Available online at www.sciencedirect.

com

Available online at www.sciencedirect.com

Available online at www.sciencedirect.com

ScienceDirect
Procedia Computer Science 00 (2023) 000–000
www.elsevier.com/locate/procedia
Procedia
Procedia Computer
Computer Science
Science 00 (2023)
218 (2023) 000–000
1936–1945
www.elsevier.com/locate/procedia
International Conference on Machine Learning and Data Engineering
International Conference on Machine Learning and Data Engineering
Efficient Workflow Scheduling algorithm in cloud computing using
Efficient Workflow Scheduling algorithm in cloud computing using
Whale Optimization
Whale Optimization
Sudheer Mangalampallia , Ganesh Reddy Karria , G Naga Satishb
a VIT-AP University,School ofaComputer Science and Engineering,aAmaravati, India
Sudheer Mangalampalli
b
, Ganesh Reddy Karri , G Naga Satishb
BVRIT Hyderabad College of Engineering For Women, Telangana, India
a VIT-APUniversity,School of Computer Science and Engineering, Amaravati, India
b BVRITHyderabad College of Engineering For Women, Telangana, India

Abstract
Task Scheduling is a tremendous challenge in cloud computing as tasks are varied in terms of their processing capacity and interde-
Abstract
pendent on each other i.e. workflows. Therefore, it is difficult to map workflows to suitable virtual resources in cloud environment.
Task
ManyScheduling is a tremendous
authors developed challenge inwhich
various schedulers, cloudaddresses
computing as tasks are
makespan andvaried in of
Quality terms of their
service processing
metrics. Authorscapacity and
have not interde-
addressed
pendent on each other i.e. workflows. Therefore, it is difficult to map workflows to suitable virtual resources
parameters makespan in combination with migration time, which also affects performance of cloud computing environment. In thisin cloud environment.
Many authors
manuscript, developed
a new various schedulers,
workflow-scheduling which addresses
mechanism is proposed,makespan and Quality
which takes of of
priorities service
tasks metrics. Authors
and schedules have
tasks not addressed
effectively on to
parameters
correspondingmakespan
virtual in combination
resources. Whalewith migration time,
optimization whichasalso
algorithm the affects performance
methodology of cloud
to model computing environment.
this algorithm.Extensive In this
simulations
manuscript,
carried out ona new workflow-scheduling
workflowsim simulator. Itmechanism
was evaluatedis proposed, which takes
aganist existing PSO,priorities
CS, ACO, of tasks and schedules
GA algorithms. tasks from
Finally, effectively on to
simulation
corresponding virtual resources. Whale optimization algorithm as the methodology to model this
results, it was identified that makespan, migration time and energy consumption were minimized to a good extent. algorithm.Extensive simulations
carried out on workflowsim simulator. It was evaluated aganist existing PSO, CS, ACO, GA algorithms. Finally, from simulation
results, it was identified that makespan, migration time and energy consumption were minimized to a good extent.
© 2023 The Authors. Published by Elsevier B.V.
This is an
© 2023 open
The accessPublished
Authors. article under the CC BY-NC-ND
by Elsevier B.V. license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
© 2023 The under
Peer-review Authors. Published by
responsibility of Elsevier
the B.V. committee of the International Conference on Machine Learning and Data Engi-
This is an open access article under the scientific
CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
neering.
Peer-review under responsibility of the scientific committee of the International Conference on Machine Learning and Data Engineering
Peer-review under responsibility of the scientific committee of the International Conference on Machine Learning and Data Engi-
neering.
Keywords: workflow Scheduling; makespan; migration time; Energy Consumption; Whale Optimization

Keywords: workflow Scheduling; makespan; migration time; Energy Consumption; Whale Optimization

1. Introduction
1. Introduction
Cloud computing emerged as one of the new computing model, which gives seamless services of compute, storage
and network to users as a service. It gives a high flexibility to customers to choose specific services on demand like
Cloud computing emerged as one of the new computing model, which gives seamless services of compute, storage
a subscription model. With the advent of cloud paradigm, many of industries migrated to cloud environment, which
and network to users as a service. It gives a high flexibility to customers to choose specific services on demand like
gives following advantages many of users [1].
a subscription model. With the advent of cloud paradigm, many of industries migrated to cloud environment, which
gives following advantages many of users [1].
1. On demand services
1. On demand services
∗ Sudheer Mangalampalli.
E-mail address: [email protected]
∗ Sudheer Mangalampalli.
E-mail address: [email protected]

1877-0509 © 2023 The Authors. Published by Elsevier B.V.

This is an open
1877-0509 access
© 2023 Thearticle under
Authors. the CC BY-NC-ND
Published by Elsevier license
B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/)
1877-0509
This is an ©
Peer-review 2023access
under
open The Authors.
responsibility Published
article of the by
CCElsevier
the scientific
under B.V.of license
committee
BY-NC-ND the International Conference on Machine Learning and Data Engineering.
(https://creativecommons.org/licenses/by-nc-nd/4.0)
This is an open under
Peer-review access article under the CC
responsibility of BY-NC-ND license
the scientific (http://creativecommons.org/licenses/by-nc-nd/4.0/)
committee of the International Conference on Machine Learning and
Peer-review under responsibility of the scientific committee of the International Conference on Machine Learning and Data Engineering.
Data Engineering
10.1016/j.procs.2023.01.170
 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945 1937
2 Author name / Procedia Computer Science 00 (2023) 000–000

2. Reduced efforts of network administrative team

3. Scalability
4. No upfront investment
5. Choosing only specific services rather than all services
6. Flexibility

The above are some of the advantages to the user of cloud paradigm but not limited to the mentioned above. Besides all
the above advantages, there are some limitations also there in cloud computing paradigm i.e. lack of expertise about
services in cloud platforms, Security, Vendor lock issues, Cost management of services on cloud platform but still
cloud computing paradigm is advantageous to different users in diversified domains and applications of cloud com-
puting paradigm also significantly improved over different areas. It can be used in various domains such as Health,
education, government Sectors, entertainment etc. This model can be applied in various domains and to provide
services seamlessly to users is a challenge as each of domain need different kind of services. Services provided by
cloud provider to users depends on their requirement as some companies need Storage or compute or network services
depends on their requirement. To provide seamless virtual services to users depend on their requirement is a huge chal-
lenge as every cloud provider will have millions and billions of customers around the world. In order to provide wide
variety of services to users and to handle interdependent tasks i.e. workflows, an effective scheduler is needed which
maps these workflows to appropriate virtual resources. Therefore, task scheduling in particular workflow scheduling
is tremendous challenge in cloud computing. Many of earlier authors developed various scheduling algorithms, which
addresses parameters such as makespan, resource utilization and Quality of service. They have used different bio and
nature inspired algorithms[2, 3, 4, 5, 6] to develop scheduling algorithms, but there is a chance to reduce makespan,
migration time, energy consumption. In this paper, a new workflow-scheduling model is proposed, which considers
task priorities based on size, processing capacity. whale optimization is used to model the scheduler and it addresses
parameters makespan, migration time,energy consumption. Main Contributions of authors in this paper are presented
below.

1. Workflow scheduling mechanism developed by considering task size and processing capacity.
2. Scheduling problem solved by Whale optimization algorithm.
3. Workload given to algorithm is generated randomly from the simulator.
4. Workflowsim was used for simulating this scheduling model.
5. Finally, parameters addressed i.e. makespan, migration time and Energy consumption while effectively schedul-
ing interdependent tasks i.e. workflows.

Remaining structure of manuscript is organized as follows. Section 2 is about Related Works, Section 3 discusses
proposed system architecture, Section 4 discusses proposed Workflow Scheduling algorithm using whale optimization,
Section 5 about Simulation, discussion of results, Section 6 about Conclusion and Future work.

2. Related Works

In [2], a scheduling mechanism proposed which considers both balanced and imbalanced workflows. It addresses
parameters named as data transfer and communication costs. DiCSPM methodology was used to solve scheduling
problem. It compared with existing GWO, BRS, PSO algorithms. From simulation results, it was shown improve-
ment over existing algorithms in data transfer, communication costs. In [3], proposed a scheduling approach which
addresses makespan. This scheduling model developed based on calculation of priorities for interdependent tasks.
A hybridized approach i.e. HybridCSPSO was used to model the algorithm. Workflowsim was used as simulator
for experimentation. It was evaluated aganist PSO, CS algorithms. From simulation results, authors identified that
hybridized workflow scheduling model minimizes energy consumption, makespan while effectively scheduling work-
flows to appropriate virtual resources. In [4], an effective workflow scheduling mechanism was developed which
addresses makespan, resource utilization. To effectively map incoming workflows onto virtual resources, authors
carefully calculated priorities of tasks and thereby scheduled tasks. This mechanism was modeled by using cat swarm
optimization. Workflowsim was used as a simulator for this approach. It was evaluated aganist baseline approaches
1938 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945
Author name / Procedia Computer Science 00 (2023) 000–000 3

PSO, CS algorithms. From extensive simulations, it was revealed that makespan, energy consumption was mini-
mized over baseline approaches. In [5], a scheduling mechanism was developed which addresses parameters such
as makespan, total execution cost while balancing of tasks for heterogeneous systems. Scheduling model was devel-
oped using a hybridized approach known as GA-PSO. Random load used for experimentation. It was compared over
baseline algorithms. Results revealed that makespan is greatly minimized with 16% over GA and 4% over PSO. Ex-
ecution cost is minimized with 13% over GA and 4% over PSO. In [6], authors proposed a grid workflow scheduling
mechanism which addresses QoS parameters with large scale workflows. Methodology used in this approach is ANT
colony optimization. Gridsim was used as simulator for experimentation. For simulation 120 realtime workflows were
considered and evaluated aganist existing algorithms such as PSO, GA. Finally from simulation results QoS param-
eters were greatly improved over existing approaches. In [7], a many objective scheduling strategy was proposed by
authors which was developed based on PSO algorithm. It mainly addresses parameters makespan, reliability, cost and
energy consumption. It was simulated on MATLAB. This many objective approach was evaluated aganist variations
of PSO algorithms. From simulation results, it was observed that it greatly minimizes above parameters with proposed
approach. In [8], a scheduling model was developed which addresses makespan, cost, energy consumption, degree of
imbalance, reliability and utilization of virtual resources. This scheduling model was developed with neural network
combined with NSGA-II. Realtime workflows are used from pegasus workflow management system[9] This proposed
model was evaluated aganist baseline models such as PSO, CS. From simulation results, it was revealed that proposed
model shown impact over existing models for above parameters. In[10], a scheduling framework developed based on
security requirements and task completion time. It was simulated on Cloudsim. The proposed approach was evaluated
over existing scheduling algorithms HEFT and MHEFT. Extensive simulations were performed with scientific work-
loads and from the results it is revealed that proposed scheduling algorithm improves security and concurrent running
of tasks. In [11], a multi objective workflow scheduling algorithm was designed by combining GSA and HEFT algo-
rithms which addresses makespan and cost. It was simulated on Matlab and proposed approach was evaluated over
ANOVA test. Results revealed that proposed approach shown a huge impact over existing mechanisms. In [12], Multi
objective scheduling approach was formulated to address makespan and cost. Methodology used to design scheduling
model is HEFT-ACO. Workflowsim was used for entire simulation. Amazon instances and realtime workflows are
used for simulation. It was evaluated aganist existing baseline algorithms HEFT, ACO. From simulation results, it
was shown that proposed approach was outperformed over baseline approaches for metrics i.e. makespan and cost.
In [13], a workflow scheduling mechanism which addresses time and cost as parameters in cloud computing. Hyper
heuristic approach with time driven mechanism is used to solve scheduling problem. An explicit customized testbed
was used to conduct experiments and a total of 9 scenarios were designed and all were evaluated aganist baseline
models such as GA, PSO, HHSA. Realtime Scientific workflows were taken in this mechanism. From the conducted
experiments, it was revealed that this CTDHH approach greatly minimizes completion time and cost of resources used
in cloud.In[14], a deadline constrained scheduling algorithm was proposed by using a modified GA algorithm by tun-
ing genetic operators. Fitness function was explicitly written as an adaptive function to map with dynamic workloads
in cloud computing. Simulation was done on Cloudsim. Realtime Scientific workflows and Synthetic workflows were
considered in this approach. Finally, it was evaluated over baseline approaches and then it shows a huge impact over
existing approaches for metrics makespan, cost of resources.
 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945 1939
4 Author name / Procedia Computer Science 00 (2023) 000–000

Table 1. Analysis of Existing works done by various authors.

Authors Methodology Parameters

[2] DiCSPM data transfer, communication costs

[3] HybridCSPSO Energy Consumption, makespan
[4] CS makespan, Energy Consumption
[5] GA-PSO makespan, Execution Cost
[6] ACO QOS
[7] PSO makespan, energy consumption, reliability, cost
[8] Neural Network makespan, cost, energy consumption, degree of imbalance, reliability
[10] MHEFT Time for Task Completion
[11] GSA and HEFT makespan
[12] HEFT-ACO makespan
[13] CTDHH cost of resources
[14] modified GA algorithm cost of resources

From the above literature, inferences that can be taken out that many of existing authors proposed scheduling
algorithms and addressed parameters i.e. makespan, energy consumption, resource utilization but migration time can
be addressed as a parameter which impacts makespan indirectly while minimizing makespan and energy consumption.

3. Problem Definition and Proposed System Architecture

Assume a DAG (Direct-Acyclic-Graph) for workflow i.e. indicated as w = (V, C) V represents set of vertices
which are considered as tasks which indicates tk = {t1 , t2 , t3 , ...tk }, C represents dependencies between various tasks as
considered for workflows. For any scheduling process virtual resources are needed vn = {v1 , v2 , v3 .....vn }. These virtual
resources will sitting in physical hosts. These are indicated as phi = {ph1 , ph2 , ph3 ....phi } and finally all resources
are running in datacenters named as DC j = {DC1 , DC2 , DC3 ....DC j }. Problem is formulated in such a way that when
tk tasks are incoming onto cloud thereby scheduler need to consider priority of task and schedule tasks to vn virtual
resources which are running in phi physical hosts which inturn they are placed in DC j datacenters which should
minimize makespan, migration time and energy consumption. In the above figure 1, workflow used in simulation

Fig. 1. Workflow of simulation represented as DAG

is represented. In the figure 1, tk represents tasks in workflow i.e. interdependent tasks. dm,n represents flow of data
in between tasks. Figure 2 represents proposed system architecture in which initially all interdependent tasks i.e.
workflows are comes onto cloud interface where task manager calculates incoming tasks prioroties based on type
of task, size of task and processing capacity. Then, afterwards these tasks are suitably mapped to appropriate VMs
which are sitting in physical hosts which inturn resided in corresponding datacenters. Initially all tasks are taken by
broker on behalf of task manager which takes all tasks and calculates priorities of interdependent tasks and then fed to
scheduler which need to map all tasks to appropriate VMs while tracking resources i.e. virtual resources in resource
1940 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945
Author name / Procedia Computer Science 00 (2023) 000–000 5

manager. Scheduler immediately send tasks which are coming onto it onto execution queue and based on priorities
of tasks and availability of virtual resources scheduler will map tasks to appropriate VMs.In the figure 2, there is a
component named as Resource manager which tracks of all resources and how many of them are allocated, utilized
and not utilized. In this paper, authors carefully presented a mathematical modeling which gives overall scenario about
how algorithm is designed and proved mathematically.

Fig. 2. Proposed System Architecture

3.1. Mathematical Modeling

In the above figure2, after giving interdependent tasks as input to cloud console, priority of tasks are calculated
as our main aim is suitable mapping of tasks and VMs.Therefore, priority of incoming tasks are calculated here.
Basically, It depends on size of task, run time capacity of a task i.e. processing capacity as some of the tasks varied by
their size at runtime. It is indicated below using equation 1.

t pr = tksize /vnproc (1)

where t pr indicates priorities of incoming tasks, tksize indicates length or size of task, vnproc indicates a capacity of a VM.

3.1.1. makespan model

After calculating task priorities, parameters makespan, migration time, energy consumption are modeled math-
ematically in the below equations respectively. In the following equation 2, makespan is evaluated as it is one of
important parameter for scheduling of tasks and how much time a task is executing on a VM. It is evaluated as follows
in the below equation.

mak = avan + ek (2)

where, mak is makespan for k tasks, avan indicates availabality of n virtual machines and ek indicates execution time
of k tasks.

3.1.2. migration time model

After evaluation of makespan, migration time is calculated in this modeling as it impacts scheduling of tasks
indirectly. Whenever migration time increases, i.e. if a VM is overloaded with tasks and migrates a task to another
VM and at that point of time, if it is taking more time then it can also impacts makespan of task indirectly. In below
equation3, migration time is calculated.

migT = tkmove /T (3)

 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945 1941
6 Author name / Procedia Computer Science 00 (2023) 000–000

3.1.3. energy Consumption model

In this subsection, energy consumption is evaluated as if energy is minimized more at vendor side, then cloud user
will be benifited for paying of lesser price for all services they consume. Therefore, energy consumption evaluated by
using below equation4.

enconsumption = enconsumption (vn ) (4)

3.1.4. fitness function

In this modeling, a fitness function is used to optimize scheduling parameters which addresses multiple parameters
i.e. a multiobjective optimization.

f (x) = min mak (x), migT (x), enconsumption (x) (5)
x

3.2. Whale optimization algorithm

This subsection, discusses about whale optimization algorithm[15]. This is inspired from blue humpback
whales[15]. Initially whale population generated randomly and looks for prey around them. It initializes population of
agents and it is represented as qr and from those population of agents best agent is to be identified and it is represented
as q∗ . Fitness function evaluated by using equation5. After evaluating it, whale identifies prey then surrounds it. When
a prey is avaiable, whales will assumes that it is current best solution and updates best agent by following equation.

→
−s =| →
− −−−→ −−−→
m.q∗ (x) − qr (x) | (6)
−−−→
where x represnts present iteration of search,→ −
m represents vector coefficient ,qr (x) represents search agent position
−−−→
and q∗ (x) indicates search agent’s best position. Then next best search agent evaluated using below equation7.
−−−−−−→ −−∗−→ →
q(x + 1) = q (x) − −z .→
−s (7)

where →
−z indicates vector coefficient and both of vector coefficients i.e. →
−z and →
−
m are evaluated as follows.
→ −a .→
−z = 2→ − →
b − −a (8)

→
− →−
m = 2. b (9)

where →−a , →
−
b are random vectors and their values ranges from 2 to 0 and 0 to 1 respectively. After calculation of vector
coefficients and identification of search agents whales look for next updation and this can be done in two phases either
shrinking or Spiral updation. In Shrinking phase, →−z is set between range of -1 and 1.
In spiral updation phase agent position can be calculated using following equation10.
−−−−−−→ → −−−→
q(x + 1) = −
s f .hur .cos(2πr) + q∗ (x) (10)

→
− −−−→ −−→
s f =| q∗ (x) − q( x) | (11)

−−−→ −−→
where q∗ (x) indicates search agent’s best position and q( x) represents agent’s current iteration. Then if Q < 0.5 it uses
shrinking process and if Q >= 0.5, then it uses spiral process. In this Q is a random number in range of -1 and 1. After
exploitation process, exploration process will starts and it is evaluated by using following equations.

→
−s =| →
− −−−→ −
m.q∗rand − →
q | (12)

−−−−−−→ −−∗−−−→ →
q(x + 1) = qrand(x) − −z .→
−s (13)
1942 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945
Author name / Procedia Computer Science 00 (2023) 000–000 7

4. Proposed workflow Scheduling algorithm

This section presents proposed whale scheduler.

Algorithm 1 Proposed Whale Scheduler

Input: Workflow designed by using DAG w = (V, C), tk = {t1 , t2 , t3 , ...tk }, vn = {v1 , v2 , v3 .....vn }, phi =
{ph1 , ph2 , ph3 ....phi } and DC j = {DC1 , DC2 , DC3 ....DC j }
Output: Accurate mapping of workflows to VMs while minimizing mak (x), migT (x), enconsumption (x)
Start
Generate random population
Calculation of priorities of tasks by using eqn1
Calculation of fitness function by using equation5
Update z,m,r,Q
if (Q < 0.5) then
if (| z |≤ 1) then
Updation of Search agent’s position using eqn 7.
else
Updation of agent randomly using eqn 13.
end if
end if
if (Q ≥ 0.5) then
Search agent position can be updated using equn10.
end if
Calculate makespan, migration time and energy consumption using equations 2, 3 and 4 respectively.
if (search agent is outside region) then
Update → −q ∗
Y= Y+1
end if
End

Flow of algorithm is explained here. Initially, generated whale population randomly and used search agents to
identify prey. In this algorithm tasks are considered as whales and prey is considered as virtual reosurces in the cloud.
After the first step, calculation priorities of tasks using eqn1 and then evaluated fitness function using equation 5. To
search for prey search agents are used and applied a condition in such a way that if Q < 0.5 then update positions of
search agents using eqn7 else update search agent’s position using eqn13. If Q ≥ 0.5 then update positions of search
agents using eqn10. After updation of search agents then evaluate makespan, migration time and energy consumption
using eqn 2, 3 and 4 respectively. Identify the solutions i.e. values achieved through these parameters if current
solution is the best solution, then state that it is a best solution. Otherwise update →
−q ∗ . This process continued till the
best solution is identified.

5. Simulation and Results

This section clearly discusses about Simulations and Results. Entire simulation performed in workflowsim [16].
Simulation settings were considered from [17]. Input to workflowsim was generated in a random manner. The param-
eters addressed in this approach are makespan, migration time and Energy consumption.
 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945 1943
8 Author name / Procedia Computer Science 00 (2023) 000–000

Table 2. Configuration Settings for Simulation.

Name Quantity

RAM 16GB
No. of Tasks 100-1000
Hard disk 256GB
Physical Host Processor i5
Length of Tasks 800,000
Hypervisor Xen
Operating System Linux
No. of Data Centers 2

5.1. Calculation of makespan

makespan is important parameter for cloud computing paradigm because if execution time is more for a correspond-
ing task on VM, then it impacts a lot on efficacy of a scheduler. makespan is evaluated aganist existing algorithms
PSO, ACO, CS and GA. In the simulation, when calculating makespan it ran for different tasks such as 100, 500 and

Fig. 3. Calculation of makespan using proposed whale scheduler

1000. It was evaluated aganist ACO, PSO, CS , GA and proposed whale algorithm. For 100, 500 and 1000 tasks with
ACO algorithm makespan generated is 1267.8,2578.8,3345.7 respectively. For 100, 500 and 1000 tasks with PSO
algorithm makespan generated is 1167.8, 2267.9, 3246.8 respectively.For 100, 500 and 1000 tasks with CS algorithm
makespan generated is 998.46, 1687.9, 1965.9 respectively. For 100, 500 and 1000 tasks with GA algorithm makespan
generated is 834.7, 1267.6, 1456.9 respectively.For 100, 500 and 1000 tasks with proposed whale scheduler algorithm
makespan generated is 756.9,1245.8,1089.9 respectively. From above figure3, it was clearly identified that proposed
whale scheduler minimizes makespan with baseline existing algorithms.

5.2. Calculation of migration time

migration time is aslo an important in cloud computing paradigm as it indirectly impacts makespan so that it is
also one of important parameter to be considered for cloud paradigm. Migration time is evaluated aganist existing
algorithms PSO, ACO, CS, GA. In the simulation, when calculating migration time it ran for different tasks such as
100, 500 and 1000. It was evaluated aganist ACO, PSO, CS, GA and proposed whale algorithm. For 100, 500 and
1000 tasks with ACO algorithm migration time generated is 867.9,945.9,1214.5 respectively. For 100, 500 and 1000
tasks with PSO algorithm migration time generated is 641.2, 823.9, 1098.9 respectively.For 100, 500 and 1000 tasks
with CS algorithm migration time generated is 598.9, 769.5, 1034.8 respectively.For 100, 500 and 1000 tasks with CS
1944 Sudheer
AuthorMangalampalli
name / ProcediaetComputer
al. / Procedia Computer
Science Science
00 (2023) 218 (2023) 1936–1945
000–000 9

Fig. 4. Calculation of migration time using proposed whale scheduler

algorithm migration time generated is 553.5, 699.7, 1001.8 respectively. For 100, 500 and 1000 tasks with proposed
whale scheduler algorithm migration time generated is 523.9,609.9,913.8 respectively. From above figure4, it was
clearly identified that proposed whale scheduler minimizes migration time with baseline existing algorithms.

5.3. Calculation of Energy Consumption

Fig. 5. Calculation of Energy Consumption using proposed whale scheduler

Energy consumption is one of parameter as it plays an important role in cloud paradigm as it effects cloud provider
and user. If more energy is consumed, user needs to pay more price for the services consumed in cloud paradigm.
Energy consumption is evaluated aganist existing algorithms PSO, ACO, CS and GA.
In the simulation, when calculating Energy consumption, it ran for different tasks such as 100, 500 and 1000. It was
evaluated aganist ACO, PSO, CS, GA and proposed whale algorithm. For 100, 500 and 1000 tasks with ACO algorithm
energy consumed is 11.7,10.8,15.8 respectively. For 100, 500 and 1000 tasks with PSO algorithm energy consumed
is 10.2, 13.1, 16.9 respectively.For 100, 500 and 1000 tasks with CS algorithm energy consumed is 9.12, 11.45, 14.98
respectively. For 100, 500 and 1000 tasks with GA algorithm energy consumed is 7.24,10.92, 13.26 respectivelyFor
100, 500 and 1000 tasks with proposed whale scheduler algorithm energy consumed is 6.25,10.02,11.13 respectively.
From figure5, it was clearly identified that proposed whale scheduler minimizes energy consumption with baseline
existing algorithms.
 Sudheer Mangalampalli et al. / Procedia Computer Science 218 (2023) 1936–1945 1945
10 Author name / Procedia Computer Science 00 (2023) 000–000

6. Conclusion and Future work

Workflow scheduling is challenging job in cloud model as incoming workflows varies with respect to their size and
processing capacities and moreover that these workflows are interdependent tasks. In order to schedule these variable
and dynamic workflows in cloud computing, there is an immense need of an effective scheduler. Many of existing au-
thors proposed various scheduling algorithms addresses parameters makespan, resource utilization. In this manuscript,
a new scheduling mechanism i.e. pwhale scheduler is proposed which takes incoming priorities of workflows and
carefully schedules on to appropriate virtual resources. whale optimization algorithm used for solving scheduling
problem. Proposed algorithm implemented on workflowsim and finally evaluated efficacy our algorithm aganist ACO,
PSO, CS algorithms. From Simulation results, it was identified that proposed scheduling algorithm greatly minimizes
makespan, migration time, energy consumption over existing approaches.There is a limitation identified in this work
i.e. it is not predicting number of upcoming tasks and type of tasks. Therefore, there is a need of ML model to pre-
dict upcoming workflows onto cloud console. In future, our idea is to employ a ML approach to address scheduling
problem in cloud computing and deploy it in a realtime cloud environment.

References

[1] Sadiku MN, Musa SM, Momoh OD. Cloud computing: opportunities and challenges. IEEE potentials. 2014 Jan 9;33(1):34-6.
[2] Alawad NA, Abed-alguni BH. Discrete island-based cuckoo search with highly disruptive polynomial mutation and opposition-based learning
strategy for scheduling of workflow applications in cloud environments. Arabian Journal for Science and Engineering. 2021 Apr;46(4):3213-
33.
[3] Mangalampalli S, Pokkuluri KS, Kocherla R, Rapaka A, Kota NR. An Efficient Workflow Scheduling Algorithm in Cloud Computing Using
Cuckoo Search and PSO Algorithms. InInnovations in Computer Science and Engineering 2022 (pp. 137-145). Springer, Singapore.
[4] Mangalampalli S, Pokkuluri KS, Satish GN, Kumar KV. An Effective Workflow Scheduling Algorithm in Cloud Computing Using Cat Swarm
Optimization. ECS Transactions. 2022 Apr 24;107(1):2523.
[5] Manasrah AM, Ba Ali H. Workflow scheduling using hybrid GA-PSO algorithm in cloud computing. Wireless Communications and Mobile
Computing. 2018 Jan 1;2018.
[6] Chen WN, Zhang J. An ant colony optimization approach to a grid workflow scheduling problem with various QoS requirements. IEEE
Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews). 2008 Oct 31;39(1):29-43.
[7] Saeedi, Sahar, et al. ”Improved many-objective particle swarm optimization algorithm for scientific workflow scheduling in cloud computing.”
Computers & Industrial Engineering 147 (2020): 106649.
[8] Ismayilov G, Topcuoglu HR. Neural network based multi-objective evolutionary algorithm for dynamic workflow scheduling in cloud com-
puting. Future Generation computer systems. 2020 Jan 1;102:307-22.
[9] Bharathi S, Chervenak A, Deelman E, Mehta G, Su MH, Vahi K. Characterization of scientific workflows. In2008 third workshop on workflows
in support of large-scale science 2008 Nov 17 (pp. 1-10). IEEE.
[10] Abazari F, Analoui M, Takabi H, Fu S. MOWS: multi-objective workflow scheduling in cloud computing based on heuristic algorithm. Simu-
lation Modelling Practice and Theory. 2019 May 1;93:119-32.
[11] Choudhary A, Gupta I, Singh V, Jana PK. A GSA based hybrid algorithm for bi-objective workflow scheduling in cloud computing. Future
Generation Computer Systems. 2018 Jun 1;83:14-26.
[12] Belgacem A, Beghdad-Bey K. Multi-objective workflow scheduling in cloud computing: trade-off between makespan and cost. Cluster Com-
puting. 2022 Feb;25(1):579-95.
[13] Alkhanak EN, Lee SP. A hyper-heuristic cost optimisation approach for Scientific Workflow Scheduling in cloud computing. Future Generation
Computer Systems. 2018 Sep 1;86:480-506.
[14] Iranmanesh A, Naji HR. DCHG-TS: a deadline-constrained and cost-effective hybrid genetic algorithm for scientific workflow scheduling in
cloud computing. Cluster Computing. 2021 Jun;24(2):667-81.
[15] Natesan G, Chokkalingam A. An improved grey wolf optimization algorithm based task scheduling in cloud computing environment. Int. Arab
J. Inf. Technol.. 2020 Jan 1;17(1):73-81.
[16] Chen W, Deelman E. Workflowsim: A toolkit for simulating scientific workflows in distributed environments. In2012 IEEE 8th international
conference on E-science 2012 Oct 8 (pp. 1-8). IEEE.
[17] Mangalampalli S, Swain SK, Mangalampalli VK. Prioritized Energy Efficient Task Scheduling Algorithm in Cloud Computing Using Whale
Optimization Algorithm. Wireless Personal Communications. 2021 Aug 25:1-7.