Pervasive and Mobile Computing 92 (2023) 101785

Contents lists available at ScienceDirect

Pervasive and Mobile Computing

journal homepage: www.elsevier.com/locate/pmc

An efficient dynamic decision-based task optimization and

scheduling approach for microservice-based cost
management in mobile cloud computing applications
∗
Mahmood ul Hassan a , Amin A. Al-Awady a , Abid Ali b , ,
Muhammad Munawar Iqbal c , Muhammad Akram d , Jahangir Khan e ,
Ali Ahmad AbuOdeh a
a
Department of Computer Skills, Deanship of Preparatory Year, Najran University, Najran, Kingdom of Saudi Arabia
b
Department of Computer Science, University of Engineering and Technology, Taxila 47080, Pakistan
c
Department of Computer Science, Govt. ANK(S) Degree College, KTS 22620, Haripur, Pakistan
d
Department of Computer Science, College of Computer Science and Information Systems, Najran University, Najran, Kingdom of
Saudi Arabia
e
Department of Computer Science, Applied College Mohyail Asir, King Khalid University, Kingdom of Saudi Arabia

article info a b s t r a c t

Article history: The use of smartphones and mobile devices has increased significantly, as have Mobile
Received 22 November 2021 Cloud Applications based on cloud computing. These applications are used in various
Received in revised form 23 March 2023 fields, including Augmented Reality, E-Transportation, 2D/3-D Games, E-Healthcare, and
Accepted 1 April 2023
Education. While existing cloud-based frameworks provide such services on Virtual
Available online 6 April 2023
Machines, they incur problems such as overhead, lengthy boot time, and high costs.
Keywords: To address these issues, the paper proposes a Dynamic Decision-Based Task Schedul-
Cloud computing ing Approach for Microservice-based Mobile Cloud Computing Applications (MSCMCC)
Mobile cloud computing that can run delay-sensitive applications and mobility with less cost than existing
Task offloading approaches. The study focuses on Task Offloading problems on heterogeneous Mobile
Task sequencing Cloud servers. It proposes a Task Offloading and Microservices based Computational
Task scheduling Offloading (TSMCO) framework to solve Task Scheduling in steps such as Resource
Microservices
Matching, Task Sequencing, and Task Offloading. Experimental results show that the
proposed MSCMCC and TSMCO enhance Mobile Server Utilization while minimizing
costs and improving boot time, resource utilization, and task arrival time for various
applications. Specifically, the proposed system effectively reduces the cost of healthcare
applications by 25%, augmented reality by 23%, E-Transport tasks by 21%, and 3-D games
tasks by 19%, the average boot-time of microservices applications by 17%, resource
utilization by 36%, and tasks arrival time by 16%.
© 2023 Elsevier B.V. All rights reserved.

1. Introduction

Edge Computing and Mobile Cloud Computing (MCC) enhance the processing capabilities of mobile devices. Due to
the limitations of mobile devices, such as battery life, processing power, resource utilization, and cost, task scheduling

∗ Corresponding author.
E-mail addresses: [email protected] (M.u. Hassan), [email protected] (A.A. Al-Awady), [email protected] (A. Ali),
[email protected] (M.M. Iqbal), [email protected] (M. Akram), [email protected] (J. Khan), [email protected]
(A.A. AbuOdeh).

https://doi.org/10.1016/j.pmcj.2023.101785
1574-1192/© 2023 Elsevier B.V. All rights reserved.
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

is a crucial aspect of mobile computing. Applications like 2D/3D E-Gaming, Augmented Reality, E-Healthcare, and E-
Transport require significant modifications to be processed efficiently on mobile devices [1,2]. Contemporary mobile
devices can gather, handle, and dispatch data even in unfavorable conditions for data collection. The data processing
environment relies on shared characteristics. The motivation for implementing MCC is to broaden the potential of mobile
devices and incorporate task scheduling capabilities within them [3]. Due to their demanding operational requirements,
mobile devices encounter challenges such as limited CPU, low bandwidth, and battery power. As a solution, these devices
delegate their tasks to Mobile Cloud Computing (MCC) to alleviate the burden on their computational and power resources.
MCC is interconnected with Edge Computing, Fog Computing, and Cloudlet. This study focused on the task offloading of
microservices-based applications in MCC. In [4], a proposed framework and model aim to improve processing performance
and enhance application capabilities to save battery life for devices with limited reserves. This framework focuses on
offloading tasks to achieve these goals [5,6]. Using MCC to regulate execution performance, the suggested framework
amplifies application effectiveness and advances work quality. In a study conducted in [7], the research findings show
that the boot time of microservices-based Virtual Machines (VMs) is 28 s, which is a significant delay and adds to the
processing overhead of these applications. The limitations highlighted in this study provide insights into the feasibility of
using a microservices-based task offloading framework, as explored in this research [8].

1.1. Motivation

Over the past few years, resource-intensive and battery-consuming applications have become increasingly prevalent in
mobile devices, including online games, cloud-based apps, and other demanding applications. These applications require
minimal local services to be executed correctly and quickly. At the same time, user-centric applications require the
support of heavy virtual machines (VMs), which can be costly due to their usage-based payment model. Mobility, cost,
and interactivity are the main challenges for the existing mobile cloud computing (MCC) paradigm in providing these
services. Additionally, scheduling resources efficiently for mobile tasks is a significant challenge for MCC in microservices-
based applications [9]. Scheduling tasks is a crucial aspect of mobile cloud computing as it poses significant challenges
related to the limited capabilities of mobile devices, storage constraints, processing capabilities, and network bandwidth
requirements. Conversely, cloud-based mobile cloud computing has vast processing abilities, unrestricted bandwidth, and
no storage restrictions, making it necessary to employ task scheduling for mobile-based microservices applications.
This paper discusses the problem of cost-efficient task scheduling in the context of Mobile Cloud Computing (MCC) for
Internet of Things (IoT) applications. Specifically, the focus is on the MCC-based cloud network, intending to reduce the
cost of mobile application services. The study concentrates on the computational and communication costs associated with
task scheduling and offloading, common issues in mobile applications that comprise independent and fine-grained sub-
tasks. Fine-grained refers to the independent nature of each task, with specific attributes and data that make the workload
efficient. Each task is associated with vector attributes such as CPU instructions, data size, and execution deadline, while
the MCC services are evaluated based on their price and speed. The research considers several questions during the task
scheduling process.

(i) How to select the cost-efficient MCC VM for every task to meet the processing requirements of every task?
(ii) How are offloaded tasks executed sequentially before scheduling on MCC?
(iii) How to accomplish optimal task scheduling to reduce and execute the tasks under the operating costs and defined
deadlines?

1.2. Summary/contributions

The main contribution of the proposed system is to save time and computational energy using the mobile cloud
computing approach as follows.

(i) Our proposed method, called Microservices Container-based MCC system (MSCMCC), facilitates computational
offloading between mobile devices and resource-rich MCC models. To improve the performance, we implemented a
Docker container that reduces the overhead for services and lowers the VMs’ boot time.
(ii) While considering MCC servers with different attributes, we also considered the individual quality-of-service
requirements for each task. We selected the MCC servers with VMs that met the service demand and developed
a service-matching algorithm to execute instructions based on service requirements and tasks.
(iii) We adopted the FCFC and SJF task sequence algorithm to schedule randomly generated tasks from mobile devices
and put slack time and task size in sequences.
(iv) The computational cost and communication time for scheduling fine-grained tasks from local mobile devices towards
MCC VMs pose a significant challenge. Therefore, we proposed a microservices-based cost-efficient task scheduler
algorithm that reduces the overhead cost of mobile tasks to MCC servers.
(v) Our novel contribution involves sequencing tasks based on their priority, with tasks requiring special consideration
during offloading given the highest priority.
The remaining sections of this paper are structured as follows: In Section 2, previous research on task scheduling and
fault tolerance is reviewed. Section 3 outlines our research approach and offers solutions to the problems discussed in
Section 1. Section 4 presents our proposed method and its simulations using an MCSMCC approach. Finally, Section 5
concludes our proposed system, emphasizing its simplicity in fault tolerance methodology.
2
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

2. Literature review

The popularity of computational offloading of mobile applications among mobile cloud users has increased with the
rise of mobile devices and mobile applications. This method enables mobile devices with computationally intensive tasks
to shift their workload to the cloud for execution on mobile cloud servers. [10]. Outsourcing mobile tasks is one way to
enhance the performance of mobile devices and their batteries. To achieve this, mobile cloud support transfers heavy-
duty application tasks away from the mobile device. Previous research has focused on optimizing mobile application
performance while reducing computational costs and conserving the processing power of mobile devices [11].
In [12], the authors’ CloneCloud framework aims to enhance the battery life and execution speed of mobile devices by
leveraging Centric Cloud Servers to handle computationally intensive tasks. In the paper [13], the proposed framework by
the authors aims to conserve the mobile device’s battery by only running computationally intensive tasks on it. Another
approach, MAUI, uses profiling technologies to determine whether to execute the task locally or on the mobile cloud
server. The key distinction is to make the optimal decision for offloading the task based on runtime. The primary goal of
this approach is to efficiently offload the task during its runtime.
As far as we know, there has not been a microservices-based MCC framework introduced yet that caters to time-
sensitive and intricate tasks. In this context, we suggest a new MCSMCC mobile cloud computing framework designed
to perform tasks with the least energy and cost. Furthermore, we present a cost-conscious framework CCCOF that
aims to optimize the cost-centric and computational offloading procedures. CCCOF ensures the quality of service by
running services within the given time constraints while keeping the costs of resources at a minimum. In [14], the
ThinkAir approach, introduced by the authors, offers a way to transfer mobile device tasks to the mobile cloud through
a computational offloading framework. This can be achieved by utilizing mobile virtualization technology and facilitating
task offloading. The authors also developed meta-heuristic algorithmic methods to tackle task-scheduling challenges in
conjunction with the existing task-scheduling frameworks [15,16].
The resolution offered by computation offloading frameworks is unsustainable due to cloud network latencies. As
a result, there is a greater need for problem-oriented cloudlet-based computational offloading resources that cater to
the latency challenges of wireless-based access networks. Cloudlet frameworks have addressed this issue by bringing
mobile devices closer to computational resources. In [17], the virtual machine-based Cloudlet framework proposed by
Satyanarayana et al. offers elasticity, scalability, and mobility. This framework brings Cloudlets closer to mobile phones,
serving as a single hop toward cell phones. Another researcher presented Rattrap [18], a mobile cloud-centric environment
that was proposed to offload the computational tasks of mobile devices to the cloud using an Android-based system.
Virtual machines with containers were utilized in this research to reduce the boot time of cloud platforms’ monolithic
services. However, while this approach effectively reduces service boot time, it fails to meet the fine-grained resource
requirements of mobile devices for resource-intensive applications.
Certain research contributions improve the cost-effectiveness of services provided by mobile applications. In [19], the
authors leverage various mobile cloudlet services with varying characteristics to enhance cost-effectiveness. The authors
evaluate computation time, communication time, and deadline cost as essential factors. In [20–22], The study investigates
cost-efficient and energy-efficient task offloading in mobile cloud computing. Its primary aim is to preserve the battery
life of mobile devices by transferring tasks to the mobile cloud. The authors perceive resources as storage and propose a
mobile cloud-based computational offloading framework for task scheduling that is presented as an effective computing
model. The [23–25] research objective is to analyze the resource consumption pattern of mobile applications in mobile
cloud computing. Various cost pricing models, such as spot instances, on-demand, and on-reserved systems, are discussed
to ensure efficient collaboration for resource consumption. The primary goal of these studies is to minimize resource rental
expenses and deliver mobile services within their designated time frames.
Furthermore, the authors in [26–28] aim to improve the performance of mobile cloud applications by exploring cost-
efficient and cost-effective real-time analysis techniques to mitigate the running costs associated with task scheduling.
The studies aim to optimize task execution by identifying effective workflow categories and reducing data transfer and
execution time. Additionally, the studies consider mobile device services’ computational and communication costs during
task scheduling in MCC [29]. In [30], the authors employ mobile edge computing to reduce service latency, optimize
revenue, and enhance the quality of services. They present the benefits of increased revenue and service utilization.
Although the authors enhance the usage, service latency, revenue, and utility value, they do not consider cost management,
resource matching, task deadline management, server utilization ratio, and the time to boot up microservices. Authors
in [31] aim to create a mathematical model for a microservice scheduling framework that enhances satisfaction, reduces
network delay, energy consumption, average price, and failure rate, and improves network throughput. However, the
authors do not focus on addressing issues such as service delivery latency, cost management of microservices, resource
matching, task failure ratio, service utilization, and task sequencing, as these parameters are outside the scope of their
research. In [32], the author introduced an auction mechanism proposed as a technique to model the interaction among
Mobile Edge Computing systems. The approach employs a justified methodology to manage offloading tasks and measure
performance. Table 1 summarizes the literature, which includes Task Sequencing, Server Utilization, Task Failure Ratio,
Handling Task Deadlines, Boot Time, Cost Management, Resource Matching, and Service Delivery. This approach considers
these parameters, which were not considered by previous approaches.
This related work section proposes several algorithms, and our proposed methodology aims to tackle the problem of
migrating microservices for content-sensitive applications. While multiple techniques are suggested to address this issue,
3
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Table 1
Based on their characteristics, the proposed system can be compared to related work.
Papers Task VM server Fault Deadlines Load Process cost Resource Latency
sequence use tolerance time organization complementary
Lee et al. (2019) ✔ ✔ – ✔ ✔ ✔ – –
Raju et al. (2016) ✔ ✔ – – ✔ – – ✔
Abd et al. (2019) ✔ ✔ ✔ – ✔ – – ✔
Park et al. (2016) ✔ ✔ ✔ ✔ – ✔ ✔ ✔
Al-Syed et al. (2016) ✔ ✔ – – – – – –
Keshanchi et al. (2017) ✔ – – ✔ – – ✔ –
Peng et al. (2019) ✔ – ✔ – ✔ – ✔ –
Tang et al. (2018) ✔ – ✔ – – – – ✔
Lin et al. (2014) – – ✔ ✔ – – – ✔
Guo et al. (2022) – – ✔ ✔ – ✔ – –
Wei et al. (2023) – ✔ – – ✔ – – –
Proposed methodology ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔

they do not consider the computational cost requirements and demands for future needs. In the next section, we will
discuss the standard methods used to address the problem and provide a detailed description of the problem.
As the use of microservices increases on mobile devices, the tasks required become more resource-intensive in
today’s environment. Instead of performing entire tasks, only the necessary operations must be performed. The primary
motivation behind offloading microservices-based tasks is to reduce the resources used and increase cost-effectiveness
through virtual machine-based resources. Previous studies have focused solely on task scheduling on collective tasks with
individual services. However, it is essential to consider cost-effectiveness and resource management for each microservice.
This approach offloads tasks to the Mobile Cloud Computing (MCC) virtual machine.
To the best of our knowledge, there has not been any suggestion of a framework for a Dynamic Decision-Based Task
Scheduling Approach in Microservice-based Mobile Cloud Computing Applications. This research proposes the MSCMCC
framework, which enables microservices-based task execution on mobile devices at a very low cost. The framework
ensures Quality of Service for microservices-based mobile device applications over mobile cloud computing. MSCMCC
enhances application efficiency by providing an effective resource constraints framework that enables task execution
under a deadline while minimizing application cost.

2.1. Problem description

The literature has explored task offloading in mobile cloud computing [25,33]; however, task scheduling has become a
concern. To address this issue, we developed the Cost-Centric and Computational Offloading Framework (CCCOF), which
considers communication, offloading cost, and processing power during scheduling and offloading. Our paper aims to
reduce computation and communication costs while maximizing processing power on both MCC and mobile devices.
To address this problem, we propose the MSCMCC framework, which consists of three layers, as shown in Fig. 1. The
Task Scheduling Layer, which includes the System Monitoring Agent, Cloud Computing Agent, and Task Sequencing using
sequence algorithms, is responsible for scheduling decisions. The MSCMCC framework is higher-order, while TSMCO and
CCCOF are lower-order frameworks within the proposed methodology. CCCOF employs a cost model to compute the
resource access and cost for microservices-based tasks. The cost is computed for tasks that must be scheduled on the
MCC server.
Furthermore, TSMCO has been developed to match resources between offloaded tasks and MCC servers. The advantage
of optimizing tasks is that it allows for the pair-wise processing of diverse applications. In the subsequent section of the
paper, we elaborate on the proposed architecture for offloading tasks using MSCMCC, CCCOF, and TSMCO.

3. Proposed microservices-based MCC architecture

The MSCMCC framework proposed in Fig. 1 comprises three layers: Mobile Users Layer, Task Scheduling Layer, and
Mobile Cloud Layer. Typically, heterogeneous mobile users generate tasks to offload, which are then passed to the Task
Scheduling Layer via an API. The Task Scheduling Layer consists of four main modules that receive the offloaded tasks. The
Cloud Computing Agent (CCA) manages and handles all offloaded tasks and is supported by the System Monitoring Agent,
Task Sequences, and Task Scheduling Handler. Acting as an intermediary between mobile devices and system resources,
CCA collects data such as configuration information, metrics, and logs from mobile device APIs. All components of the
MCC virtual network, including the System Monitoring Agent, Task Sequences, and Task Scheduling Agent, exist on MCC
servers. CCA utilizes the functionality of all three modules to measure performance and manage workload requests.
We have primarily utilized FCFC and SJF algorithms for task sequencing from mobile devices. It is important to note
that each task in this scheduling includes vector attributes such as execution time, execution deadline, data size, and
CPU instructions. Rules for task sequencing have been proposed based on sorting algorithms to ensure minimum cost
overhead rules are utilized. The System Monitoring Agent maintains a table that includes tasks and their resource list,
4
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 1. Proposed MSCMCC system architecture.

updated after every event in the Task Scheduling Layer, with information about resources consumed or left from the
system or task completion. In the Mobile Cloud Layer, tasks are swapped among MCC servers, resulting in improved early
task scheduling and an enhanced cost function. The task-swapping mechanism among MCC servers enhances resource
handling and other task provision mechanisms.
The Mobile Cloud Layer comprises various types of Mobile Cloud servers that utilize VMs to execute and manage
resource allocation. These VMs are in the topmost layer of MCC server layers. Within each VM, the MCC Server engine
adds or removes microservices as needed to handle tasks. These microservices are managed through containers and
communicated via REST-API services by the container MCC Server engine. To load any necessary data or functions required
for microservices offloading, DB/Libraries are connected to the MCC servers.

3.1. Task clarification using microservices

The Mobile Devices produced independent tasks from different mobile device applications. These tasks contain data
embedded in their applications. All tasks are running independently and do contain their data and processes. Based on
the characteristics of these tasks, every task contains a deadline for execution, data size, and CPU time to execute before
scheduling from MCC. A random method loads these tasks onto the task scheduling and communication mechanism.
Randomly these tasks are loaded onto the MCC system as these tasks before scheduling are not preempted on MCC.

3.2. Tasks mobility

We have integrated the MCC cloud servers at the end of the MCC network. We have introduced a Mobility Module in
CCA that allows the MCC network to choose the mobile users or subscribers for packet delivery. One of the packet delivery
schemes is Location Management. CCA also manages all mobile subscribers’ connections, including the attachment points
for the final task scheduling. Handoff Management is a crucial factor that determines the scenario for handoff management.
The Task Mobility Section deals with the issues and features of location and handoff management problems. In MCC,
Mobile Devices change their location using Mobility options, which trigger their connection from one access point to
another. This process is known as Handoff Management. We divide the handoff management into three primary stages in
MCC Mobile Device connections. First, the activation process of MCC Mobile Devices is caused by the changing condition
of Mobile Devices, the MCC network agent, and the MCC network. Second, when a new Mobile Device connection is
established, CCA performs additional routing and discovers a new handoff connection tool. Third, the data is delivered
from the old connection points to other connections to support the quality of service operations and handling procedures.
The old connection refers to the connection of Mobile Devices in the previous handoff process, while the new connection
refers to the handoff towards the new MCC access point.
5
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

3.3. Microservices-based MCC servers characterization of resources

The MCC paradigm employs Mobile Cloud Layer containers as a standard methodology for running Mobile Applications
in the MCC network. This emerging approach has resulted in significant advancements in managing group containers.
The primary challenge, however, is deploying microservices containers through the MCC server layer. Microservices are
small, self-contained services that interact with external resources through well-defined APIs. Self-established teams in
MSCMCC own these microservices. The approach utilizes heterogeneous mobile cloud servers with VMs on each server
to optimize performance. The MCC server engine is responsible for adding or removing microservices quickly. The Mobile
Cloud Layer comprises heterogeneous MCC servers with devices and processors with various computational capabilities,
runtime, bandwidth, costs, and VMs (container microservices).

3.4. Runtime microservices in MSCMCC

Microservices is a methodology for developing runtime software that involves arranging an application as a series
of loosely coupled services, a variation of the service-oriented architecture model. Within microservices, protocols are
lightweight, and services are fine-grained. In contrast, MSCMCC utilizes a single VM per MCC server, while a single VM
can support multiple microservices at once in the microservices architecture. Every microservice in the microservices
architecture runs a single task at a time. CCA employs cost-efficient microservices to handle computational tasks, resulting
in efficient outcomes. Each microservice in CCA contains libraries and resources related to its specific task. During task
offloading in MSCMCC, mobile application tasks are handled through fine-grained microservices architecture.

3.5. Problem formulation

We began by selecting a set of delay-sensitive tasks, T = {t1 , t2 , . . .. . ., tn }, to be offloaded to the MCC server. These
tasks were then scheduled using the MCC server. Since we were dealing with microservices, each task was accompanied
by certain attributes, denoted as ti = {Tw , Tdata , Td , Tstorage }. The Tw attribute indicated the task’s workload, Tdata showed
the data associated with the task during its transmission from the mobile device to the MCC server VM, Td represented
the task’s deadline, and Tstorage indicated the storage requirements for the task.
Assuming the MCC cloud configuration, we have a set of N MCC servers, denoted as M = {m1 , m2 , . . . , N}. Each MCC
wj
server mj has a set of attributes, including BMCC , which represents the bandwidth between the MCC Centric Agent and
the MCC server for task offloading, ξj , which represents the computing rates of the jth MCC server, Scj , which represents
j
the total storage capacity of the MCC server j, and VMmic , which represents the deployed VMs for microservices through
j
the MCC server j, capable of handling tasks. Each VMmic comprises multiple containers for executing the microservices to
perform various tasks, with each microservice having its database and libraries required for execution.
j
To schedule a task at MCC server j, the required bandwidth ti is denoted by Bi , and the resources, such as Storage,
RAM, Bandwidth, etc., are denoted by Cj . The offloaded task cost to MCC server j is denoted by ti , and the task’s state is
denoted by yj = 1. Notations for limited page space are shown in this paper, and the remaining notations are illustrated
in Table 2. Eq. (1) shows that the binary variable sij determines whether task ti is scheduled through MCC server Mj , with
sij being either zero or one for all application tasks. The proposed servers decide on task scheduling.

1,
{
ti ← Mi
sij = (1)
0, otherwise
In this problem context, each task is assigned to a specific Mj , and the MCC server schedules tasks concurrently. The
task assignment ti to MCC server M is illustrated in Eq. (2), where the number 1 represents the ready tasks assigned to
the MCC server for processing on its VMs. It indicates a positive task assignment to the appropriate server VM out of N
tasks.
N
∑
sij = 1 (2)
j=1

Each MCC server has a finite resource capacity, ensuring that the demands for task offloading do not surpass the server’s
capabilities. To verify the task scheduling requirements for offloading, Eq. (3) presents the scenario for assessing the task
offloading capabilities of scj and sij against scj . The aim is to ensure that no task processing requirements go beyond the
limitations of the MCC server.
N
∑
Scj *Sij ≤ Scj (3)
i=1

The MCC resource server possesses limited resources, including many virtual machines (VMs) available to execute mi-
croservices for coordinating tasks. Therefore, Eq. (4) indicates a decision-making process that assigns fewer computational
6
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Table 2
Notations with description.
Symbols Description
N N represents the task set that includes all mobile applications.
M M refers to the total number of MCC servers in the model.
sij The notation sij indicates that the ith task is assigned to the jth MCC server.
yj The notation yj denotes the state of the MCC server Mj , which can be ON or
OFF.
ti The notation ti represents the ith task of the mobile application.
Mj The notation Mj represents the jth MCC server in the model.
Scj Scj refers to the storage demand of the tasks ti .
Td Td represents the deadline for the task ti .
Tw Tw represents the computational workload or data of the task ti .
Tdata Tdata represents the size of the task ti during transmission from the mobile
device to the MCC server.
j j
Bi The notation Bi Represents the total bandwidth the task demands to the MCC
server Mj .
Scj The notation Scj represents the capacity of the jth MCC server.
j j
VMmic The notation VMmic represents the total number of VMs deployed at the MCC
server Mj .
ξj The notation ξj represents the computational speed of an individual VM under
the MCC server Mj .
wj wj
BMCC The notation BMCC represents the bandwidth between the MCC Centric Agent
and the MCC server Mj during the task offloading ti .
Cj The notation Cj represents the cost of the jth fog server in the method.

tasks. Each requested task, which consists of microservices, must utilize less computation than the capacity of the VMs.
N
∑ j
Tw *Sij ≤ VM mic (4)
i=1

Each designated task, denoted as ti , is assigned to an optimal MCC server, Mj , by the Cloud Computing Agent. The execution
of the task on the assigned server is determined using Eq. (5), which considers the task’s resource, storage, and time
requirements to select the appropriate server, Mj , for the final decision.
M
∑ Tw
Ttei = *Sij (5)
RMj
j=1

As a result of the Cloud Computing Agent’s decision to assign a task ti for computation on the MCC server, the task now
has additional options for computational offloading. It can transmit the results to the Mobile Device from the MCC server.
The computational process is depicted in Eq. (6).
⎛ ⎞
T− Dateenter
i T− Dateleave
i
RTT = ⎝ + ⎠ (6)
w j up w j down
BMCC BMCC
ij ij

Eq. (6) defines the input and output data sizes of task ti in terms of T − Dateenter
i and T − Dateleave
i , respectively, after being
w jij up wj dow n
processed by the MCC server Mj . The bandwidth for uplink and downlink, denoted by BMCC and BMCC Respectively,
ij
it represents the link rates for offloading data from the mobile device to the MCC server and receiving the results back.
The Round-Trip Time (RTT ) is the time between sending and receiving data for all tasks (microservices). We have measured
the bandwidth required for each task, and Eq. (7) shows the bandwidth requirements for each task for RTT .

Sij RTT + Ttei ≤ Td

( )
(7)
Eq. (7) outlines the bandwidth requirements. However, since each task varies, so does the necessary bandwidth. Eq. (8)
demonstrates the bandwidth inequality of task ti . The bandwidth is evaluated based on each VM Time, Data, and
computational speed under MCC server Mj . Eq. (8) is utilized to determine the bandwidth inequality.
w ij T − datai
BMCC > i
(8)
Tw
Td − ξj

To attain the desired performance of the suggested system, we ensure that each task denoted as ti , which originates
from mobile devices, is completed before its deadline. This approach reduces the overall cost of the MCC server, Mj . The
7
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Table 3
MCC servers unit price.
MCC servers On-demand δj MCC server
state
wj
BMCC /Cost Scj /Cost CCPU /Cost ∂j
MC1 0.4 0.2 2 0/1
MC2 0.6 0.4 4 0/1
MC3 0.8 0.6 6 0/1

calculation of the necessary communication bandwidth between the MCC server and mobile application is expressed in
Eq. (9). Following this, the communication bandwidth for task ti is determined by evaluating the inequality bandwidth,
which determines the actual and anticipated bandwidth required for offloading the task.
wij T − datai
BMCC = i
(9)
Tw
Td − ξj

Each MCC server, denoted as Mj , has a finite amount of bandwidth and a limited number of VMs, in addition to the
individual task bandwidth. Therefore, the task allocation to Mj must use less bandwidth than the server can handle. Eq. (10)
states that the total bandwidth consumed by the tasks should not exceed the bandwidth required by Mj to execute wj
tasks. The variable wij represents the necessary bandwidth for executing the tasks in the correct order.
N
wij wj
∑
BMCC *Sij ≤ BMCC (10)
i=1

The Cloud Computing Agent (CCA) functions as a connector and monitor to the MCC server, overseeing its performance
regularly. The cost of the MCC server is determined by two primary factors: its state and the resources required for
microservices for each offloaded task. CCA handles the tasks of the user’s mobile application, and its cost is based on
its operational state and the computational capabilities requested from the MCC server. We use the binary variable ∂j to
indicate the status of the MCC Server, as expressed in Eq. (11). This variable reflects whether the server is on or off for
task processing.
1,
{
Mj ← on
∂j = (11)
0, off,

3.6. Cost-Centric and Computational Offloading Framework (CCCOF)

This section outlines the MCC Cost Model, which comprises the Cost-Centric and Computational Offloading Framework
(CCCOF) and defines the quality of service. Microservices are not standalone resource computational applications.
CCCOF includes a cost model that explains the on-demand resource access method, ensuring connectivity based on
the microservices of the business applications framework. Eq. (12) demonstrates the on-demand cost model for mobile
applications, which calculates the processing demand for each selected application used in the simulation.
Cj = δj *sij *Ttei (12)
Table 3 presents the unit price for computational work, δ _j, for each MCC server. The resource constraints for each server
are then computed using Eqs. (13)–(25), which collectively determine the constrained resources for MCC servers MMC1
to MMC3 . The state of each server is monitored through CPU costs, bandwidth requirements, and microservices-based
processing tasks, which must be distributed against every resource. The decision to offload tasks is made using the
parameters in Table 3, represented by Sij = {0, 1}. Eq. (13) calculates each task’s minimum required resources (Rc) based
on On-Demand δj and the MCC Server State ∂j . Eqs. (14), (16), (17), (18), and (19) are then used to compute the resources
demand, the time for resource computation, the task of mobile applications, and the comparison and execution of Cloud-
based resources tasks. Eqs. (20) and (21) determine the capacity of the jth MCC server for resources and set it to 1 to show
that resources are ready to offload the task to MCC servers. Finally, Eqs. (22), (23), and (24) consider server, bandwidth,
and VM-based resources required for each microservices-based task to be offloaded to the cloud. Ultimately, Sij = {0, 1}
decides to offload the task and the required resources.
M N
∑ ∑
min Rc = ∂j *Cj ∀i ∈ N (13)
i=1 j=1
M N
∑ ∑
subject to min Rc = Sii *∂j *Cj ∀i ∈ N (14)
i=1 j=1

8
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

j
T0 = 0, ∀ {j = 1, 2 . . . , N } (15)
N
j j
∑ j
Tk = Tk − 1∗ Sk ∗Ttek ∀ {j = 1, 2 . . . , N } (16)
k=1
N
∑ Tw
Ttei = Sij × ∀ {i = 1, 2 . . . , N } (17)
ξj
j=1
N
∑ j
MC i = Tk ∗Sij ∀ {i = 1, 2 . . . , N } (18)
j=1

MC i + RTT ≤ Tdi (19)

N
∑
Sij = 1, ∀ {i = 1, 2 . . . , M } (20)
j=1
M
∑
Sij = 1, ∀ {i = 1, 2 . . . , N } (21)
j=1
M
∑
Sij *Sci ≤ Scj , ∀j ∈ 1, 2, . . . N , (22)
i=1
M
∑ j
Sij ≤ VM mic , ∀j ∈ 1, 2, . . . N , (23)
i=1
M
wj
∑ j
Sij *Bi ≤ BMCC , ∀j ∈ 1, 2, . . . N , (24)
i=1

Sij = {0, 1} (25)

3.7. Task Scheduling and Microservices based Computational Offloading (TSMCO) framework

The TSMCO framework comprises various components and presents a new approach to task scheduling and
microservices-based computational offloading. It includes the Resource Matching Algorithm and Task Scheduling Algo-
rithm, which work together to determine the processing of tasks and create a sequential list of completed tasks from
9
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 2. The TSMCO Framework has been implemented with a Resource Matching Algorithm and Task Scheduling Algorithm to receive microservices-
based tasks from mobile devices. This framework is used for task offloading, and the MCC Scheduler utilizes virtual machines to process the offloaded
tasks.

mobile devices. The framework offers a resource matching system that matches the required resources of incoming tasks
with MCC VM servers, and once a match is found, it proceeds to sequence the tasks. Fig. 2 provides a visual representation
of the TSMCO framework components, with the first module being resource matching for incoming mobile device tasks
and matching them with MCC servers for processing. Task Sequencing is a crucial component of TSMCO, as it uses sorting
algorithms to sort tasks into different categories for cost-effective scheduling. Task sequencing provides input for task
scheduling, with the task sequence ti scheduled on the MCC server MCj if Sij = 1 or Sij = 0. The task sequencing process
continues until all tasks are scheduled and processed according to their deadlines using MSCMCC. All mobile tasks are
passed through multiple components to complete their executions.
Algorithm 1 is defined to schedule tasks on each virtual machine using heterogeneous MCC servers. The goal is to
minimize Rc by scheduling and sequencing tasks according to the given parameters.

3.8. MCC server resource attaining

The proposed system aims to optimize costs and involves using heterogeneous MCC servers. The ideal approach is to
choose the most effective edge MCC server for processing all extended sequential tasks to achieve the best results. Our
task scheduling objective for microservices applications is to minimize processing and computation costs for the MCC. One
of the challenges is selecting an MCC server with the minimum cost δj . This can be achieved using Eq. (26) and Eq. (27),
which demonstrate the unit cost δj and smaller MCC server costs, respectively.
Cj
δj = (26)
ρj
Eq. (26) utilizes the ρj to determine the size of the MCC server Mj based on its processing power, memory, and task
processing capacity. The resulting output of Eq. (26) is the overall cost associated with the chosen MCC server. Eq. (27)
determines the unit cost δj , which considers the demands for MCC tasks, MCC VM, and MCC bandwidth handling. The
unit cost is computed to facilitate the offloading of tasks with a cost-effective approach.

MCC Scj MCC VM j MCC Bwj

ρj = ∑ N + ∑N mic
j
+ ∑N MCC
wj
(27)
i=1 Scj i=1 VM mic i=1 BMCC
10
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Once the servers have successfully processed the initial level tasks, the leftover resources on the MCC server Mi are not
required to be wasted and are represented by the average of µMi . The dot product γi denotes the task that maximizes the
products and their primary operations. The values of γi can be determined using Eqs. (28)–(31), which comprehensively
explain the proposed system.

γi = µMi *βτij (28)

γi = Sci *Rlj *qi + qi *ν +l
j c*blj *qi (29)
µ Mi = ∗qi *ν ∗ ,
Rlj l
qi blj
( )
j *qi (30)
∑
µMi = YMj − β t∗ j (31)

The symbol β _(t_(*j)) represents the scheduling of resource management tasks on the MCC server Mj , which has
varying types and quantities of resources. The goal is to match tasks with the optimal and most suitable resources in
the heterogeneous MCC server system. Tasks are characterized by vector attributes like the deadline, data size, and
workload, while resources have vector attributes like bandwidth, cost, VM capacity, and storage. To accomplish this
resource matching, we utilize two algorithms: The technique for Order of Preference by Similarity to the Ideal Solution
(TOSS) model and the Analytic hierarchy processing (AHP) model. Our approach uses Algorithm 2 to match MCC server
resources to tasks, where resources and tasks are inputted sequentially. The resulting frequent list array, FLIST[], stores
fulfilled task requirements that are then allocated to the MCC server MCj .

Step 4 of Algorithm 2 checks the resource requirements of all incoming tasks ti for the MCC server. If the resources
match the requirements of the MCC server, the algorithm returns true. Then, it returns false. Once a true result is obtained,
the matching list is added to the Frequent List PLIST [kj , ti ]. This process is repeated for all possible tasks received from
mobile devices to match the MCC server’s heterogeneous requirements.

3.9. MCC server task sequencing

The tasks arrive randomly from various mobile devices, and their arrival rate is tracked through the Poisson Process.
The task sequences have no specific order; they are allocated randomly without any rule. All tasks must be sequenced,
and the sequenced tasks should be properly formatted and provided in the correct sequential order. Three parameters
are considered to sequence the tasks: task size, deadline, and slack time. Four rules have been developed and deployed
in the proposed system to sort all incoming unordered tasks.-

1. The First Come First Served (FCFS) algorithm involves arranging tasks based on their arrival times. The tasks are
sorted and placed in a queue in the order they arrive, without any specific priority due to the nature of FCFS.
Even late tasks are arranged based on their arrival times. The effectiveness of FCFS in addressing task lateness is
determined by its adherence to this sorting approach.

FCFS = Td − Mi (32)
11
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 3. The Cloud Computing Agent utilizes FCFS, SJF, and SSJ task scheduling algorithms to adjust task sequences for applications based on their
arrival time.

2. The SJF (Shortest Job First) scheduling algorithm arranges tasks based on their computation time. The shortest tasks
are prioritized for execution ahead of longer tasks, resulting in minimized lateness for the shorter tasks. The longer
tasks are scheduled for execution later.
3. The SSF (Shortest Size First) approach, as outlined in Ref. [34], involves sorting tasks based on their size, with shorter
tasks given priority over longer ones. This method uses pre-emptive task scheduling for task sequences, resulting
in efficient and dependable data provision.

The tasks offloaded from mobile devices arrived at the MCC in a random order, and initially, they were arranged using a
first-come, first-served (FCFS) sequence. The FCFS method arranges the tasks cost-efficiently, following the order sequence
rules. Fig. 3 illustrates the different task sequencing techniques available for task offloading, including FCFS, SJF, and
SSF, each producing different scheduling and sequencing results. The cloud computing agent chooses the best method
to achieve optimal task sequencing based on the problem objectives. The cloud computing agent selects the scheduling
algorithm and computes the task values based on the task size, deadline, and slack time to decide which tasks to prioritize.
Once the task values are calculated, they are matched with the computational capabilities of FCFS, SJF, and SSF, as
explained in Refs. [30], [31], and [32], respectively. Since tasks are processed in batches, the cloud computing agent
selects the best sequencing method (FCFS, SJF, or SSJ) for each batch of incoming tasks based on decision parameters
that determine whether the tasks are computed based on their sequence numbers. Fig. 3 provides an overview of the
offloaded tasks, the responsibilities of the cloud computing agent, and the task sequencing adjustment for FCFS, SJF, and
SSJ tasks.

3.10. Task scheduling

Until now, we have completed the resource matching and task sequencing. We need to adopt a task-scheduling
methodology to address the problem. However, the initial implementation of the task scheduling policies described in
the paper is insufficient for calculating the mobile application’s cost. The cost can be evaluated in various ways, and
due to concurrent changes in the MCC cloud network, the initial selection may not be appropriate for task scheduling.
Furthermore, the instability of MCC resources is another reason not to finalize the initial solution. Therefore, a new and
improved solution is necessary to enhance task scheduling. For instance, we can consider executing tasks T1 and T2 on
12
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 4. The task offloading for MSCMCC is done through the Mobile Cloud Computing Agent Console using the REST API. Afterward, the tasks are
transmitted to the Edge X Foundry Fog Server Runtime Platform, which is processed on the MCC VM server.

heterogeneous MCC servers with VMs K1 and K2 . The required resources for T1 and T2 include a deadline of 24 and 50,
data sizes of 15 Mb and 35 Mb, and CPU requirements of 12 and 32, respectively.
At first, MCC server K1 is assigned task T1 , while MCC server K2 is assigned task T2 . The overall cost of the application
is the combined sum of the costs for both MCC servers. Eq. (33) illustrates the total cost of the application, indicating the
reduced cost necessary for the application to offload the task.
2
∑
ωtotal = Ki ∗Ti (33)
i=1

If all tasks are scheduled on the MCC server K2 , the total cost of the application is determined solely by the cost of
the MCC server K2 . Although the tasks are executed on the same server, this reduces the computational costs of the
application. Fig. 4 illustrates multiple solutions for sequencing a single task, and we are willing to accept the worst
solution for the processes. The scheduler’s challenge is to select the best solution that reduces the system’s total internal
cost. In the example, the scheduler chooses an MCC server with resources that match the cost in the resource list, but
additional optimization is necessary to achieve high-cost reduction. Initially, the MCC server had a high cost of available
resources, so the main challenge was to reduce the resource utilization cost while utilizing the maximum resources.
Therefore, we propose an improved Task Scheduling methodology that significantly improves the resource utilization of
MCC servers. The primary objective of the scheduling is to assign tasks to MCC servers with the lowest cost during the
initial scheduling phase, which is the only way the Task Scheduling Algorithm can enhance the MCC server’s performance.
The scheduler eliminated extra and expensive costs during the initial collaboration stage. However, we also propose task
scheduling algorithms that enhance scheduling performance and solve the resource optimization problem on the MCC
Server. Algorithm 3 defines task scheduling on the MCC server with optimized resource utilization. It takes as input a set
of tasks that must be scheduled on a heterogeneous MCC server. The algorithm executes as follows:

• Online 1–6 involves the declaration of variables. MCC servers are then sorted in descending order based on the values
of ρj and γi (derived from Eqs. (27) and (29)).
• Following this, the MCC server is loaded with all applications, and their respective resource and task requirements
are specified on lines 7 to 10.

13
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

• The task involves selecting the smallest MCC server δj from a set of T tasks, including some unpublished tasks G. The
unscheduled tasks are assigned to the MCC server based on their resource requirements and available resources the
server. Then, the smallest unit cost server Mj is selected from the available MCC servers Mi . If the scheduled cost of
the available fog servers meets the resource demand, the most significant task is scheduled on the MCC server Mi .
However, if the server cannot meet the requirements, it is selected to fulfill the demand. Steps 11–20 outline these
procedures.
• We select MCC servers Mg1 and Mg2 and apply the cost function W. Swapping a task between the MCC servers is
only allowed if the resources on Mg1 fulfill the requirements of task ti on Mg2 . The task scheduling variable executes
the task swapping on both servers. Once all the tasks on Mg2 are completed, we update its state and choose a new
server from E with the lowest cost to efficiently propose new information.
14
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Table 4
Simulation parameters.
S.No. Parameters Used values
1 Windows OS Docker Engine
2 Language Python, JAVA, XML
3 Processor X64 bit
4 Android Phone LG W11, W31
5 Simulation Time 14 h
6 Simulation Repetition 180 Times
7 Android OS Orio
8 App Interface Android Orio Interface
9 Simulation Evaluation Single and Multiple Factor Anova
10 Service EC2 (Amazon On Demand)
w j(up)
11 BMCC 50 ≈ 1200 Mbps
w j(dow n)
12 BMCC 1200 ≈ 50 Mbps

Table 5
MCC server specifications.
Cloud name VM Core MIPS/Core VMs Storage Cost
j1 i3 1000 2 800 0.6
j2 i5 3000 4 2600 0.7
j3 i7 5000 8 4000 0.9
j4 i9 10000 14 10000 0.05

Table 6
Mobile device applications workload analysis.
Applications workload Tw,i (MB) Communication cost N
E − Transport 40.2 4G:0.7$ 650
2D/3D Games 31.2 3G:2.5$ 790
Augmented Reality 25.1 4G:1.5$ 650
HD Video Streaming 52.8 5G:5$ 900
Healthcare Applications 16.5 5G:2.6$ 830

3.11. Time complexity of TSMCO

The TSMCO comprises various components such as Resource Matching, Task Sequences, and Task Scheduling. These
components are analyzed individually, and their combined use is optimized to determine the time complexity. Firstly,
Resource Matching involves utilizing the TOPSIS and AHP methods for task matching on heterogeneous servers. The time
complexity for Resource Matching is calculated as O(M × T ), where M is the MCC server resources for multi-criteria,
and T is the number of tasks arranged in pair-wise matching. Secondly, Task Sequencing involves sorting all tasks based
on shortest size, deadlines, and lateness, using the O(m log n) method where N is the number of sorted tasks and M is
the employed method for task sorting. Finally, Task Scheduling involves scheduling all MCC servers based on descending
order of δj and Cj . The time complexity for Task Scheduling is measured as O(log M). O(log M) + N for all MCC servers
according to their descending order of price and load. Here, N represents the task-swapping process in the time complexity
of different MCC servers.

4. Performance evaluation

We have generated practical results from various simulators for mobile and microservices-based applications to assess
the effectiveness of the TSMCO and MSCMCC frameworks. Table 4 describes the simulation parameters.
The process has been broken down into several parts based on the simulation parameters specified in Table 4. These
parts include the implementation of MSCMCC, calibration of metric parameters and components, comparison of the
TSMCO offloading framework, and the comparison of algorithms and task scheduling. Resource specifications for the MCC
servers are listed in Table 5, while Table 6 analyzes the workload for mobile applications.

4.1. Comparison framework and approaches

The results are compared to the existing approach based on primary considerations. Fig. 4 illustrates the implemen-
tations of MSCMCC via Mobile Application Services, Mobile Cloud Computing Agent Console, and Edge Mobile Server
Runtime Platform.
15
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 5. The time it takes for microservices to start up.

Fig. 6. A scheme for partially offloading 3000 tasks in the proposed work.

• We have developed a VM-based offloading framework for scheduling MCC tasks compared with existing frameworks
in [35–38]. We aim to offload the entire mobile application, including all microservices, to MCC cloud servers.
Baseline-1 compares the results of our proposed solution, MSCMCC and TSMCO, regarding microservices boot
time, task arrival time, CPU utilization, cost, deadline computations, and fault rate. We have implemented the
techniques in [39–42] with similar results, but our proposed technique improves these results through practical
simulation-based testing.
• Baseline-2/Hypothesis-2 involved utilizing virtual machines to develop a dynamic computational offloading-based
framework. This framework was compared to strategies detailed in [34,43,44] to evaluate and contrast their
performance. The study’s goal was to offload complete mobile applications to heterogeneous servers based on the
availability of resources. The obtained results were then compared to existing methodologies. Fig. 4 depicts the
MSCMCC framework and TSMCO framework and provides a comprehensive explanation of the proposed system.
Additionally, Figs. 5, 6, 7, 9, 10, 11, and 12 were used to support the implementation of the proposed system.
• Hypothesis-3/Baseline-3 involves comparing the cost and task deadline estimation with an additional baseline-3
chosen from the methodology implemented in [45]. The cost function is implemented using the CCCOF framework.
The findings reveal that the cost of the existing methodology is less than that of the cost model presented in Fig. 9.

16
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 7. Resource utilization of the CPU.

4.2. Performance metrics

This paper presents an implementation plan for experimental results recommended by the collaborative efforts of
various components. Table 4 outlines the application tasks generated from this collaborative process. The experiment
involves five distinct applications, each with its type considered for analysis. To ensure timely execution, every task has
a set deadline. The task deadlines in this experimental setup are determined using the following equation.
a ,i
Td = Pa,i + γ + Pa,i (34)

To determine the early completion time and the deadline for the final task interval, it is necessary to establish the
a,i
deadline for the task Td . The degree of stringency in the task deadline, denoted by γ , is indicated by five values: 0.2,
0.4, 0.6, 0.8, and 1, resulting in five distinct task deadlines (D1, D2, D3, D4, and D5) for each task on a mobile device.
The algorithm’s performance is evaluated using performance metrics verified through Eq. (31). The Relative Percentage
Division (RPD) statistical analysis method is employed to compute the performance division technique for MTOP, VFCN,
and CTOS. The paper’s findings effectively assess the energy consumption of different devices provide an algorithm for
evaluating computational throughput using efficient parameters. Eq. (34) defines the RPD estimation for the proposed
method.
Pa∗ + Pa
RPD (%) = × 100% (35)
Pa∗
where, Pa shows the objective function.

4.3. MSCMCC implementation

a,i
Defining the early finished time and task deadline is necessary for Td task. Our implementation of a Mobile Cloud-
based application on mobile devices involved using Android Studio, specifically the Huawei Y9 2019 Mobile Model
emulator, as shown in Fig. 5. We assessed the Edge X Foundry using an open-source platform. We built the MSCMCC
framework, comprising three main components: the Mobile Users layer, Mobile Cloud Agent Control layer, and Mobile
Cloud Resources Layer. Mobile applications send their related tasks to the Mobile Cloud Computing Agent Console using
REST API. The JSON format interprets requests and responses from the Mobile Cloud Computing Agent via a Gateway
Interface. The console interface processes the API request promptly and determines the type of services required to
execute the offloaded task based on its characteristics. The Mobile Cloud Monitoring System monitors the task list and
system stability. At the same time, Task sequencing orders the tasks logically, and Task Scheduler schedules tasks to
heterogeneous Mobile Cloud servers for execution. The Run Time is a complete system environment based on the system
scenario. The Java Runtime Virtual Machine (JVM) runs the Java program efficiently, like Windows Docker Virtual Machine.
Autonomous Microservices are created using containers registered with a Mobile Cloud server through registry services to
consume services efficiently. Using REST API for inter-services communication among microservices minimizes overhead.
17
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 8. The microservices overhead.

Fig. 9. The impact of Microservices on boot time, measured in milliseconds, and its correlation with the random arrival of 3000 tasks.

4.4. Comparison of offloading frameworks

The Microservices Container-Based Mobile Cloud Computing (MSCMCC) framework proposed in this study demon-
strates superior performance compared to existing techniques. The framework reduces bootup time and improves resource
utilization due to the utilization of lightweight VMs instead of heavyweight ones. The study shows through simulation
and experimental results that the MSCMCC technique effectively enhances the overhead of the service for the arrival of
tasks in percentage value. In addition, the proposed approach exhibits less time for microservices-based applications, less
CPU usage by around 22%, and less boot time than existing techniques by 17%. The results indicate that the MSCMCC
framework effectively improves overhead, bootup time, and resource utilization. The study finds that the lightweight
VM utilization during scheduling makes the proposed task scheduling framework efficient for delay-sensitive mobile
applications. Furthermore, the proposed system minimizes the cost of healthcare applications by 25%, augmented reality
by 23%, E-Transport tasks by 21%, and 3-D games tasks by 19%. On average, the boot time of microservices applications
is reduced by 17%, resource utilization by 36%, and task arrival time by 16%.

4.5. Task sequencing

Task sequencing rules, including FCFS, SJF, and SSJ, are utilized to arrange tasks in sequential order for scheduling. These
rules work in Fig. 8(a), while Fig. 8(b) shows the mean plot of alpha with a 95% HSD interval and the mean plot for random
18
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 10. A scheme that partially offloads processing resources is proposed to optimize CPU utilization when dealing with 3000 randomly arriving
tasks.

Fig. 11. The average plot for alpha along with a 95% HSD interval and the average plot for the arrival of random mobile tasks with a 95% HSD
interval.

mobile tasks arrival with 95% HSD intervals. Our proposed approach effectively compares the results among these rules
and extends the results for RPD value in percentage. To implement the results properly, our proposed technique uses SSJ,
which effectively improves the results compared to existing methods. The results demonstrate that our approach enhances
the proposed technique’s working with an effective scenario. Additionally, Fig. 8(b) illustrates that RPD’s significance for
SSJ is lower than SSTF and SPF for practical elaboration.

4.6. Task scheduling

The efficient management of costs in mobile cloud computing task scheduling is critical for mobile applications. This
study considers task deadlines and mobile application costs, specifically computation and communication costs, during
task scheduling using mobile cloud servers. This research aims to ensure timely task completion while minimizing mobile
application costs. Figs. 9 and 10 demonstrate the TSMCO framework that reduces mobile application execution costs. The
proposed scheme and scenario ensure mobile applications are within their deadlines. Fig. 10 displays the deadlines for
various tasks, including Healthcare applications, Augmented Reality Tasks, E-Transport Tasks, and 3D-Game Application
Tasks. These tasks have soft deadlines that allow for scheduling decisions and fault tolerance. If the initial deadline is
19
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 12. The RPD% is a factor to be considered in meeting the deadlines for Healthcare, Augmented Reality, E-Transport, and 3-D Game application.

Fig. 13. TSMCO was compared with three other techniques to evaluate the total cost incurred by healthcare applications (100 to 825 tasks),
Augmented reality (200 to 600 Tasks), E-Transport (100 to 640 Tasks), and 3D-Games (200 to 750 tasks) in terms of RPD%.

not met due to an MCC server’s unavailability, a secondary deadline is set for the task to be offloaded and processed.
The proposed system is evaluated by setting up five different deadlines for various tasks, and the iterative features
continuously improve the system’s performance until it reaches the optimal solution (see Figs. 13–15).

4.7. Task failure ratio

Our task scheduling mechanism demonstrates an improvement in the task failure ratio compared to existing techniques
that only utilize basic approaches for task scheduling. Figs. 11 and 12 illustrate the comparison of the task failure ratio
between our proposed and existing techniques, indicating that our approach outperforms existing techniques in the
dynamic allocation of Mobile Cloud Servers.
The TSMCO framework that has been proposed demonstrates efficient performance in dynamically allocating resources
to various situations. It has significantly improved the cost of the system while also providing efficient resource constraints
for meeting deadlines. Additionally, the framework effectively executes mobile tasks within the given deadlines and
expands the execution cost.

5. Conclusion and future work

In this study, we propose a novel task scheduling approach for a microservices-based mobile cloud computing
framework. Our proposed system, MSCMCC, utilizes microservices applications to handle delay-sensitive tasks and
includes a mobility-aware framework to reduce application costs. Additionally, we introduce the TSMCO framework,
which effectively manages task offloading through three steps: resource provision and matching, task sequencing, and
task offloading. Our experimental results demonstrate that MSCMCC and TSMCO effectively utilize the mobile device
and MCC server resources while reducing microservices-based application boot time and overhead time. Compared to
existing techniques, the proposed method achieves lower overhead time for microservices-based tasks and lower cost
for each application in baselines one and two. Moreover, our results highlight the efficient utilization of server resources
achieved through MSCMCC and TSMCO. This can reduce mobile server bootup time, minimize microservices latency, and
decrease server cost to reduce latency in mobile cloud servers.
20
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Fig. 14. The ratio of task failures during task scheduling.

Fig. 15. Mobile cloud servers with limited resources can fail in tasks.

We plan to implement a microservices-based task offloading framework for IoT and mobile applications that prioritizes
privacy. This framework will be designed to deploy on mainstream platforms such as Azure, Amazon, and Google and
will consider both security and transient failures. Additionally, we aim to incorporate machine learning or ANN for task
scheduling decision-making.

Funding statement

The authors are thankful to the Deanship of Scientific Research at Najran University, Saudi Arabia for funding this work
under the research code NU/DRP/SERC/12/1.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.
21
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

Data availability

The data that has been used is confidential.

Acknowledgment

The authors extend their appreciation to the Deanship of Scientific Research at Najran University for funding this work,
under the Distinguished Research funding Program grant code NU/DRP/SERC/12/1.

References

[1] A. Amini Motlagh, A. Movaghar, A.M. Rahmani, Task scheduling mechanisms in cloud computing: A systematic review, 33 (6), 2020, e4302.
[2] N. Parajuli, A. Alsadoon, P. Prasad, R.S. Ali, O.H.J.M.T. Alsadoon, Applications, A recent review and a taxonomy for multimedia application in
mobile cloud computing based energy efficient transmission, 79 (41), 2020, pp. 31567–31594.
[3] V.K. Kaliappan, S. Gnanamurthy, C.S. Kumar, R. Thangaraj, K.J.M.T.P. Mohanasundaram, Reduced power consumption by resource scheduling in
mobile cloud using optimized neural network, 2021.
[4] A. Aliyu, et al., Mobile cloud computing: taxonomy and challenges, 2020, 2020.
[5] S. Guo, B. Xiao, Y. Yang, Y. Yang, Energy-efficient dynamic offloading and resource scheduling in mobile cloud computing, in: IEEE INFOCOM
2016-the 35th Annual IEEE International Conference on Computer Communications, IEEE, 2016, pp. 1–9.
[6] X. Wang, K. Wang, S. Wu, S. Di, K. Yang, H. Jin, Dynamic resource scheduling in cloud radio access network with mobile cloud computing, in:
2016 IEEE/ACM 24th International Symposium on Quality of Service, IWQoS, IEEE, 2016, pp. 1–6.
[7] M. Chen, S. Guo, K. Liu, X. Liao, B. Xiao, Robust computation offloading and resource scheduling in cloudlet-based mobile cloud computing, 20
(5), 2020, pp. 2025–2040.
[8] X. Liu, Y. Li, H.-H. Chen, Wireless resource scheduling based on backoff for multiuser multiservice mobile cloud computing, 65 (11), 2016, pp.
9247–9259.
[9] P. Akki, V.J.W.P.C. Vijayarajan, Energy efficient resource scheduling using optimization based neural network in mobile cloud computing, 114
(2), 2020.
[10] J. Zare, S. Abolfazli, M. Shojafar, A. Kamsin, Resource scheduling in mobile cloud computing: taxonomy and open challenges, in: 2015 IEEE
International Conference on Data Science and Data Intensive Systems, IEEE, 2015, pp. 594–603.
[11] E. Ahmed, A. Gani, M. Sookhak, S.H. Ab Hamid, F. Xia, Application optimization in mobile cloud computing: Motivation, taxonomies, and open
challenges, 52, 2015, pp. 52–68.
[12] B.-G. Chun, S. Ihm, P. Maniatis, M. Naik, A. Patti, Clonecloud: elastic execution between mobile device and cloud, in: Proceedings of the Sixth
Conference on Computer Systems, 2011, pp. 301–314.
[13] E. Cuervo, et al., Maui: making smartphones last longer with code offload, in: Proceedings of the 8th International Conference on Mobile
Systems, Applications, and Services, 2010, pp. 49–62.
[14] S. Kosta, A. Aucinas, P. Hui, R. Mortier, X. Zhang, Thinkair: Dynamic resource allocation and parallel execution in the cloud for mobile code
offloading, in: 2012 Proceedings IEEE Infocom, IEEE, 2012, pp. 945–953.
[15] A.A. Heidari, S. Mirjalili, H. Faris, I. Aljarah, M. Mafarja, H.J. Chen, Harris hawks optimization: Algorithm and applications, 97, 2019, pp. 849–872.
[16] I. Attiya, M. Abd Elaziz, S.J.C. i. Xiong, neuroscience, Job scheduling in cloud computing using a modified harris hawks optimization and
simulated annealing algorithm, 2020, 2020.
[17] M. Satyanarayanan, P. Bahl, R. Caceres, N. Davies, The case for vm-based cloudlets in mobile computing, 8 (4), 2009, pp. 14–23.
[18] S. Goundar, A. Bhardwaj, S. Chand, Control, and communications, task offloading concept using cloud simulations in mobile computing, 12 (3),
2021, pp. 243–263.
[19] M.A. Sahito, A. Kehar, Dynamic content enabled microservice for business applications in distributed cloudlet cloud network, 9 (7), 2021.
[20] A. Lakhan, Q.-U.-A. Mastoi, M. Elhoseny, M.S. Memon, M.A. Mohammed, Deep neural network-based application partitioning and scheduling
for hospitals and medical enterprises using IoT assisted mobile fog cloud, 2021, pp. 1–23.
[21] X. Ma, A. Zhou, S. Zhang, Q. Li, A.X. Liu, S. Wang, Dynamic task scheduling in cloud-assisted mobile edge computing, 2021.
[22] A.M. Rahmani, et al., Towards data and computation offloading in mobile cloud computing: Taxonomy, overview, and future directions, 2021,
pp. 1–39.
[23] R.B. Mulinti, M.J. P.-t. P. N. Nagendra, Applications, An efficient latency aware resource provisioning in cloud assisted mobile edge framework,
14 (3), 2021, pp. 1044–1057.
[24] A. Lakhan, et al., Cost-efficient service selection and execution and blockchain-enabled serverless network for internet of medical things, 18
(6), 2021, pp. 7344–7362.
[25] A. Ali, et al., An efficient dynamic-decision based task scheduler for task offloading optimization and energy management in mobile cloud
computing, 21 (13), 2021, p. 4527.
[26] X. Ma, H. Xu, H. Gao, M. Bian, Real-time multiple-workflow scheduling in cloud environment, 2021.
[27] A. Lakhan, M.A. Dootio, T.M. Groenli, A.H. Sodhro, M.S.J.E. Khokhar, Multi-layer latency aware workload assignment of e-transport iot applications
in mobile sensors cloudlet cloud networks, 10 (14), 2021, p. 1719.
[28] N. Chaurasia, M. Kumar, R. Chaudhry, O.P. Verma, Comprehensive survey on energy-aware server consolidation techniques in cloud computing,
2021, pp. 1–56.
[29] L. Abualigah, A. Diabat, P. Sumari, A.H. Gandomi, Applications, deployments, and integration of internet of drones (IoD): a review, 2021.
[30] A. Samanta, Z.J. Chang, Adaptive service offloading for revenue maximization in mobile edge computing with delay-constraint, 6 (2), 2019, pp.
3864–3872.
[31] A. Samanta, J. Tang, Dyme: Dynamic microservice scheduling in edge computing enabled IoT, 7 (7), 2020, pp. 6164–6174.
[32] A. Samanta, F. Esposito, T.G. Nguyen, Fault-tolerant mechanism for edge-based IoT networks with demand uncertainty, 8 (23), 2021, pp.
16963–16971.
[33] C. Tang, M. Hao, X. Wei, W. Chen, Energy-aware task scheduling in mobile cloud computing, Distrib. Parallel Databases 36 (3) (2018) 529–553.
[34] F.J.D.S. Yang, Applications, Cloud computing virtual resource dynamic system allocation and application based on system architecture, 30 (5),
2021, pp. 753–770.
[35] A. Lakhan, D.K. Sajnani, M. Tahir, M. Aamir, R. Lodhi, Delay sensitive application partitioning and task scheduling in mobile edge cloud
prototyping, in: International Conference on 5G for Ubiquitous Connectivity, Springer, 2018, pp. 59–80.
[36] C. Tang, M. Hao, X. Wei, W.J.D. Chen, P. Databases, Energy-aware task scheduling in mobile cloud computing, 36 (3), 2018, pp. 529–553.

22
M.u. Hassan, A.A. Al-Awady, A. Ali et al. Pervasive and Mobile Computing 92 (2023) 101785

[37] X. Ma, S. Zhang, W. Li, P. Zhang, C. Lin, X. Shen, Cost-efficient workload scheduling in cloud assisted mobile edge computing, in: 2017 IEEE/ACM
25th International Symposium on Quality of Service, IWQoS, IEEE, 2017, pp. 1–10.
[38] S. Wu, C. Niu, J. Rao, H. Jin, X. Dai, Container-based cloud platform for mobile computation offloading, in: 2017 IEEE International Parallel and
Distributed Processing Symposium, IPDPS, IEEE, 2017, pp. 123–132.
[39] J. Lee, J. Gil, Adaptive fault-tolerant scheduling strategies for mobile cloud computing, J. Supercomput. 75 (8) (2019) 4472–4488.
[40] D.N. Raju, V. Saritha, Architecture for fault tolerance in mobile cloud computing using disease resistance approach, Int. J. Commun. Netw. Inf.
Secur. 8 (2) (2016) 112.
[41] S.K. Abd, S.A.R. Al-Haddad, F. Hashim, A.B. Abdullah, S. Yussof, Energy-aware fault tolerant task offloading of mobile cloud computing, in: 2017
5th IEEE International Conference on Mobile Cloud Computing, Services, and Engineering, MobileCloud, IEEE, 2017, pp. 161–164.
[42] J. Park, H. Yu, H. Kim, E. Lee, Dynamic group-based fault tolerance technique for reliable resource management in mobile cloud computing,
Concurr. Comput.: Pract. Exper. 28 (10) (2016) 2756–2769.
[43] R. Tiwari, R. Sille, N. Salankar, P. Singh, Utilization and energy consumption optimization for cloud computing environment, in: Cyber Security
and Digital Forensics, Springer, 2022, pp. 609–619.
[44] N.K. Mishra, P. Himthani, G.P. Dubey, Priority-based shortest job first broker policy for cloud computing environments, in: Proceedings of
International Conference on Communication and Computational Technologies, Springer, 2021, pp. 279–290.
[45] B. Keshanchi, A. Souri, N.J. Navimipour, An improved genetic algorithm for task scheduling in the cloud environments using the priority queues:
formal verification, simulation, and statistical testing, J. Syst. Softw. 124 (2017) 1–21.

23