environment is the hybrid
shop. According to Fig. 6, the type of task/workflow that are most Cloud-fog with a rate of 35 %.
schedule in the Cloud, Fog, and Edge computing environment is scien
tific workflow. • RQ6: What are the most widely used simulation tools for evaluating
tasks/workflows scheduling algorithms in Cloud, fog, and edge
• RQ2: What are the most used QoS metrics in scheduling problems in computing environments?
Cloud, Fog, and Edge computing environments?
According to Fig. 10, Cloudsim is the most used simulation tools.
As shown as in Fig. 7, the makespan is the factor which have been
16
• RQ7: What kind of data are used in the evaluation of algorithms in the cloud. If the communication is between the Fog and the cloud
scheduling problems? Considering the above-mentioned works, manager, the workflow or sub-workflow received from the Fog is for
some works used real-world workflow trace to evaluate their algo warded to one of the datacenters for processing, with the awareness that
rithm, namely: HPC2N and NASA in the fog [5], Ligo, Cybershake, the requirements given by the Fog have been met. When the commu
Epigenomics and Montage data sets in the cloud [118], Ligo, nication is direct between the Fog and a cloud datacenter, the choice of
Montage, Epigenomics, Sipht, Cybershake in the fog [27], HPC2N, datacenter is based on the reputation of the datacenter in relation to the
NASA and Synthetic workload in the cloud-fog [123]. satisfaction it provides to the Fog. It would be interesting to see whether
• RQ8: What are the challenges of scheduling tasks/workflows in the the consideration of reputation when choosing a datacenter would help
Cloud, Fog, and Edge computing environment? the Fog to execute the workflow in a way that is aware of the re
quirements of both the customer and the providers of the Fog and cloud
The challenges of scheduling tasks/workflows in distributed resources.
computing environments include, among others, mapping tasks/VMs to
client requirements and supplier resource availability, and optimizing 7.6. Security
service quality metrics without disturbing each other.
There are well-studied security issues for cloud computing that
7. Challenges and open issues require significant security measures to protect it [78]. However, these
measures are not suitable for fog computing due to its characteristics,
In this section, we present open issues and challenges of workflow including mobility, heterogeneity, and large-scale geodistribution [126,
scheduling in a Cloud-Fog computing environment. 127,78]. Currently, only a few researchers focus on security issues in fog
computing, and there are some open issues like external attacks, and
7.1. Optimization metrics intrusion detection [19].
Most of the algorithms mentioned are multi-objective algorithms 7.7. End user privacy
that optimize makespan, cost, and energy. These metrics are aimed at
satisfying both the customer and the provider. However, to improve the Preserving end-user privacy is a significant issue facing fog
quality of service and reassure the customer, it would be interesting to computing, as being closer to endusers, they can collect more sensitive
consider metrics such as reliability, throughput, availability of resources data [126,78]. It is very difficult to maintain centralized control of fog
in the Fog-Cloud. nodes because they are spread across multiple zones [78]. Unsecured fog
nodes are points of failure through which a hacker can break into the
7.2. Type of jobs network and collect user data [128,129,78]. Protecting the privacy of
fog user is a challenging issue that requires more research [78].
Based on the analytical discussion, we conclude that several works
have been done in scientific workflows. With the emergence of smart 8. Conclusion
factories and IoT, research works based on workfloworiented industry in
cloud, fog and edge environments must be conducted. The internet of things has attracted a lot of attention recently. The
industrial internet of things is booming. It has not gained popularity like
7.3. Benchmark data the IoT. IoT devices are dynamic in nature and have limited storage and
processing capabilities. The explosion of the IoT paradigm has promoted
Based on the answer of the 7th research question, most authors the use of fog computing and edge computing as the cloud environment
working on workflows evaluate their approaches using simulated data, has shown limitations such as high latency and network failure. In this
as the data produced by real-world applications sometimes remain paid paper, we studied the concepts of Industry 4.0, its challenges and
and not easily accessible for confidentiality, privacy, security and other enabling technologies; we reviewed the notion of workflows scheduling.
issues. There is a real need to design benchmark data in order to easily We surveyed various existing scheduling algorithms and we have clas
evaluate and compare algorithms. sified them in a table according to the year of publication, the optimized
metrics, the type of jobs, the used computational environments, the used
7.4. Collaboration between fog and cloud simulators and the categorization. Our analysis indicates that few re
searchers focused on scheduling industrial task on computing environ
A new paradigm that combines cloud and fog computing seeks to ment system. Finally, some open issues and challenges worthy of further
give users quick and effective services. Fog computing is an extension of study and research, including: security, optimization metrics, collabo
the cloud to the periphery of the network, that enables the deployment ration between fog and cloud, type of jobs, reputation, end-user privacy,
of applications closer to the end-users, it must always interact with and, benchmark data were presented.
Cloud servers. By moving apps closer to the network’s edge, this strategy
lowers their latency and bandwidth needs. Task offloading, intrusion Declaration of Competing Interest
detection, and virtual simulation are just a few examples of how cloud
and fog computing can work together. In a Fog, only certain localized The authors declare that they have no known competing financial
applications are provisioned and synchronized with the cloud [78]. interests or personal relationships that could have appeared to influence
the work reported in this paper.
7.5. Reputation
Data availability
In a hybrid Fog-Cloud execution environment, a workflow submitted
to the Fog can be fully executed in the Fog if it feels capable of delivering No data was used for the research described in the article.
the results while guaranteeing the QoS and the gain of the Fog providers.
In case of inability, the Fog can collaborate with the Cloud to pass all or
part of the workflow to it for execution. If the cloud consists of several
data centres, the communication between the Fog and the cloud can
either take place with the cloud manager or directly with a datacenter in
17
References [25] Jean Etienne Ndamlabin Mboula, Vivient Corneille Kamla, Muhammad
Hafizuddin Hilman, and Clement Tayou Djamegni. Energy-efficient workflow
scheduling based on workflow structures under deadline and budget constraints
[1] Mostafa Ghobaei-Arani, Alireza Souri, Ali A Rahmanian, Resource management
in the cloud. arvix,Distributed, Parallel, and Clust. Comput.ing, January 2022. htt
approaches in fog computing: a comprehensive review, J. Grid Comput. 18
ps://doi.org/10.48550/arXiv.2201.05429.
(2019), https://doi.org/10.1007/s10723-019-09491-1.
[26] Omed Hassan Ahmed, Joan Lu, Aram Mahmood Ahmed, Amir Masoud Rahmani,
[2] Laith Abualigah, Ali Diabat, Mohamed Abd Elaziz, Intelligent workflow
Mehdi Hosseinzadeh, Mohammad Masdari, Scheduling of scientific workflows in
scheduling for big data applications in iot cloud computing environments, Clust.
multi-fog environments using markov models and a hybrid salp swarm algorithm,
Comput. 24 (2021) 1–20, https://doi.org/10.1007/s10586-021-03291-7.
IEEE Access 8 (2020), https://doi.org/10.1109/ACCESS.2020.3031472.
[3] Laith Abualigah, Ali Diabat, P. Sumari, A.H. Gandomi, Applications,
[27] Danial Javaheri, Saeid Gorgin, Jeong-A. Lee, Mohammad Masdari, An improved
deployments, and integration of internet of drones (iod): a review, IEEE Sens. J.
discrete harris hawk optimization algorithm for efficient workflow scheduling in
21 (2021), https://doi.org/10.1109/JSEN.2021.3114266.
multi-fog computing, Sustain. Comput. 36 (2021) 40, https://doi.org/10.1016/j.
[4] Emiliano Sisinni, Abusayeed Saifullah, Song Han, Ulf Jennehag, Mikael Gidlund,
suscom.2022.100787. December.
Industrial internet of things: challenges, opportunities, and directions, IEEE
[28] Samia Ijaz, Ehsan Ullah Munir, Saima Gulzar Ahmad, M. Mustafa Rafique, Omer
Trans. Ind. Inf. 14 (2018), https://doi.org/10.1109/TII.2018.2852491.
F. Rana, Energy-makespan optimization of workflow scheduling in fog-cloud
[5] Mohamed Abd Elaziz, Laith Abualigah, Rehab Ali Ibrahim, Ibrahim Attiya, Iot
computing, Computing 103 (2021), https://doi.org/10.1007/s00607-021-00930-
workflow scheduling using intelligent arithmetic optimization algorithm in fog
0.
computing, Hindawi Comput. Intell. Neurosci. (2021) 14, https://doi.org/
[29] Marwa Mokni, Sonia Yassa, Cooperative agents-based approach for workfow
10.1155/2021/9114113, page.
scheduling on fog-cloud computing, J. Ambient Intell. Humaniz. Comput. 13
[6] Almarshadi Mohammed Hueesin, Saleh Mahmoud Ismail, Effects of precision
(2022), https://doi.org/10.1007/s12652-021-03187-9.
irrigation on productivity and water use efficiency of alfalfa under different
[30] Marwa Mokni, Sonia Yassa, Jalel Eddine Hajlaoui, Mohamed Nazih Omri,
irrigation methods in arid climates, J. Appl. Sci. Res. 7 (2011). https://ucanr.
Rachid Chelouah, Multi-agent system-based fuzzy constraints offer negotiation of
edu/sites/adi/files/197037.pdf.
workflow scheduling in fog-cloud environment, Computing (2022), https://doi.
[7] Hak-Jin Kim, Kenneth A. Sudduth, John W. Hummel, Soil macronutrient sensing
org/10.1007/s00607-022-01148-4.
for precision agriculture, J. Environ. Monit. 11 (2009), https://doi.org/10.1039/
[31] Marwa Mokni, Sonia Yassa, Jalel Eddine Hajlaoui, Mohamed Nazih Omri,
b906634a.
Rachid Chelouah, Multi-objective fuzzy approach to scheduling and offloading
[8] Nathaniel D Mueller, James S Gerber, Matt Johnston, Deepak K Ray,
workflow tasks in fog–cloud computing, Simul. Model. Pract. Theory 123 (2023),
Navin Ramankutty, Jonathan A Foley, Closing yield gaps through nutrient and
https://doi.org/10.1016/j.simpat.2022.102687.
water management, Nature (2012) 490, https://doi.org/10.1038/nature11420,
[32] Zouina Sarfraz, Azza Sarfraz, Hamza Mohammad Iftikar, Ramsha Akhund, Is
ctober.
covid-19 pushing us to fifth industrial revlution (society 5.0)? Pak. J. Med. Sci. 37
[9] Imen BOUZARKOUNA, PhD thesis, INSA Rouen Normandie, 2021.
(2) (2021) https://doi.org/10.12669/pjms.37.2.3387.
[10] Laith Abualigah, Ali Diabat, A novel hybrid antlion optimization algorithm for
[33] Tannis Liviniuk. Industry 5.0 in engineering and construction, June 2022. https
multi-objective task scheduling problems in cloud computing environments,
://www.trilliumgroup.io/post/industry-5-0-in-engineering-and-construction,
Clust. Comput. 24 (2020) 205–223, https://doi.org/10.1007/s10586-020-03075-
[Accessed on 31st October 2022].
5.
[34] Amr Adel, Future of industry 5.0 in society: human-centric solutions, challenges
[11] Mohamed Abd Elaziz, Laith Abualigah, Ibrahim Attiya, Advanced optimization
and prospective research areas, J. Cloud Comput. 11 (40) (2022), https://doi.
technique for scheduling iot tasks in cloud-fog computing environments, Future
org/10.1186/s13677-022-00314-5.
Gener. Comput. Syst. 124 (2021) 142–154, https://doi.org/10.1016/j.
[35] Justyna Matuszak. Is your business ready for industry 5.0 ?, March 2022. https
future.2021.05.026.
://knowhow.distrelec.com/manufacturing/is-your-business-ready-for-indus
[12] Amit Kishor, Chinmay Chakraborty, Wilson Jeberson, Intelligent healthcare data
try-5-0/, [Accessed on 31st October 2022].
segregation using fog computing with internet of things and machine learning,
[36] Short history of manufacturing: from industry 1.0 to industry 4.0, January 2023.
Int. J. Eng. Syst. Model. Simul. 12 (2/3) (2021) 188–194, https://doi.org/
https://kfactory.eu/short-history-of-manufacturing-from-industry-1-0-to-indu
10.1504/IJESMS.2021.10036745.
stry-4-0/, [Accessed on 12th January 2023].
[13] Michaela Iorga, Larry Feldman, Robert Barton, Michael J. Martin, Nedim Goren,
[37] Industry 4.0, January 2023. https://www.gtai.de/en/invest/industries/industr
and Charif Mahmoudi. Fog computing conceptual model recommendations of the
ial-production/industrie-4-0, [Accessed on 12th January 2023].
national institute of standards and technology. NIST Special Publication 500-325,
[38] Hatem Azzam. Industry 4.0 - the fourth industrial revolution, July 2019. https
2018. https://doi.org/10.6028/NIST.SP.500-325.
://www.linkedin.com/pulse/industry-40-fourth-industrial-revolution-hat
[14] Flavio Bonomi, Rodolfo Milito, Jiang Zhu, Sateesh Addepalli, Fog computing and
em-azzam/, [Accessed on 12th January 2023].
its role in the internet of things, in: Proceeding of the first edition of the MCC
[39] Xun Xu, Yuqian Lu, Birgit Vogel-Heuser, Lihui Wang, Industry 4.0 and industry
workshop on Mobile cloud computing, ACM, 2012. https://dl.acm.org/doi
5.0—Inception, conception and perception, J. Manuf. Syst. 61 (2021), https://
/epdf/10.1145/2342509.2342513.
doi.org/10.1016/j.jmsy.2021.10.006.
[15] Anass Ben Moula. Les differences entre cloud computing, edge computing et fog
[40] Ron Davies. Industry 4.0 digitalisation for productivity and growth, September
computing, Mars 2022. https://fr.linkedin.com/pulse/les-diff %C3 %A9rences-
2015. https://www.europarl.europa.eu/RegData/etudes/BRIE/2015/568
entre-cloud-computing-edge-et-fog-anass-ben-moula?, [Online,Accessed on 29
337/EPRSBRI(2015)568337EN.pdf, [Accessed on 12th January 2023].
July 2022].
[41] Yuqian Lu, Xun Xu, Lihui Wang, Smart manufacturing process and system
[16] Ahmed M Alwakeel, An overview of fog computing and edge computing
automation—a critical review of the standards and envisioned scenarios,
securityand privacy issues, Sensors 21 (2021), https://doi.org/10.3390/
J. Manuf. Syst. 56 (2020), https://doi.org/10.1016/j.jmsy.2020.06.010. July.
s21248226.
[42] Shiyong Wang, Jiafu Wan, Daqiang Zhang, Li Di, Chunhua Zhang, Towards smart
[17] Edge computing: definition’ et cas d’usage de la technologie, January 2021.
factory for industry 4.0: a self-organized multi-agent system with big data based
https://www.journaldunet.fr/web-tech/dictionnaire-de-L-iot/, [Accessed on 18
feedback and coordination, Comput. Netw. 101 (2016), https://doi.org/10.1016/
July 2022].
j.comnet.2015.12.017. June.
[18] Ali Shakarami, Hamid Shakarami, Mostafa Ghobaei-Arani, Elaheh
[43] Ray Y. Zhong, Xun Xu, Eberhard Klotz, Stephen T. Newman, Intelligent
Nikougoftarand Mohammad FarajiMehmandar, Resource provisioning in edge/
manufacturing in the context of industry 4.0: a review, Engeneering 3 (2017),
fog computing: a comprehensive and systematic review, J. Syst. Archit. 122
https://doi.org/10.1016/J.ENG.2017.05.015. October.
(2022), https://doi.org/10.1016/j.sysarc.2021.102362. January.
[44] Rabeea Basir, Saad Qaisar, Mudassar Ali, Monther Aldwairi, Muhammad
[19] Bushra Jamil, Humaira Ijaz, Mohammad Shojafar, Kashif Munir,
Ikram Ashraf, Aamir Mahmood, Mikael Gidlund, Fog computing enabling
Rajkumar Buyya, Resource allocation and task scheduling in fog computing and
industrial internet of things: state-of-the-art and research challenges, Sensors
internet of everything environments: a taxonomy, review, and future directions,
(2019), https://doi.org/10.3390/s19214807.
ACM Comput. Surv. 10 (01) (2022), https://doi.org/10.1145/3513002. January.
[45] Roberto Figlie, Riccardo Amadio, Marios Tyrovolas, Chrysostomos Stylios,
[20] Cheol-Ho Hong, Blesson Varghese, Resource management in fog/edge computing:
Łukasz Pasko, Dorota Stadnicka, Anna Carreras-Coch, Agust´ın Zaballos,
a survey on architectures, infrastructure, and algorithms, ACM Comput. Surv.
Joan Navarro, Daniele Mazzei, Towards a taxonomy of industrial challenges and
(CSUR) 52 (2019), https://doi.org/10.1145/3326066.
enabling technologies in industry 4.0, arXiv Comput. Soc. (2022), https://doi.
[21] Sonia Yassa, Jerémie Sublime, Rachid Chelouah, Hubert Kadima, Geun-Sik Jo,
org/10.48550/arXiv.2211.16563. November.
Bertrand Granado, A genetic algorithm for multi-objective optimisation in
[46] Aurelie Edouard, Samir Lamouri, Virginie Fortineau, yves Sallez,
workflow scheduling with hard constraints, Int. J. Metaheuristics 2 (2013),
Alexandre Berger, Utilisation des approches de l’industrie 4.0 et de l’internet
https://doi.org/10.1504/IJMHEUR.2013.058475. December.
physique pour le deploiement d’entrepots urbains, in: 13eme Conference
[22] Sonia Yassa, Rachid Chelouah, Hubert Kadima, Bertrand Granado, Multi-
Internationale De Modelisation, Optimisation Et Simulation (MOSIM2020), 2021.
objective approach for energyaware workflow scheduling in cloud computing
April, https://hal.archives-ouvertes.fr/hal-03193070.
environments, Sci. World J. (2013) 2013, https://doi.org/10.1155/2013/
[47] Emmanuel Ohene Opare, Ethan Struhs, Amin Mirkouei, A comparative state-of-
350934.
technology review and future directions for rare earth element separation,
[23] Juan Wang and Di Li. Task scheduling based on a hybrid heuristic algorithm for
Renew. Sustain. Energy Rev. 143 (110917) (2021), https://doi.org/10.1016/J.
smart production line with fog computing. Sensors, 19(1023), 2019. https://doi.
RSER.2021.110917.
org/10.3390/s19051023.
[48] Yushan Siriwardhana, Pawani Porambage, Madhusanka Liyanage,
[24] Zhenyu Yin, Fulong Xu, Yue Li, Chao Fan, Feiqing Zhang, Guangjie Han,
Mika Ylianttila, A survey on mobile augmented reality with 5 g mobile edge
Yuanguo, A multi-objective task scheduling strategy for intelligent production
computing: architectures, applications, and technical aspects, IEEE Commun.
line based on cloud-fog computing, Sensors (2022), https://doi.org/10.3390/
Surv. Tutor. 23 (2) (2021), https://doi.org/10.1109/COMST.2021.3061981.
s22041555.
18
[49] Thomas B Sheridan, Human–robot interaction: status and challenges, Hum. [77] Mohammed Alodib, Qos-aware approach to monitor violations of slas in the iot,
Factors 58 (2016), https://doi.org/10.1177/0018720816644364. J. Innov. Digit. Ecosyst. 3 (2016), https://doi.org/10.1016/j.jides.2016.10.010.
[50] Mohd Javaid, Abid Haleem, Ravi Pratap Singh, Rajiv Suman, Artificial [78] Hany F. Atlam, Robert Walters, Gary Wills, Fog computing and the internet of
intelligence applications for industry 4.0: a literature-based study, J. Ind. Integr. things: a review, Big Data Cogn. Comput. 2 (2018), https://doi.org/10.3390/
Manag. 7 (2022), https://doi.org/10.1142/S2424862221300040. bdcc2020010.
[51] T. Stock, G. Seliger, Opportunities of sustainable manufacturing in industry 4.0, [79] Mohammed Al-Khafajiy, Thar Baker, Hilal Al-Libawy, Atif Waraich,
in: Global Conference on Sustainable Manufacturing - Decoupling Growth from Carl Chalmers, Omar Alfandi, Fog computing framework for internet of things
Resource Use, Elsevier, 2016, https://doi.org/10.1016/j.procir.2016.01.129. applications, in: 2018 11th International Conference on Developments in
[52] Vasja Roblek, Maja Mesko, Alojz Krapez, A complex view of industry 4.0, Sage eSystems Engineering (DeSE), 2018, https://doi.org/10.1109/DeSE.2018.00017.
Open J. 6 (2016), https://doi.org/10.1177/2158244016653987. [80] Muhammad Waqas, Yong Niu, Manzoor Ahmed, Yong Li, Depeng Jin, Zhu Han,
[53] Robot revolution and industrial iot initiative —English, September 2015. http Mobility-aware fog computing in dynamic environments: understandings and
s://www.jmfrri.gr.jp/english/, [Accessed 02 May 2023]. implementation, IEEE Access 7 (2018), https://doi.org/10.1109/
[54] Le gouvernement accel ere` son action pour l’industrie du futur, May 2021. ACCESS.2018.2883662.
https://www.lafrenchfab.fr/2021/05/12/industrie-du-futur-gouvernement- [81] Shreya Ghosh, Anwesha Mukherjee, Soumya Kanti Ghosh, Rajkumar Buyya,
accelere/, [Accessed 02 May 2023]. Mobi-iost: mobility-aware cloudfog-edge-iot collaborative framework for time-
[55] Smartfactorykl. https://smartfactory.de/en/, [Accessed 02 May 2023]. critical applications, IEEE Trans. Netw. Sci. Eng. 7 (2019), https://doi.org/
[56] Siuly Siuly, Varun Bajaj, Abdulkadir Sengur, Yanchun Zhang, An advanced 10.1109/TNSE.2019.2941754.
analysis system for identifying alcoholic brain state through eeg signals, Int. J. [82] Mark F. Tompkins, Master’s thesis, Massachusetts Institute of Technology, 2003.
Autom. Comput. 16 (2019), https://doi.org/10.1007/s11633-019-1178-7. [83] Altex soft. Schedule optimization: technical approaches and use cases, July 2021.
[57] Stephanini group. The fourth industrial revolution: industry 4.0 challenges and https://www.altexsoft.com/blog/schedule-optimization/, [Accessed on 29 May
opportunities for your business, January 2021. https://stefanini.com/en/insigh 2023].
ts/news/the-fourth-industrial-revolution-industry-4-0-challenges-and-opp, [84] Somayeh Mohammadi, Hossein Pedram, Latif PourKarimi, Integer linear
[Accessed 04 May 2023]. programming-based cost optimization for scheduling scientific workflows in
[58] David Hollingsworth. The workflow reference model. Workflow Management multi-cloud environments, J. Supercomput. (2018), https://doi.org/10.1007/
Coalition Specification, TC00-1003, January 1995. http://www.wfmc.org/Dow s11227-018-2465-8. June.
nload-document/TC00-1003-The-Workflow-Reference-Model.html. [85] Zahra Movahedi, Bruno Defude, Amir mohammad Hosseininia, An efficient
[59] Bouafia Zouheyr, Badr Benmammar, Mourad Hakem, Algorithmes population-based multiobjective task scheduling approach in fog computing
d’ordonnancement des tâches dans un environnement cloud, Mediter. systems, J. Cloud Comput. 10 (2021), https://doi.org/10.1186/s13677-021-
Telecommun. J. 5 (June 2015). https://www.researchgate.net/publication/2 00264-4. October.
90430460_ Algorithmes d’ordonnancement des tâches dans un environnement [86] Ahwazi. Amin Ziagham, Master’s thesis, UIT the Arctic University of Norway,
Cloud. [Accessed 27 December 2023]. 2022.
[60] Dadmehr Rahbari, Mohsen Nickray, Scheduling of fog networks with optimized [87] Ismael Martinez, Abdallah Jarray, Abdelhakim Senhaji Hafid, Scalable design and
knapsack by symbiotic organisms search, in: 2017 21st Conference of Open dimensioning of fogcomputing infrastructure to support latency-sensitive iot
Innovations Association (FRUCT), 2017, pp. 278–283, https://doi.org/10.23919/ applications, IEEE Internet Things J. 7 (2020), https://doi.org/10.1109/
FRUCT.2017.8250193, pages. JIOT.2020.2979705. June.
[61] Jean Lucas de Souza Toniolli, PhD thesis, CONCORDIA UNIVERSITY, 2020. [88] Jose Santos, Tim Wauters, Bruno Volckaert, Filip De Turck, Towards end-to-end
[62] Dougani Bentabet, Dennai Abdeslem, Makespan optimization of workflow resource provisioning in fog computing over low power wide area networks,
application based on bandwidth allocation algorithm in fog-cloud environment, J. Netw. Comput. Appl. 175 (2021), https://doi.org/10.1016/j.
Res. Sq. (2022), https://doi.org/10.21203/rs.3.rs-1809172/v1. July. jnca.2020.102915. November.
[63] Sonia Yassa, These de doctorat, Universit` e de Cergy Pontoise, 2014. [89] Nasim Soltani, Behzad Soleimani, Behrang Barekatain, Heuristic algorithms for
[64] Narayana Potu, Sreedhar Bhukya, Chandrashekar Jatoth, task scheduling in cloud computing: a survey, Int. J. Comput. Netw. Inf. Secur. 9
Premchand Parvataneni, Quality-aware energy efficient scheduling model for fog (8) (2017), https://doi.org/10.5815/ijcnis.2017.08.03.
computing comprised iot network, Comput. Electr. Eng. 97 (2022), https://doi. [90] Georgios L. Stavrinides, Helen D. Karatza, Orchestrating real-time iot workflows
org/10.1016/j.compeleceng.2021.107603. in a fog computing environment utilizing partial computations with end-to-end
[65] Mohammad Masdari, Sima ValiKardan, Zahra Shahi, Sonay Imani Azar, Towards error propagation, Clust. Comput. 24 (2021), https://doi.org/10.1007/s10586-
workflow scheduling in cloud computing: a comprehensive analysis, J. Netw. 021-03327-y. July.
Comput. Appl. 66 (2016), https://doi.org/10.1016/j.jnca.2016.01.018. [91] Emrullah Gunduzalp, Gungor Yildirim, Yetkin Tatar, Efficient task scheduling in
[66] Bertram Ludäscher, Shawn Bowers, Timothy McPhillips, Scientfic workflows. cloud systems with adaptive discrete chimp algorithm, Balkan J. Electr. Comput.
Encyclopedia of Database Systems, 2009, https://doi.org/10.1007/978-0-387- Eng. 10 (03) (2022), https://doi.org/10.17694/bajece.989467. July.
39940-9-1471. [92] Essam H Houssein, Ahmed G Gad, Yaser M Wazery, Ponnuthurai
[67] Toktam Ghafariana, Bahman Javadib, Rajkumar Buyya, Decentralised workflow Nagaratnam Suganthan, Task scheduling in cloud computing based on meta-
scheduling in volunteer computing systems, Int. J. Parallel Emerg. Distrib. Syst. heuristics: review, taxonomy, open challenges, and future trends, Swarm Evol.
(2014), https://doi.org/10.1080/17445760.2014.973876. October. Comput. 62 (2021), https://doi.org/10.1016/j.swevo.2021.100841.
[68] Jonathan McChesney, Nan Wang, Ashish Tanwer, Eyal de Lara, Blesson Varghese, [93] Mohamed Abdel-Basset, Laila Abdel-Fatah, Arun Kumar Sangaiah, Metaheuristic
Defog: fog computing benchmarks, in: SEC’19: Proceedings of the 4th ACM/IEEE algorithms: a comprehensive review. Computational Intelligence For Multimedia
Symposium on Edge Computing, ACM, 2019, https://doi.org/10.1145/ Big Data On the Cloud With Engineering Applications, Elsevier, 2018, https://doi.
3318216.3363299. November. org/10.1016/B978-0-12-813314-9.00010-4.
[69] Lade Prasanth, Ghosh Rumi, Srinivasan Soundar, Manufacturing analytics and [94] Xin-She Yang, Su Fong Chien, Tiew On Ting, Computational intelligence and
industrial internet of things, Internet Things 32 (2017), https://doi.org/10.1109/ metaheuristic algorithms with applications, Swarm Evol. Comput. 62 (2014),
MIS.2017.49. https://doi.org/10.1155/2014/425853.
[70] Voulodimos Athanasios, Kosmopoulos Dimitrios, Vasileiou Georgios, [95] Aleksandra Swiercz, Hyper-heuristics and metaheuristics for selected bio-inspired
Sardis Emmanuel, A dataset for workflow recognition in industrial scenes, in: combinatorial optimization problems, editor, in: Javier Del Ser Lorente (Ed.),
IEEE 2011 18th IEEE International Conference on Image Processing (ICIP 2011), Heuristics and Hyper-Heuristics, Chapter 1, IntechOpen, Rijeka, 2017, https://
IEEE, 2011, https://doi.org/10.1109/icip.2011.6116362. doi.org/10.5772/intechopen.69225.
[71] Dong Yuan, Yun Yang, Jinjun Chen, Computation and storage in the cloud. [96] John H Drake, Ahmed Kheiri, Ender Ozcan, Edmund K Burke, Recent advances in
Understanding the Trade-Offs, Elsevier, Boston, 2013, pp. 109–113, https://doi. selection hyper-heuristics, Eur. J. Oper. Res. 285 (2020), https://doi.org/
org/10.1016/B978-0-12-407767-6.00002-0, pages. 10.1016/j.ejor.2019.07.073.
[72] Poonam Singh, Maitreyee Dutta, Naveen Aggarwal, A review of task scheduling [97] Mohammad Babrdel Bonab, PhD thesis, Universiti Teknologi Malaysia, 2016.
based on meta-heuristics approach in cloud computing, Knowl. Inf. Syst. 52 [98] Sabihe Kabirzadeh, Dadmehr Rahbari, Mohsen Nickray, A hyper heuristic
(2017), https://doi.org/10.1007/s10115-017-1044-2. algorithm for scheduling of fog networks, in: Proceeding of the 21st Conference of
[73] Andrew Stuart Tanenbaum, Maarten Van Steen, Distributed Systems: Principles Open Innovations Association (FRUCT), Helsinki, Finland, 2017, https://doi.org/
and Paradigms, Pearson Prentice Hall, 2007. https://books.google.cm/books? 10.23919/FRUCT.2017.8250177.
id=DL8ZAQAAIAAJ. [Accessed 27 December 2023]. [99] Dadmehr Rahbari, Analyzing meta-heuristic algorithms for task scheduling in a
[74] Amir Vahid Dastjerdi, Harshit Gupta, Rodrigo Neves Calheiros, Soumya fog-based iot application, Algorithms 15 (2022), https://doi.org/10.3390/
Kanti Ghosh, Rajkumar Buyya, Chapter 4 - fog computing: principles, a15110397.
architectures, and applications, editors, in: Rajkumar Buyya, Amir [100] Chun-Wei Tsai, Wei-Cheng Huang, Meng-Hsiu Chiang, Ming-Chao Chiang, Chu-
Vahid Dastjerdi (Eds.), Internet of Things, Morgan Kaufmann, 2016, pp. 61–75, Sing Yang, A hyper-heuristic scheduling algorithm for cloud, IEEE Trans. Cloud
https://doi.org/10.1016/B978-0-12-805395-9.00004-6. pages. Comput. 2 (2) (2014), https://doi.org/10.1109/TCC.2014.2315797.
[75] Samadi Yassir, Zbakh Mostapha, Tadonki Claude, Workflow scheduling issues [101] Willy Malvaut-Martiarena, PhD thesis, Université de Grenoble, 2011.
and techniques incloud computing: a systematic literature review, in: Cloud [102] Jose Octavio Gutierrez-Garcia, Kwang Mong Sim, Family of heuristics for agent
Computing and Big Data: Technologies, Applications and Security, 49, Springer, based elastic cloud bag-of-tasks concurrent scheduling, Future Gener. Comput.
2019, https://doi.org/10.1007/978-3-319-97719-516. Syst. 29 (2013), https://doi.org/10.1016/j.future.2012.01.005. September.
[76] Mala Kalra, Sarbjeet Singh, A review of metaheuristic scheduling techniques in [103] J.G. Nithya, G. Jayapratha, A multi-agent brokering approach and jumper firefly
cloud computing, Egypt. Inform. J. 16 (2015), https://doi.org/10.1016/j. algorithm for job scheduling in cloud computing, in: 2014 International
eij.2015.07.001. Conference on Intelligent Computing Applications (ICICA), IEEE Computer
Society, 2014, https://doi.org/10.1109/ICICA.2014.20. March.
19
[104] Dineshan Subramoney, Clement N. Nyirenda, A comparative evaluation of [117] Vincenzo De Maio, Dragi Kimovski, Multi-objective scheduling of extreme data
population-based optimization algorithms for workflow scheduling in cloud-fog scientific workflows in fog, Future Gener. Comput. Syst. 106 (2020) 171–184,
environments, in: 2020 IEEE Symposium Series on Computational Intelligence https://doi.org/10.1016/j.future.2019.12.054.
(SSCI), 2020, https://doi.org/10.1109/SSCI47803.2020.9308549. [118] Zhenyu Wen, Saurabh Garg, Gagangeet Singh Aujla, Khaled Alwasel,
[105] Hassan Asghar, Eun-Sung Jung, A survey on scheduling techniques in the edge Deepak Puthal, Schahram Dustdar, Albert Y. Zomaya, Rajiv Ranjan, Running
cloud: issues, challenges and future directions, arxiv (2022), https://doi.org/ industrial workflow applications in a software-defined multicloud environment
10.48550/arXiv.2202.07799. February. using green energy aware scheduling algorithm, IEEE Trans. Ind. Inf. 17 (2021),
[106] Shuran Sheng, Peng Chen, Zhimin Chen, Lenan Wu, Yuxuan Yao, Deep https://doi.org/10.1109/TII.2020.3045690. August.
reinforcement learning-based task scheduling in iot edge computing, Sensors [119] Ruimiao Ding, Xuejun Li, Xiao Liu, Jia Xu, A cost-effective time-constrained
(2021), https://doi.org/10.3390/s21051666. multi-workflow scheduling strategy in fog computing, in: Service-Oriented
[107] Junhyung Moon, Minyeol Yang, Jongpil Jeong, A novel approach to the job shop Computing, 11434, 2019, https://doi.org/10.1007/978-3-030-17642-6 17. April.
scheduling problem based on the deep q-network in a cooperative multi-access [120] Harvinder Singh, Sanjay Tyagi, Pardeep Kumar, Sukhpal Singh Gill,
edge computing ecosystem, Sensors 21 (4553) (2021), https://doi.org/10.3390/ Rajkumar Buyya, Metaheuristics for scheduling of heterogeneous tasks in cloud
s21134553. computing environments: analysis, performance evaluation, and future
[108] Omar Elzeki, Magdi Rashad, Mohamed Abu Elsoud, Overview of scheduling tasks directions, Simul. Model. Pract. Theory 111 (2021), https://doi.org/10.1016/j.
in distributed computing systems, Int. J. Soft Comput. Eng. (IJSCE) 2 (2012). simpat.2021.102353. May.
https://www.researchgate.net/publication/284679321. Overview of scheduling [121] Jiuyun Xu, Zhuangyuan Hao, Ruru Zhang, Xiaoting Sun, A method based on the
tasks in distributed computing systems. combination of laxity and ant colony system for cloud-fog task scheduling, IEEE
[109] Goudarzi Mohammad, Palaniswami Marimuthu, Buyya Rajkumar, Scheduling iot Access 7 (2019), https://doi.org/10.1109/ACCESS.2019.2936116. August.
applications in edge and fog computing environments: a taxonomy and future [122] Hongyan Cui, Xiaofei Liu, Tao Yu, Honggang Zhang, Yajun Fang, Zongguo Xia,
directions, ACM Comput. Surv. (2022), https://doi.org/10.1145/3544836 jun. Cloud service scheduling algorithm research and optimization, Secur. Commun.
[110] Mehdi Hosseinzadeh, Marwan Yassin Ghafour, Hawkar Kamaran Hama, Bay Vo, Netw. 2017 (2017), https://doi.org/10.1155/2017/2503153. January.
Afsane Khoshnevis, Multiobjective task and workflow scheduling approaches in [123] Mohamed Abd Elaziz, Laith Abualigah, Ibrahim Attiya, Advanced optimization
cloud computing: a comprehensive review, J. Grid Comput. 18 (2020), https:// technique for scheduling iot tasks in cloud-fog computing environments, Future
doi.org/10.1007/s10723-020-09533-z. August. Gener. Comput. Syst. 124 (2021), https://doi.org/10.1016/j.future.2021.05.026.
[111] Mostafa Ghobaei-Arani, Alireza Souri, Fatemeh Safara, Monire Norouzi, An May.
efficient task scheduling approach using moth-flame optimization algorithm for [124] Ambika Aggarwal, Sunil Kumar, Ashutosh Bhatt, Mohd Asif Shah, Solving user
cyber-physical system applications in fog computing, Trans. Emerg. Telecommun. priority in cloud computing using enhanced optimization algorithm in workflow
Technol. 31 (2020), https://doi.org/10.1002/ett.3770. February. scheduling, Comput. Intell. Neurosci. 2022 (2022), https://doi.org/10.1155/
[112] Khaled Matrouk, Kholoud Alatoun, Scheduling algorithms in fog computing: a 2022/7855532. August.
survey, Int. J. Netw. Distrib. Comput. 9 (2021), https://doi.org/10.2991/ijndc. [125] Dineshan Subramoney, Clement N. Nyirenda, Multi-swarm pso algorithm for
k.210111.001. January. static workflow scheduling in cloud-fog environments, IEEE Access 10 (2022),
[113] Rongbin Xu, Yeguo Wang, Yongliang Cheng, Yuanwei Zhu, Ying Xie, https://doi.org/10.1109/ACCESS.2022.3220239. November.
Abubakar Sadiq Sani, Dong Yuan, Improved particle swarm optimization based [126] Mukherjee Mithun, Matam Rakesh, Shu Lei, Maglaras Leandros, Ferrag Mohamed
workflow scheduling in cloud-fog environment, in: Business Process Management Amine, Choudhury Nikumani, and Kumar Vikas. Security and privacy in fog
Workshops. BPM 2018 342, Springer, 2019, https://doi.org/10.1007/978-3-030- computing: challenges, IEEE Access 5 (2017), https://doi.org/10.1109/
11641-5-27. ACCESS.2017.2749422.
[114] Higinio Mora Mora, David Gil, Flexible framework for real-time embedded [127] Hu Pengfei, Ning Huansheng, Qiu Tie, Song Houbing, Wang Yanna, Yao Xuanxia,
systems based on mobile cloud computing paradigm, Mobile Inf. Syst. 2015 Security and privacy preservation scheme of face identification and resolution
(2015), https://doi.org/10.1155/2015/652462. June. framework using fog computing in internet of things, IEEE Internet Things J. 4
[115] Mao Yuyi, Zhang Jun, B. Letaief Khaled, Joint task offloading scheduling and (2017), https://doi.org/10.1109/JIOT.2017.2659783.
transmit power allocation for mobile-edge computing systems, in: 2017 IEEE [128] Lingling Wang, Guozhu Liu, Lijun Sun, A secure and privacy-preserving
Wireless Communications and Networking Conference (WCNC), 2017, https:// navigation scheme using spatial crowdsourcing in fog-based vanets, Sensors 17
doi.org/10.1109/WCNC.2017.7925615. (2017), https://doi.org/10.3390/s17040668.
[116] Salin Guan, Hirofumi Matsuda, Morikazu Nakamura, Takeshi Shikanai, [129] Al Hamid Hadeal Abdulaziz, Rahman Sk Md Mizanur, Hossain M. Shamim,
Takeo Okazaki, Scheduling for farm work planning based on petri net model and Almogren Ahmad, Alamri Atif, A security model for preserving the privacy of
siulated annealing, Agric. Inf. Res. 16 (2007), https://doi.org/10.3173/ medical big data in a healthcare cloud using a fog computing facility with pairing-
AIR.16.188. based cryptography, IEEE Access 5 (2017), https://doi.org/10.1109/
ACCESS.2017.2757844.
20

1 s2.0 S2452414X2300119X Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S2452414X2300119X Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S2452414X2300119X Main

Uploaded by

Copyright:

Available Formats

Journal of Industrial Information Integration 38 (2024) 100546

Contents lists available at ScienceDirect

Journal of Industrial Information Integration

Industry 4.0 and industrial workflow scheduling: A survey

1. Introduction understanding of the manufacturing process while enabling efficient and

Available online 20 December 2023

Fig. 1. 3-tier fog-cloud architecture.

Fig. 2. Flow of the article.

Fig. 3. Timeline of industrial revolutions [32].

2.3.1. IoT and related technologies 2.3.7. Augmented reality

2.3.2. Cloud computing 2.3.8. Additive manufacturing

2.5. Challenges 2.5.6. Latency

Fig. 4. Taxonomy of scheduling algorithms based on [19].

Fig. 5. Taxonomy of scheduling scheme in the layers of computing.

Mohammadi et al., 2018 MP ILP-DC Cost CPLEX Scientific Cloud

In this paper [106], the problem of scheduling computationally

Fig. 7. Metrics used.

Fig. 10. Simulation tools.

Fig. 11. Publications per year.

focused by several scheduling schemes in 25 %.

• RQ3: What is the most common type of algorithm categorization

According to Fig. 8, the most common category of algorithm used to

You might also like