Journal of Information & Optimization Sciences

ISSN 0252-2667 (Print), ISSN 2169-0103 (Online)

Vol. 44 (2023), No. 6, pp. 1023–1033
DOI : 10.47974/JIOS-1226

Task scheduling in cloud-fog computing using discrete binary particle

swarm meta-heuristic with modified sigmoid function

Y. Nasir Ahmed *
S. Pakkir Mohideen †
Department of Computer Applications
B. S. Abdur Rahman Crescent Institute of Science and Technology
Vandalur, Chennai
Tamil Nadu 600048
India

Mohammad Pasha §
Department of Information Technology
Muffakham Jah College of Engineering and Technology
Hyderabad
Telangana (TS) 500034
India

Abstract
Cloud-Fog IoT networking is a resourceful technology to aid the processing of IoT end
device requests. These devices generate tasks that need optimized computing and reduced
latency for applications that operate in real-time environments. This article sets forth a cloud-
fog task scheduler that schedules diverse tasks on vertically scaled Cloud and Fog virtual
machines. For the proposed Binary Particle Swarm Optimizer (BPSO) based scheduler, an
apt choice identified is to employ a modified sigmoid function with the logarithm decreasing
inertia weight policy to deliver an optimal scheduling scheme. Moreover, the parameters of
the BPSO are tuned inferring the best practices prescribed in literature. The results show that
proposed method caters better than existing heuristic techniques to improve makespan and
load imbalance.

Subject Classification: Primary 68T20, Secondary 68M20.

Keywords: Discrete binary particle swarm algorithm modeling, Meta-heuristic algorithms modified
sigmoid function, Logarithm decreasing inertia weight.

* E-mail: [email protected] (Corresponding Author)

†
E-mail: [email protected]
§
E-mail: [email protected]

©
1024 Y. N. AHMED, S. P. MOHIDEEN AND M. PASHA

1. Introduction
The type and amount of IoT tasks generated change dynamically and
the interactions are far from being predictable. Cloud services provided
users with virtual servers through virtual machines (VMs). Over the years
to cut costs, the demand for these costly energy-mongering virtual servers
has transformed into calls for cheap energy-saving, narrow function-based
modules or micro-services [1]. The idea of moving the micro-services to
the edge devices helps and complements the vision of green computing.
This further supplement the needs arising from real-time or near real-time
applications that have emerged due to the connectivity boom as part of a
growing need for a smart-eco system [2].
As the number of these diverse tasks (compute-intensive and low
latency) increase so will be the number of VM requests. Thus, vertical
scaling of VMs to handle these requests efficiently is needed. This forms
the problem of scheduling a large group of diverse tasks on an array of
heterogeneous VMs. Further the time taken to evaluate the best (task,
VM) pair upsurges rapidly. For such mapping problems, evolutionary
computational algorithms can attain optimal or near-optimal resolutions
in a viable time. Hence these are part of the selected choice for such
computationally hard (NP-hard) problems [3].
In this research, a meta-heuristic task scheduler namely Binary PSO-
based scheduler [4] with improved load balancing strategies (namely
IBPSO and MIBPSO) is proposed to optimize makespan and reduce load
imbalance. The experimented heuristics are implemented to evaluate the
task assignment and scheduling methods in cloud and fog computing
environments.
The subsequent sections of the present work are organized as follows:
in Section 2, a summary of related works in heuristic task scheduling
algorithms. Section 3 briefs the scheduling problem and its formulation
with respect to BPSO and improvised algorithms. Section 4 presents the
Simulation setup and various scenarios implemented for cloud and fog
environments. The penultimate Section 5 presents the analyses of results
and observations. To conclude, Section 6 covers-up the theme of the work.

2. Related Works
The time sensitive services deployed on the cloud computing requires
fastest turnaround time from the datacenter to IoT device and vice versa
and has been attained by the fog computing paradigm [1].The proposed
CLOUD-FOG COMPUTING USING WITH MODIFIED SIGMOID FUNCTION1025

fog-cloud system has enabled critical heart beat tasks medical applications
to collect and analyze so that overall cost may be reduced subsequently
[4]. Particle Swarm Optimization is projected as optimal solution towards
scheduling workflows for scientific and also the web applications [5].
Recommended multi-objective task-scheduling problem which optimizes
the two parameters in fog-cloud system first is the makespan and the
second is total costs [6]. Maximum usage of fog computing by subsequent
decrease in response time and also decreasing the cost towards cloud
compute resources is addressed [7].
Network delay, cost, processing time are offered in this paper to
address and meet the deadline of the deployed resources on fog-cloud
setting [8]. Classifier method is used to impart the workloads with respect
to power consumption and delays at the edge of the networks in fog-cloud
systems [9]. Task imbalance at fog nodes is addressed by implementing two
algorithms and colony and particle swarm optimization [10]. Effective and
managed energy consumption is attained by even distribution of tasks at
the nodes in fog-cloud system which gives reduced latency and increased
quality of service [11]. Usage of Harris Hawks’s optimization algorithm
enhanced the quality of service for Industrial IoT and fog computing
users through cloud computing [12]. Heuristic approach to place various
applications on the fog-cloud system to acquire the low latency and low
energy consumption. [13]. Heuristic task scheduling which is of three
staged is recommended and attained the best results towards performance
in defined service level agreements [14].
Computational nodes received many healthcare tasks to both cloud
and fog devices in order to reduce the schedule time in-turn usage of
task features, thus attaining maximum mobility [15]. Maximum profit is
offered at the integrated fog-cloud system by applying profit enhancement
algorithm to the smart systems [16]. A cost effective blockchain based fog
cloud is proposed to enhance the visibility of the patient’s health problems
and reduces their charges as the processing is done in the system [17].

3. Discrete Binary Particle Swarm Meta-Heuristic with Modified

Sigmoid Function and Logarithm Decreasing Inertia Weight.
Particle swarm meta-heuristic, i.e. PSO is a remarkable technique that
solves a numerically hard polynomial problem by intelligently repeating
steps to improvise a candidate solution, making it approach a well-defined
fitness measure. The Adapted form of this is what is dubbed as its Binary
counterpart, named as BPSO, in short. The BPSO provisions solutions for
1026 Y. N. AHMED, S. P. MOHIDEEN AND M. PASHA

practical discrete optimization problems [4]. The significant distinction

of the two forms is that the velocities and positions of the individual
solutions are characterized as far as the progressions of probabilities and
the particles are framed by zero or one. The velocity of the particle needs to
be constrained to the interval [0, 1). A calculated sigmoid transformation
can be utilized to restrict the velocity of a particle.

3.1 Task Scheduling in Cloud-Fog systems

The system consists of independent tasks (T) whose requested
computation size and data size is specified in million instructions and
bytes respectively. These tasks are to be mapped to Virtual Machines
(VMs) whose Processing power as well as bandwidth, specified in
million instructions per second (MIPS) and Megabytes per second (MBps)
respectively vary according to where they are located. The Task scheduler
first filters tasks based on suitability for execution on Cloud or Fog Nodes
as shown in Fig. 3.1.

Fig. 3.1
Architectural Elements of Cloud-Fog Computing

Table 1.1
Symbols used in the problem formulation and implementation.

R(Ti,VMj) Runtime for the task ti when executed on VMj

Tl Task length in million instructions (mi)
VMnpe is Number of processing elements in VM
VMmips Processing power of each core in VM
Tfs File-size to be sent for processing the task
Contd...
CLOUD-FOG COMPUTING USING WITH MODIFIED SIGMOID FUNCTION1027

VMbw Bandwidth of the VM

CT(VMi) Completion time of all tasks assigned to VMi
P {i, j} Particle position
V {i, j} New velocity of particle
Inertia weight calculated using linearly decreasing property,
w, wMax, wMin
Maximum and Minimum inertia weight resp.
iterMax Maximum number of iterations specified
iterCurr Current iteration
c1, c2 Cognitive/local coefficient and Social/global coefficient resp.
r1, r2 Random numbers in the interval [0, 1)
lbp{i, j} Local best position value of a particle in the current iteration
gbp{i, j} Global best position value of a particle across all iterations

Algorithm1: Binary PSO pseudocode

With respect to the BPSO algorithm given in Algorithm 1, whose

parameters are defined in Table 1.1, initially we calculate the running times
of all these tasks on all VMs. The resultant matrix of all tasks mapped
to all VMs is termed as the completion time matrix. The task completion
time is calculated using the details of each Task-VM pair. It is given by the
following equation
1028 Y. N. AHMED, S. P. MOHIDEEN AND M. PASHA

 Tl Tfs 
R(Ti , VMj )
=  +  Equation (1)
 (VMnpe∗ VMmips) VMbw 

Further the completion time of all tasks assigned to a VM can be

obtained using the following
CT (VMj ) R(Ti , VMj ) ∗ P{i , j}
= Equation (2)

P {i, j} denotes the particle position which is a binary value of either

1 (if selected task is assigned to the particular VM) or 0 (if selected task is
not assigned to the particular VM).
The maximum of the completion time on a VM gives the makespan
for executing all tasks assigned across all VMs. Thus, Makespan is given
by
Makespan(T ,VM )= Max(CT (VMi )) Equation (3)

The positions and velocities of all particles of PSO are initialized

by calling the CreateParticles () method (Algorithm 1 Line 4). Here the
position values are assigned a binary value of either 1, if the selected task
is assigned to the particular VM) or 0 otherwise. The velocity is a uniform
distributed random number in [0,1). These positions are updated to the
best positions by comparing the value obtained from the fitness function.
Next the particles positions and velocities are updated (Algorithm 1 Line
5) using the PSO equations as following
V {i , j}= ( ω ∗V {i − 1, j − 1}) + (c1∗ r1 ∗ (lbp{i , j} − p{i , j}))
+ (c 2 ∗ r 2 ∗ ( gbp{i , j} − p{i , j})) Equation ( 4)

ω = ω Max − ( ω Max− ωMin) ∗ (iterMax− iterCurr )/ iterMax)

Equation (5)

 1   1 
P{i , j} =
  0 ⇔
=  < r{i , j}
 (1+ exp(− V { i , j }))   (1+ exp(− V { i , j })) 
 1 
1 ⇔
=  ≥ r{i , j} Equation (6)
 (1+ exp(− V {i , j })) 

The update of local best particle position is compared for all particles
from the current iteration. As shown in Fig.3.2, we modified the sigmoid
function to reflect on the cognitive/local coefficient and social/global
coefficients. Following this, the global best particle positions are obtained
by comparison across all iterations. The VMs for which the position
values are set are selected for allocation of tasks. To enhance the solutions
CLOUD-FOG COMPUTING USING WITH MODIFIED SIGMOID FUNCTION1029

Fig. 3.2
Modified Sigmoid Functions used to reflect on the cognitive/local coefficient
and social/global coefficients of BPSO.

obtained by BPSO, we use a rebalancing method to identify the heaviest

loaded VMs in the solution set. We swap the assigned tasks from these
with the tasks from the lightest loaded VMs where the reassignment
outputs a reduced task completion time, hence improving the makespan.

4. Experimental Setup
The Experiments were carried on a PC with Intel Core i7-3770 CPU,
3.9 GHz, Memory 16 GB, Cloudsim 3.0 simulator on Windows OS.
The Virtual Machine configuration of Cloud (CVM) and Fog (FVM) nodes
is as follows
x1= CVM , 9k ≤ pp(x1) ≤ 10 k , 1k ≤ bw(x1) ≤ 2k ;
x 2= FVM , 1k ≤ pp(x 2) ≤ 2k , 9k ≤ bw(x 2) ≤ 10 k

with pp(x) is in MIPS, bw(x) is in MBps units. The Task configuration for
Cloud (CT, CW) and Fog (FT, FW) is
cb = CT , 9k ≤ ts(cb) ≤ 2k , 1k ≤ fs(cb) ≤ 2k ;
fb = FT , 1k ≤ ts( fb) ≤ 2k 9k ≤ fs( fb) ≤ 10 k ;
cw = CT , 9k ≤ ts(cw) ≤ 2k 9k ≤ fs(cw) ≤ 10 k ;
fw = FT , 9k ≤ ts( fw) ≤ 2k , 9k ≤ fs( fw) ≤ 10 k

with ts(x) defined in MI, fs(x) is in Bytes. Table 4.1 presents BPSO and
Modified sigmoid transformation parameters used in experiments.
1030 Y. N. AHMED, S. P. MOHIDEEN AND M. PASHA

Table 4.1
The BPSO parameters and their corresponding values

Algorithm Sigmoid Particles Rounds c1, c2 W

 1 
BPSO   10 100 1.49, 1.49 04-0.9
 (1+ exp(− V {i , j })) 

 1 
MBPSO-1   10 100 2.2, 2.2 -0.434
 (1+ exp(2.2073− V {i , j })) 

 1 
MBPSO-2  
 2.2073 + exp( 2.2073− V {i , j })  10 100 -2.2, 2.2 -0.434
( )

5. Experimental Evaluation and Results

The Task count is varied as 100, 500, 1000 and 2000 tasks. The VM
number is varied as 10, 25 and 50 VMs. The experiments to evaluate the
makespan and load imbalance were performed for (i) Min-Min heuristic,
(ii) BPSO (Binary PSO), (iii) MBPSO (Modified Binary PSO), (iv) IBPSO
(Improved Binary PSO) and (v) MIBPSO (Modified Improved Binary PSO)
Fig. 5.1 plots the makespan of one heuristic (Min-Min) and four
meta-heuristic algorithms namely BPSO, IBPSO, MBPSO, MIBPSO under

Fig. 5.1
Makespan comparison for implemented algorithms with varying Task and
VM Count in Cloud and Fog scenarios
CLOUD-FOG COMPUTING USING WITH MODIFIED SIGMOID FUNCTION1031

Fig. 5.2
Degree of Imbalance comparison for implemented algorithms with varying
Task and VM Count in Cloud and Fog scenarios

varying Task and VM count. Top row plots depict Cloud task scheduling
for 10, 25 and 50 VM count (let-center--right) whereas bottom row is for Fog
task scheduling similarly. With the increase in VM count, the makespan of
all scheduling algorithms decreases when load balancing is applied. The
reason is that, at VM count 10, the tasks starting from task 11 are mapped
to VMs that are already allocated. The number of free VMs increases for a
given set of tasks as the VM count increases. Moreover, it can be seen that
the MBPSO outperforms BPSO producing nearly 50% lesser makespan for
the same set of tasks and VMs. The improved versions IBPSO, MIBPSO
further improve makespan to 80% or more.
Fig. 5.2 plots the degree of load imbalance on VMs produced by
various algorithms in a Cloud and Fog scenarios. Top row plots depict
Cloud task scheduling for 10, 25 and 50 VMs respectively, whereas bottom
row is for Fog task scheduling. Load imbalance increases with the increase
in VM count. The degree of imbalance is reduced with IBPSO and MIBPSO
algorithms due to application of rebalancing the solution obtained by
BPSO and MBPSO algorithms.

6. Conclusion
This article presented implementation of Discrete Binary PSO and
Modified Discrete Binary PSO algorithms to address efficient task
1032 Y. N. AHMED, S. P. MOHIDEEN AND M. PASHA

scheduling in cloud and fog deployment scenarios. To enhance the

result further, the selected algorithms were load balanced. The relative
improvement of makespan achieved across all scenarios is in the range of
80% to 93% for VM count of 50. The proposed version of BPSO algorithm
can perform faster when compared to Original BPSO for similar settings in
all performed experiments. It demonstrates the effect of tuning the BPSO
parameters to reflect in customizing the sigmoid transformation. In the
future work other meta-heuristic algorithms will be tested to enhance
makespan, load balancing and other metrics.

References

[1] Aburukba, Raafat O., TahaLandolsi, and Dalia Omer, A heuristic

scheduling approach for fog-cloud computing environment with
stationary IoT devices, Journal of Network and Computer Application, 180
(2021): 102994.
[2] Kratzke, Nane. “A brief history of cloud application
architectures.” Applied Sciences 8.8 (2018): 1368.
[3] Mastoi, Qurat-ul-ain, et al, A novel cost-efficient framework for
critical heartbeat task scheduling using the Internet of medical things
in a fog cloud system, Sensors 20.2 (2020): 441.
[4] Poli, Riccardo, James Kennedy, and Tim Blackwell. “Particle swarm
optimization.” Swarm intelligence 1.1 (2007): 33-57.
[5] Farid, Mazen, et al, A survey on QoS requirements based on particle
swarm optimization scheduling techniques for workflow scheduling
in cloud computing, Symmetry 12.4 (2020): 551.
[6] Ali, Ismail M., et al, An automated task scheduling model using non-
dominated sorting genetic algorithm II for fog-cloud systems, IEEE
Transactions on Cloud Computing (2020).
[7] Mokni, Marwa, et al,Cooperative agents-based approach for
workflow scheduling on fog-cloud computing, Journal of Ambient
Intelligence and Humanized Computing (2021): 1-20.
[8] Naha, Ranesh Kumar, et al, Deadline-based dynamic
resource allocation and provisioning algorithms in fog-cloud
environment, Future Generation Computer Systems 104 (2020): 131-141.
[9] Abbasi, Mahdi, et al, Efficient resource management and workload
allocation in fog–cloud computing paradigm in IoT using learning
classifier systems, Computer communications 153 (2020): 217-228.
CLOUD-FOG COMPUTING USING WITH MODIFIED SIGMOID FUNCTION1033

[10] Hussein, Mohamed K., and Mohamed H. Mousa, Efficient task

offloading for IoT-based applications in fog computing using ant
colony optimization, IEEE Access 8 (2020): 37191-37201.
[11] Kaur, Mandeep, and Rajni Aron, Energy-aware load balancing in fog
cloud computing, Materials Today: Proceedings, (2020).
[12] Abdel-Basset, Mohamed, et al, Energy-aware metaheuristic algorithm
for industrial-Internet-of-Things task scheduling problems in fog
computing applications,IEEE Internet of Things Journal, 8.16 (2020):
12638-12649.
[13] Natesha, B. V., and Ram Mohana Reddy Guddeti, Heuristic-based
IoT application modules placement in the fog-cloud computing
environment, 2018 IEEE/ACM international conference on utility and
cloud computing companion (UCC Companion), IEEE, 2018.
[14] Sang, Yongxuan, et al, A three-stage heuristic task scheduling for
optimizing the service level agreement satisfaction in device-edge-
cloud cooperative computing, PeerJ Computer Science, 8 (2022): e851.
[15] Abdelmoneem, Randa M., Abderrahim Benslimane, and
EmanShaaban, Mobility-aware task scheduling in cloud-Fog IoT-
based healthcare architectures, Computer Networks, 179 (2020): 107348.
[16] Mahmud, Redowan, et al, Profit-aware application placement for
integrated fog–cloud computing environments, Journal of Parallel and
Distributed Computing, 135 (2020): 177-190.
[17] Lakhan, Abdullah, et al, Smart-contract aware ethereum and client-
fog-cloud healthcare system, Sensors, 21.12 (2021): 4093.