International Journal of Reconfigurable and Embedded Systems (IJRES)

Vol. 13, No. 2, July 2024, pp. 372~382

ISSN: 2089-4864, DOI: 10.11591/ijres.v13.i2.pp372-382  372

Hybrid fault tolerant cost aware mechanism for scientific

workflow in cloud computing

Chaya T. Doddaiah1, Mohamed Rafi2

1
Department of Computer Science and Engineering, University BDT College of Engineering, Davangere, India
2
Department of Studies in Computer Science and Engineering, University BDT College of Engineering, Davangere, India

Article Info ABSTRACT

Article history: Cloud computing provides solutions for diverse commercial and academic
applications which is the primary goal. Scientific workflows are used in the
Received Jan 30, 2023 cloud-computing environment to analyses large-scale scientific applications.
Revised Sep 5, 2023 For scientific workflows, many data is required, and a single scientific
Accepted Sep 17, 2023 workflow that includes hundreds of stages, depending on the application's
time restrictions, task failures, money limits, incorrect task organization, and
task management issues can all hinder the implementation of scientific
Keywords: methods. In light of this, a cloud-based scientific workflow management and
scheduling system that is fault-tolerant and data-oriented method are
Cloud computing proposed. This research designs a novel hybrid cost-aware fault tolerant
Cost minimization (HCFT) mechanism for minimizing the cost. Moreover, HCFT integrates
Fault-tolerant optimal clustering and efficient resource utilization through parallel and
Hybrid cost-aware fault tolerant distributed execution. Novelty of HCFT lies in novel clustering of the similar
Resource management task for improvisation, CyberShake, laser interferometer gravitational wave
observatory (LIGO), Montage, and sRNA identification protocol using high
throughput technology (SIPHT) processes are used in the simulations to
evaluate the performance of the proposed approach.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Chaya T. Doddaiah
Department of Computer Science and Engineering, University BDT College of Engineering
Davangere-577 004, Karnataka, India
Email: [email protected]

1. INTRODUCTION
The objective of "computer as a utility" has been realized via cloud computing, a new platform for
distributed computing [1]. The cloud was created by combining two significant computing technologies, cluster
computing, and grid computing [2]. Cloud computing provides specifically on-demand access to trustworthy
resources and pay-per-use customization of computer settings [3]. Cloud computing offers a range of
dynamically scaled, virtualized, abstracted, and adaptable computer resources and services [4]. Networks,
storage, servers, and applications are all made available as cloud resources in a subscription-based paradigm.
The three main architectural elements of cloud services are infrastructure as a service (IaaS), platform as a
service (PaaS), and software as a service (SaaS). Due to the demand-based availability of high-speed internet
connectivity, these services are accessible to clients outside the company SaaS.
Research and commercial applications make extensive use of cloud computing's capabilities and
services [5]. Business apps are organized according to best practices and are task-oriented. To manage such
operations, organizations adopt business models like Amazon EC2. In contrast, scientific applications are
created according to scientific principles and are data-driven. In addition to the Pegasus workflow management
system (WMS), other systems are utilized to coordinate scientific activities.

Journal homepage: http://ijres.iaescore.com

Int J Reconfigurable & Embedded Syst ISSN: 2089-4864  373

According to Faragardi et al. [6], scientific applications are data-intensive and need a substantial
amount of computing power and storage space for assessment and computation. Applications for scientific
workflows are precisely groups of computing activities structured in various ways. When organized as a
scientific process, even a single scientific application comprises several computer network tasks. Each phase
of scientific study requires a large amount of data and computational resources. Numerous additional academic
disciplines, including astronomy, biology, gravitational physics, and earthquake science, have universal
applicability. Multiple cloud models are designed to be optimized by workflow schedule; nevertheless,
customers must give priority to quality of service (QoS) satisfaction when submitting workflow applications,
including cost execution and deadline. In addition, there are issues with energy consumption, time constraints,
makespan optimization, and cost reduction due to the increasing demand for computers and services in
scientific workflow applications. Nodes are used to represent jobs in the direct acyclic graph (DAG) modeling
approach for workflows, while edges are used to show interactions between tasks [7]. DAG is therefore used
to express workflows.
The fault tolerance model utilizes a single backup in the event of a failure; as a novel concept, it was
initially enticing, but the complexity of scientific processes renders it incapable of withstanding several
failures. As a result, it is required to develop and execute a fault tolerance model that can withstand repeated
failures and increase the requirement for dependability; cost optimization is also crucial, as the ideal cost
symbolizes the model's effectiveness [8]. There are two sorts of fault occurrence methodologies for managing
the various types of failure: active and reactive fault occurrence processes. In addition, a reactive fault-tolerance
method is intended to mitigate the issue and enhance the efficacy of the fault-tolerance technique. In addition,
while incurring execution costs, this fault-tolerant technique offers system dependability [9]. The goal was to
decrease response time and energy usage. In addition, by using replication, this strategy decreases the labor
required to reject data due to system failure occurrences; yet, cost optimization remains its largest challenge.
Moreover, Shahidinejad and Barshandeh [10] created fault‐tolerant dynamic scheduling (FTDS), a fault-
tolerance technique intended to overcome the static scheduling strategy. FTDS recognizes the processor fault
and then attempts to efficiently reschedule the halted operations.
There are several actions and tasks with various limitations in scientific workflows. For management's
benefit, various workflow management techniques are used to plan and carry out scientific processes on the
intended resources [11], [12]. When managing and scheduling scientific workflows, several factors may be
used in a system that could result in performance degradation, involving the creation, organization, and control
of scientific workflows as well as task management, resource management, scheduling guidelines, and fault-
tolerant procedures [13]. Additionally, some of the processes are too big to transfer across nodes without
incurring additional costs. Five practical scientific methods for various scientific applications were thoroughly
analyzed from the start [14]. In the following fields, these techniques can be applied: i) seismic research with
CyberShake; ii) biological research with sRNA identification protocol using high throughput technology
(SIPHT); iii) astronomy using Montage; iv) genetic research with epigenomics; and v) gravitational physics
with laser interferometer gravitational wave observatory (LIGO).
The research details each scientific approach's organizational, data, and technological needs.
Additionally, words from the computer industry are used to illustrate the many structural and functional aspects
of scientific procedures. A few examples of these characteristics are the pipeline, data parallelism, data
dispersion and redistribution, data aggregation, and compositions of scientific procedures. A scalable workflow
management system used for automating research is called a WMS. It was characterized by Fan et al. [15].
Distributed computer infrastructures are changed to provide room for scientific process conceptual models. In
our research, we analyze and comprehensively address the aforementioned issues by presenting: a data-driven,
fault-tolerant model is developed.
The following are the most significant contributions of this study: i) this research work designs and
develops a hybrid cost-aware fault tolerant (HCFT) mechanism in workflow scheduling considering a cloud-
computing environment; ii) HCFT minimizes cost through novel clustering and optimal balancing mechanism,
which reduces the entire cost; moreover fault-tolerance is carried out through the clustering approach; also
parallel processing is carried out for further optimization of the process; and iii) HCFT is evaluated considering
scientific workflow to prove the model efficiency based on the execution cost for various variants of workflow
like CyberShake, Montage, SIPHT, and Inspiral.
This particular research is organized as follows: the first section starts with the background of cloud
computing and the integration of workflow scheduling along with it. The further section proceeds with the
challenges and needs for the fault-tolerant model in workflow scheduling. The section ends with a research
contribution. The second section focuses on discussing the existing fault-tolerant technique concerning cost
reduction; the third section designs the mathematical modeling of the HCFT model and is evaluated in the
fourth section.

Hybrid fault tolerant cost aware mechanism for scientific workflow in cloud … (Chaya T. Doddaiah)
374  ISSN: 2089-4864

2. RELATED WORK
Five real scientific methods with various scientific applications were thoroughly examined, starting
with Fan et al. [15]. Epigenomics is used to investigate genetics, Montage is used to study astronomy,
CyberShake is used to analyze seismic activity, SIPHT is used to study biology, and LIGO is used to study
gravitational physics. The organizational, data, and computer needs of each scientific technique are covered in
the article. Additionally, certain structural and functional features of scientific methods are examined using
computers. This includes information on data aggregation, data parallelism, data dispersion and redistribution,
and data aggregation. It also includes works that employ the scientific approach. According to research by
Ahmad et al. [16], Pegasus is a scalable WMS for the automation of research. Distributed computer systems
are constructed using abstract models of scientific procedures.
A scheduling technique for scientific workflows called dynamic scheduling of a bag of tasks-based
workflows (DSB) tries to cut costs while still following user-specified time limits. The technique breaks down
into bag of tasks (BoTs), optimizes their distribution, and then schedules them in line with priority restrictions
and data dependencies. The method satisfies the process's deadline and significantly lowers the cost of
workflow computing. Utilizing methods that quickly and efficiently utilize resources is a requirement for
planning scientific work. According to previous research [17]–[19], adaptive data-aware scheduling (ADAS)
is a scheduling technique that gives priority to resource use and workflow completion time. It is a technique
for combining task management and data for a range of jobs in the cloud. Although the suggested scheduling
approach is efficient, it does not consider error tolerance, which is an essential component of workflow
scheduling. The budget driven algorithm for generating high-quality schedules (BAGS) scheduling method
was developed in [20] to maximize process execution time while respecting budgetary limitations. The BAGS
algorithm distributes money to activities before making choices about dynamic resource provisioning and
scheduling in response to environmental changes. The approach still needs to be improved in terms of time
constraints and error-tolerance measures even if it is effective for scheduling scientific activities while adhering
to financial constraints. When organizing scientific procedures, the designed dynamic benefit weighted
scheduling (DBWS) method takes time and financial constraints into consideration.
Activities that require a lot of data and calculation are thought to utilize the most energy. To use less
energy, Juarez et al. [21] offered a real-time dynamic scheduling system for the effective execution of task-
based applications. The authors developed a polynomial-time solution that satisfies the requirements of low
energy consumption and quick execution speed by combining a resource allocation method with a set of
heuristic criteria. Authors fail to address the significant problem of accessibility of fault-tolerant systems [22].
Proclaims that when fault-tolerant practices are foregone to maintain profitability or save costs, cloud-
computing systems collapse. The fault-tolerant technique for scientific workflow systems received approval
for use in scientific operations in its second submission. For scientific processes to run properly, fault-tolerant
techniques are required since the failure of bottleneck nodes renders the entire operation meaningless [23]. If
a job is not successfully finished during execution, it is promptly resubmitted to the same resource or another.
As a result, faster, a dynamic fault-tolerant scheduling approach was created (a method for fault-tolerant
scheduling of real-time scientific workflows).
Applications for scientific methods include detailed, multi-level computation. Since all activity levels
demand the same services, this sort of computing is perfect for fault-tolerant clustering (FTC) systems. The
FTC method, developed in [24]–[26], is useful for many scientific projects. The aforementioned study offers
three techniques: dynamic clustering (DC), selective re-clustering (SR), and dynamic re-clustering (DR). DC's
initial approach is to maintain a clustering factor that is dynamically adjusted to the rate of job failure. The
second technique, known as SR, is repeatedly carrying out failed tasks inside a single position. The third
technique, known as DR, combines the first two strategies by providing failing tasks within a job a second
chance in addition to dynamically maintaining the clustering factor based on the failure rate of recognized
activities. Improvements to data-oriented scheduling methods with dynamic clustering defeat-resistant
mechanism the title of the piece is [27]. Data-oriented scheduling was integrated by the enhanced data-oriented
scheduling strategy with dynamic clustering fault-tolerant technique (EDS-DC) developers. A dynamic
clustering method with fault tolerance was offered by EDS-DC. The results of processes that were modeled
using workflows were compared to the three well-known scheduling rules minimum completion time-dynamic
clustering (MCT-DC), max-min-DC, and min-min-DC. Simulations show that EDS-DC significantly
decreased cost and manufacturing time when compared to traditional methods. For scientific activities,
Chakraborty et al. [28] proposes a QoS aware fault tolerant workflow management system (QFWMS). A
cluster-based, fault-tolerant, data-intensive (CFD) approach is given in [29] for cloud-based scientific
applications. The suggested CFD approach provides a precise strategy for obtaining outcomes from the
presentation of scientific data. The figures show that, when using the Montage workflow, the CFD approach
outperformed the alternatives [30]. Applications of scientific processes have other distinctive features aside
from pipelining, parallelism, integration, and disintegration. These programs use a lot of computation and

Int J Reconfigurable & Embedded Syst, Vol. 13, No. 2, July 2024: 372-382
Int J Reconfigurable & Embedded Syst ISSN: 2089-4864  375

information. Scientific workflow applications require workflow management and fault-tolerant scheduling
systems that are data-centric and have high data storage and processing capacities.
Scientific workflows must be gathered, categorized, and managed using a data-oriented scheduling-
based energy-efficient WMS since they are composed of several data- and computation-intensive applications
at the bottleneck node or level, a large number of workflow activities are performed, and if even one of them
fails, the execution as a whole is meaningless. A fault-tolerant system is therefore required. Given that different
workflow activity at various levels, have comparable service and resource needs, scientific operations can be
carried out using a cluster-based scheduling and fault-tolerant approach. A fault-tolerant, data-centric, and
energy-efficient system for coordinating and scheduling scientific activities is produced because of these
restrictions.

3. PROPOSED METHOD
The scientific community to perform the task using IaaS cloud as a platform has extensively utilized
workflow schedules; moreover, being fault tolerant is one of the major entities in the cloud platform, which
tends to minimize the cost per task and further workflow. This research work aims at designing a hybrid fault-
tolerant mechanism, which reduces the cost as given in Figure 1. The data here is fetched from each user, which
is further reviewed and executed. A user here refers to the individual from whom the data is collected. Here
the user submits a collection of various types of data. Once it is executed, the output is generated respectively.
The application interface (AI) here is responsible to connect the user with the proposed method. The user here
feeds the data to the AI for execution. The AI transfers data to the next component of the model, the ƥ model
that transmits the data to the workflow model for size, nodes, and edges. The AI serves as an interface to one
or more users or for workflows submitted to the proposed model.

Figure 1. Hybrid fault tolerance workflow

3.1. Designing optimal 𝐃𝐀𝐆𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰

DAGWorkflow here is depicted as ƥ, the data received by the interface for the functioning and evaluation
of various applications. For the data generated the ƥ generates a DAG graph. The ƥ functions as the data
collected by various workflows produce a reliably synthetic workflow. The DAG graph structure is represented
as shown in (1).

GDAG = (Swf , DF) (1)

GDAG denotes a DAG with features such as Swf = Swf1 , Swf2, Swf3 , … … . Swfn denotes the number of
iterations involved and DF = {(Swf(x) , Swf(y) |Swf(x) , Swf(y) ∈ Swf } denotes the DF dependencies in between the
iterations. The dependency set Swf(x) , Swf(y) consisting of the parameters in between the iterations Swf(x) and
Swf(y) . The step Swf(x) is termed the successor of the iteration Swf(y) . Any step toward the initiation is termed
the initiation step and the end step is termed the termination step shown in (2). A person can submit input and

Hybrid fault tolerant cost aware mechanism for scientific workflow in cloud … (Chaya T. Doddaiah)
376  ISSN: 2089-4864

generate an output; the ƥ receives data that contains multiple inputs used by the user. This depicts the data in
their subsequent workflows that generate a DAG which is further passed on to ϒ.

Swf(x)_initiate = Ρ (2)

Swf(x)_terminate = P (3)

3.2. Optimal-workflow execution

The executable workflow is denoted as ϒ fed with multiple workflows through the ϒ for each
resource. The ϒ rebuilt the workflow to enhance the performance. This integrates each iteration of the
workflow into a job into a single iteration henceforth the overhead is minimized. However, after that ϒ the job
includes the execution of several iterations. The iterations are arranged following their dependencies. The
iterations here are executed in a parallel and integrated sequence. At this point, the parameters for each
workflow are provided by the data for the output. Additionally, the ϒ is also responsible to produce the data
from various jobs and then estimate the storage resources, from various inputs transferred to the next model.
The ϒ transmits all the jobs to the next component.

3.3. Resource allocator

The resource allocator is denoted as ϑ allocated multiple jobs through the ϒ and then allocates
resources to it. The allocated resources are extracted from the cloud and handled separately through the
proposed technique. The ϒ is responsible to allocate the resources; this is done in a manner that the cost is
optimized. This is accomplished by a set of iterations allocated to the resources. The resource allocated for
each iteration to optimize the period to transfer data. Each iteration allocated all the resources, henceforth the
task is allocated to a resource within the optimized time limit. Here rescost(x) depicts the cost of the resource x
and DTtime denotes the data transfer time for step x for resource x. By assigning the resource to each phase
with the least amount of data transmission time possible, the list is traversed entirely.

Resource(rescost(x) ) = Step{Swf(x) |Swf(x) ← DTtime (rescost(x) )} (4)

3.4. Fault tolerant system ()

Fault tolerant system () is denoted as 𝝉 executes the iterations allocated by the ϒ which is executed
completely, the interface is used to return the results to the user. The ϒ initiates the fault-tolerant approach and
executes the failure of subsequent iterations. This model produces results upon successful execution, the 𝝉
adapts Automate_clustering () based on 𝝉 once the iteration fails. In this state, the assumption is considered
based on a 5% failure rate based on the jobs selected, henceforth the τ() is initiated in each period. The optimal
workload balancer here is responsible for cost optimization requiring resource utilization in ascending a
payload of nodes that are dispersed to use the available nodes. The resource utilization is done based on three
levels i.e Top_peak, bottom_peak, and Mid_peak.

Mid_peak (Resource usage = 0.4 to 0.8)

Resource usage (Resource(rescost(x) ) = {bottom_peak(Resource usage = 0 to 0.4) (5)
Top_peak(Resource usage = 0.8 to 1)

Here bottom_peak, denotes the lowest peak value usage and Top_peak value indicates the highest
peak value. The resources here are made available to transfer the workload from resources for the bottom_peak
value a null value when the resources with null value are terminated. The data is fed via an AI that is responsible
for providing data from two to three workflows to the proposed model. The proposed method is applied to the
ϒ which is applied to the data given as input, the ϒ is the next key component. The ϒ converts the workflow
model into appropriate resource initialization based on the iterations for the work allocated. This is then
transferred to the ϑ that sends the jobs and tasks to the workflow and then transmitted to the ϑ, this system is
responsible to schedule the resources based on the scheduling policy. The 𝝉 uses an AI, receives the
corresponding iterations, and returns the outcome to the entity. The cost-aware optimal balancer for initiating
effective cost-efficient optimization is launched when the iterations fail to execute. The energy is used based
on the ascending order of the resources, which further distributes the workload throughout the nodes for
minimal consumption of other nodes. The ϑ receives the jobs through the ϒ and schedules them by allocation
of resources to the cloud, ϒ allots the jobs upon conversion to essential tasks. The resources are fetched from
the cloud, along with the proposed system to plan and manage the tasks specifically. The ϒ allocates the
resources in a way to spend a minimal amount of money for the completion of the task. The task is completed

Int J Reconfigurable & Embedded Syst, Vol. 13, No. 2, July 2024: 372-382
Int J Reconfigurable & Embedded Syst ISSN: 2089-4864  377

by considering the list of tasks and resources, the mapping list is then generated upon the allocation of efficient
resources in the shortest period for each iteration. The integration of automated clustering of the proposed
method ensures fault tolerance mechanisms that are related to the iterations involved in a job for workflow are
combined as a cluster and execution on a single resource. The uncompleted tasks remaining in the cluster,
create clusters automatically depending on the completion of the given job and execute it repeatedly. The (6)
and (7) depict the mechanism of automatic clustering.
n
Ex(∫x=1 Swf(x) ) (6)

n
Clus(∫x=1 Swf(x) ) (7)

n
Ex (∫x=y Swf(y) ) (8)

n n
AutoClus (∫y=x Swf(y) (Fail(∫x=x Swf(x) ))) (9)

The (8) denotes the clusters for n similar tasks and executes them, henceforth the (9) denotes the job
that has failed and automatically clusters it for execution purposes. To reduce the workload, the proposed
mechanism results in evaluating the resources, arranging them in ascending order, and then automatically
clustering and transferring the workload for resource utilization of other resources and arranging them in
ascending order that transfers the workload for the resources with the least utilization of other nodes. This
categorizes the usage into three segments i.e Top_peak, bottom_peak, and Mid_peak. The workload is
transferred from the bottom_peak to the Mid_peak to make bottom_peak resources null and then the
resources with null usage are switched off. Algorithm 1 depicts the algorithm designed for minimizing the cost.

Algorithm 1. Hybrid-fault tolerant cost algorithm

Input µ(data)
Step 1 Z_1←µ (data fed to the queue), DAGWorkflow ⟶ ƥ, Executable_workflow⟶ ϒ,
Resource_Allocator →ϑ, Fault_tolerant system ⟶ 𝝉
Step 2 ƥ()
Step 3 Z_2←fetchworkflow() in the range wf1……….wfn up to Z1
Step 4 for(All the Workflow ( wf1……….wfn) do
ϒ()
Steps S1 to Sn for the Workflow
end for
ϑ()
fetchresource() from r1……….rn)
While (ϑ()) do
Steps S1 to Sn for the Workflow
for(each Steps S1 to Sn for the Workflow ( wf1……….wfn) do
fetchresource res_x and allocate it to Step S1
resx←S1
end for
end while
Step 5 Initiate_execution.τ()
While(τ ()) do
If (System Faulty)
Automate_clustering()
else
fetchresults
end if else
end while
Step 6 for(All resx to resi)do
resused←fetch(resource_used)()
If(resused(x) ==Null)then
terminate(res((x)))
else if (resused<=lowerthreshold) then
shift resused alternatively
terminate(res((x)))
end if else
end for
Step 7 𝝳←results
return 𝝳
output δ(results)

Hybrid fault tolerant cost aware mechanism for scientific workflow in cloud … (Chaya T. Doddaiah)
378  ISSN: 2089-4864

4. RESULT SECTION
In this section, the proposed model is evaluated upon further minimizing the cost optimization. The
run-time is evaluated based on allocating jobs at values 30, 50, 100, and 1,000. This is depicted in the form of
graphs for the expected outcomes shown by taking into account cost optimization. The proposed model is
tested with the CyberShake dataset by using the simulator of cloudsim. The model is deployed against 64-bit
Windows 11 operating system for 16 GB RAM that has Intel ®Core™i7 processor. It contains a 3.20 GHz
CPU, the proposed model is deployed on EclipseWS Neon.3 editor, and the code is written in JAVA.

4.1. Dataset description

The performance of the proposed model is evaluated via the different scientific areas, here CyberShake
is used to facilitate the earthquake hazards that are produced by synthetic seismograms and categorized
according to a data-centric workflow model through by wide memory-range and basic CPU necessities. The
custom Montages of the sky are based on the Montage application for the input images considered. The tasks
are differentiated by using the I/O intensive applications by not considering the CPU processing activity.
Gravitational waves are detected by using the astrophysics LIGO method. It is composed mostly of CPU-
intensive tasks with high memory requirements. Several genome-sequencing processes are automated using
the CPU-intensive epigenomics approach, which is utilized by the bioinformatics industry. In bioinformatics,
SIPHT is used to automate the search for sRNA-encoding genes. The majority of SIPHT workloads are CPU-
heavy but I/O-light. You can review examples of the design of these processes, as well as their full description
and characterization. Using the Pegasus workflow generator, five scientific processes from the actual world
were generated during experiments (CyberShake, Montage, LIGO, and SIPHT), there are 50, 100, 200, 500,
and 1,000 jobs of varying sizes; the suggested model is compared to the existing model, and all prices are
estimated in dollars.

4.2. Cost evaluation

4.2.1. CyberShake approach
The CyberShake method, which is utilized to generate the seismic hazard probability curve for
numerous areas in Southern California, takes an average of 8 hours and 51 minutes to execute. Figure 2 is a
depiction of the same topic. In addition, it creates a significant quantity of data and has a high number of jobs;
processing such a vast quantity of data demands a tremendous deal of energy, rendering the model inefficient.
The suggested model efficiently minimizes energy usage in consideration of ecosystem adoption of practices
over time (EAPT). To further evaluate the new model, it is contrasted with the current mechanism.
By considering the four CyberShake method, Table 1 compares the total expenses of the existing system
with the proposed system. In addition, the existing model for CyberShake_30 costs $22,232.6165 whereas the
proposed model costs $17,497.7745; similarly, the existing model for CyberShake_50 costs $24,669.0295
while the proposed model costs $19,036.5655. The existing model for CyberShake_100 requires $30,138.901,
but the proposed model only requires $22,674.6285 milliseconds. The existing model costs $204,102 to
perform the CyberShake_1000 job, whereas the proposed method costs $145,983.8115. Here PS stands for
proposed system and ES stands for existing system.

Figure 2. Cost comparison for CyberShake method

Table 1. Cost comparison for CyberShake method

VM ES PS Cost difference (%)
CyberShake_30 22,232.6165 17,497.7745 23.835
CyberShake_50 24,669.0295 19,036.5655 25.775
CyberShake_100 30,138.901 22,674.6285 28.267
CyberShake_1000 204,102 145,983.8115 33.202

Int J Reconfigurable & Embedded Syst, Vol. 13, No. 2, July 2024: 372-382
Int J Reconfigurable & Embedded Syst ISSN: 2089-4864  379

4.2.2. Laser interferometer gravitational wave observatory approach

The LIGO method is used to generate and analyze gravitational waves from data collected by merging
compact binary models. The LIGO workflow is represented in Figure 3. In addition, this process has four
separate DAX files; the present model for Inspiral_30 costs $22,951.4585, but the proposed model costs
$12,673.1235. Inspiral_50 and in Inspiral_100 incur additional costs of $22,519.9115 and $40,281.976,
respectively, compared to the older models' costs of $40,788.6025 and $72,942.9245. The existing model costs
789,741.133 and the proposed model costs 436,056.0895 for Inspiral_1000. Table 2 compares the values, and
Figure 3 illustrates a graph comparing the existing and proposed values.

Figure 3. Graphical comparison of LIGO method

Table 2. Cost comparison for LIGO method

VM ES PS Cost difference (%)
Inspiral_30 22,951.4585 12,673.1235 57.704
Inspiral_50 40,788.6025 22,519.9115 57.713
Inspiral_100 72,942.9245 40,281.976 57.692
Inspiral_1000 789,741.133 436,056.0895 57.707

4.2.3. Montage approach

With the NASA-developed Montage approach, several images are fed as input and are used to create
one-of-a-kind mosaics. The Montage technique requires four different dax files for a cost comparison. The
existing model forecasts costs of $846.8255 and $1,872.66 for Montage_25 and Montage_50, whereas the
proposed model, predicts costs of $482.8945 and $1,062.9835for Montage_25 and Montage_50, respectively.
Figure 4 displays the cost comparison on Montage method. Table 3 displays the cost comparison on Montage
technique. The existing model for Montage_100 and Montage_1000 costs $3,944.9945 and $41,426.565, but
the proposed model costs $2,231.8825 and $23,398.8515, respectively.

Figure 4. Cost comparison on Montage method

Hybrid fault tolerant cost aware mechanism for scientific workflow in cloud … (Chaya T. Doddaiah)
380  ISSN: 2089-4864

Table 3. Cost comparison on Montage technique

VM ES PS Cost difference (%)
Montage_25 846.8255 482.8945 54.738
Montage_50 1,872.66 1,062.9835 55.162
Montage_100 3,944.9945 2,231.8825 55.469
Montage_1000 41,426.565 23,398.8515 55.619

4.2.4. SIPHT approach

SIPHT is a Harvard project that searches the NCRI database of bacterial replicons for untranslated
RNAs. Four SIPHT DAX files, namely SIPHT_30, SIPHT_60, SIPHT_100, and SIPHT_1000, have been
examined. Existing models for SIPHT_30 and SIPH_60 cost $19,169.0855 and $40,314.883, while the
proposed models cost $10,565.216 and $22,216.6525, respectively. Existing models of the SIPHT_100, and
SIPHT_1000 cost $60,047.434 and $599,918.9015, whereas the proposed models cost $33,091.578 and
$330,580.468 respectively. Figure 5 and Table 4 give comparative cases.

Figure 5. Cost comparisons of SIPHT method

Table 4. Cost comparisons of SIPHT method

VM ES PS Cost difference (%)
Sipht_30 19,169.0855 10,565.216 57.872
Sipht_60 40,314.883 22,216.6525 57.885
Sipht_100 60,047.434 33,091.578 57.883
Sipht_1000 599,918.9015 330,580.468 57.891

4.3. Comparative analysis

In this section, several methods are compared. The cost for PS is 23.835%, 25.775%, 28.267% and
33.202 % less expensive than the prior model for CyberShake_30, CyberShake_50, CyberShake_100, and
CyberShake_1000, respectively. Similarly, the LIGO procedure's PS criteria are 57.704%, 57.713%, 57.692%,
and 57.707% for Inspiral_30, Inspiral_50, and Inspiral_1000, respectively. For Montage_25, Montage_50,
Montage_100, and Montage_1000, PS is 54.738%, 55.162%, 55.469%, and 55.619% less expensive than the
preceding model. The cost difference between ES and PS for SIPHT_30, SIPHT_60, SIPHT_100, and
SIPHT_1000 is $57.872, $57.885, $57.883, and $57.891, respectively. In addition, the comparative analysis
states that the proposed model performs better in comparison with the existing model.

5. CONCLUSION
This research work develops HCFT mechanism for cost reduction; HCFT comprises novel clustering
and optimal resource utilization mechanism for fault tolerance, which tends to reduce workflow cost. HCFT
mechanism is evaluated considering various scientific workflow with its various variant. Furthermore,
considering cost as an evaluation parameter, HCFT observes marginal improvisation over the existing model.
In case of CyberShake_30, CyberShake_50, CyberShake_100, and CyberShake_1000, HCFT-mechanism
observes 23.83%, 25.77%, and 28.26% respectively. Furthermore, considering the Inspiral workflow, HCFT-
mechanism observes improvisation of 57.704%, 57.713%, 57.692%, and 57.707% for Inspiral_30, Inspiral_50,

Int J Reconfigurable & Embedded Syst, Vol. 13, No. 2, July 2024: 372-382
Int J Reconfigurable & Embedded Syst ISSN: 2089-4864  381

and Inspiral_1000, respectively. At last, HCFT-mechanism observes improvisation of 57.872%, 57.885%,

57.883%, and 57.891% for SIPHT_30, SIPHT _60, SIPHT_100, and SIPHT _1000. HCFT observes marginal
improvisation; however, considering the complexity of scientific workflow, future work should be focused on
the implementation of a machine learning approach to predict the resources efficiently.

REFERENCES
[1] X. Li et al., “A novel workflow-level data placement strategy for data-sharing scientific cloud workflows,” IEEE Transactions on
Services Computing, vol. 12, no. 3, pp. 370–383, May 2019, doi: 10.1109/TSC.2016.2625247.
[2] A. Song, W.-N. Chen, X. Luo, Z.-H. Zhan, and J. Zhang, “Scheduling workflows with composite tasks: a nested particle swarm
optimization approach,” IEEE Transactions on Services Computing, vol. 15, no. 2, pp. 1074–1088, Mar. 2022, doi:
10.1109/TSC.2020.2975774.
[3] S. Qin, D. Pi, Z. Shao, and Y. Xu, “A knowledge-based adaptive discrete water wave optimization for solving cloud workflow
scheduling,” IEEE Transactions on Cloud Computing, vol. 11, no. 1, pp. 200–216, Jan. 2023, doi: 10.1109/TCC.2021.3087642.
[4] X. Tang, “Reliability-aware cost-efficient scientific workflows scheduling strategy on multi-cloud systems,” IEEE Transactions on
Cloud Computing, vol. 10, no. 4, pp. 2909–2919, Oct. 2022, doi: 10.1109/TCC.2021.3057422.
[5] W. Chen, R. F. da Silva, E. Deelman, and T. Fahringer, “Dynamic and fault-tolerant clustering for scientific workflows,” IEEE
Transactions on Cloud Computing, vol. 4, no. 1, pp. 49–62, Jan. 2016, doi: 10.1109/TCC.2015.2427200.
[6] H. R. Faragardi, M. R. S. Sedghpour, S. Fazliahmadi, T. Fahringer, and N. Rasouli, “GRP-HEFT: a budget-constrained resource
provisioning scheme for workflow scheduling in IaaS clouds,” IEEE Transactions on Parallel and Distributed Systems, vol. 31, no.
6, pp. 1239–1254, Jun. 2020, doi: 10.1109/TPDS.2019.2961098.
[7] M. Niu, B. Cheng, Y. Feng, and J. Chen, “GMTA: a geo-aware multi-agent task allocation approach for scientific workflows in
container-based cloud,” IEEE Transactions on Network and Service Management, vol. 17, no. 3, pp. 1568–1581, Sep. 2020, doi:
10.1109/TNSM.2020.2996304.
[8] H. Ghavamipoor, S. A. K. Mousavi, H. R. Faragardi, and N. Rasouli, “A reliability aware algorithm for workflow scheduling on
cloud spot instances using artificial neural network,” in 2020 10th International Symposium onTelecommunications (IST), Dec.
2020, pp. 67–71, doi: 10.1109/IST50524.2020.9345896.
[9] F. Yao, C. Pu, and Z. Zhang, “Task duplication-based scheduling algorithm for budget-constrained workflows in cloud computing,”
IEEE Access, vol. 9, pp. 37262–37272, 2021, doi: 10.1109/ACCESS.2021.3063456.
[10] A. Shahidinejad and S. Barshandeh, “Sink selection and clustering using fuzzy‐based controller for wireless sensor networks,”
International Journal of Communication Systems, vol. 33, no. 15, Oct. 2020, doi: 10.1002/dac.4557.
[11] Y. Ding, G. Yao, and K. Hao, “Fault-tolerant elastic scheduling algorithm for workflow in Cloud systems,” Information Sciences,
vol. 393, pp. 47–65, Jul. 2017, doi: 10.1016/j.ins.2017.01.035.
[12] A. Marahatta, Y. Wang, F. Zhang, A. K. Sangaiah, S. K. S. Tyagi, and Z. Liu, “Energy-aware fault-tolerant dynamic task scheduling
scheme for virtualized cloud data centers,” Mobile Networks and Applications, vol. 24, no. 3, pp. 1063–1077, Jun. 2019, doi:
10.1007/s11036-018-1062-7.
[13] S. Bharathi, A. Chervenak, E. Deelman, G. Mehta, M.-H. Su, and K. Vahi, “Characterization of scientific workflows,” in 2008
Third Workshop on Workflows in Support of Large-Scale Science, Nov. 2008, pp. 1–10, doi: 10.1109/WORKS.2008.4723958.
[14] A. Marahatta, Q. Xin, C. Chi, F. Zhang, and Z. Liu, “PEFS: AI-driven prediction based energy-aware fault-tolerant scheduling
scheme for cloud data center,” IEEE Transactions on Sustainable Computing, vol. 6, no. 4, pp. 655–666, Oct. 2021, doi:
10.1109/TSUSC.2020.3015559.
[15] G. Fan, L. Chen, H. Yu, and D. Liu, “Modeling and analyzing dynamic fault-tolerant strategy for deadline constrained task
scheduling in cloud computing,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 50, no. 4, pp. 1260–1274,
Apr. 2020, doi: 10.1109/TSMC.2017.2747146.
[16] Z. Ahmad, A. I. Jehangiri, N. Mohamed, M. Othman, and A. I. Umar, “Fault tolerant and data oriented scientific workflows
management and scheduling system in cloud computing,” IEEE Access, vol. 10, pp. 77614–77632, 2022, doi:
10.1109/ACCESS.2022.3193151.
[17] L. Zeng, B. Veeravalli, and A. Y. Zomaya, “An integrated task computation and data management scheduling strategy for workflow
applications in cloud environments,” Journal of Network and Computer Applications, vol. 50, pp. 39–48, Apr. 2015, doi:
10.1016/j.jnca.2015.01.001.
[18] V. Arabnejad, K. Bubendorfer, and B. Ng, “Scheduling deadline constrained scientific workflows on dynamically provisioned cloud
resources,” Future Generation Computer Systems, vol. 75, pp. 348–364, Oct. 2017, doi: 10.1016/j.future.2017.01.002.
[19] M. A. Rodriguez and R. Buyya, “Budget-driven scheduling of scientific workflows in IaaS clouds with fine-grained billing periods,”
ACM Transactions on Autonomous and Adaptive Systems, vol. 12, no. 2, pp. 1–22, Jun. 2017, doi: 10.1145/3041036.
[20] M. Ghasemzadeh, H. Arabnejad, and J. G. Barbosa, “Deadline-budget constrained scheduling algorithm for scientific workflows in
a cloud environment,” in Leibniz International Proceedings in Informatics, LIPIcs, 2017, vol. 70, pp. 19.1-19.16, doi:
10.4230/LIPIcs.OPODIS.2016.19.
[21] F. Juarez, J. Ejarque, and R. M. Badia, “Dynamic energy-aware scheduling for parallel task-based application in cloud computing,”
Future Generation Computer Systems, vol. 78, pp. 257–271, Jan. 2018, doi: 10.1016/j.future.2016.06.029.
[22] K. Ganga and S. Karthik, “A fault tolerent approach in scientific workflow systems based on cloud computing,” in 2013
International Conference on Pattern Recognition, Informatics and Mobile Engineering, Feb. 2013, pp. 387–390, doi:
10.1109/ICPRIME.2013.6496507.
[23] I. Casas, J. Taheri, R. Ranjan, L. Wang, and A. Y. Zomaya, “A balanced scheduler with data reuse and replication for scientific
workflows in cloud computing systems,” Future Generation Computer Systems, vol. 74, pp. 168–178, Sep. 2017, doi:
10.1016/j.future.2015.12.005.
[24] W. Chen and E. Deelman, “Fault tolerant clustering in scientific workflows,” in 2012 IEEE Eighth World Congress on Services,
Jun. 2012, pp. 9–16, doi: 10.1109/SERVICES.2012.5.
[25] Z. Ahmad, A. I. Jehangiri, M. Iftikhar, A. I. Umer, and I. Afzal, “Data-oriented scheduling with dynamic-clustering fault-tolerant
technique for scientific workflows in clouds,” Programming and Computer Software, vol. 45, no. 8, pp. 506–516, Dec. 2019, doi:
10.1134/S0361768819080097.
[26] Z. Ahmad, B. Nazir, and A. Umer, “A fault‐tolerant workflow management system with quality‐of‐service‐aware scheduling for
scientific workflows in cloud computing,” International Journal of Communication Systems, vol. 34, no. 1, Jan. 2021, doi:
10.1002/dac.4649.

Hybrid fault tolerant cost aware mechanism for scientific workflow in cloud … (Chaya T. Doddaiah)
382  ISSN: 2089-4864

[27] Z. Ahmad, A. I. Jehangiri, M. A. Ala’anzy, M. Othman, and A. I. Umar, “Fault-tolerant and data-intensive resource scheduling and
management for scientific applications in cloud computing,” Sensors, vol. 21, no. 21, pp. 1-19, Oct. 2021, doi: 10.3390/s21217238.
[28] D. Chakraborty, V. V. Mankar, and A. A. Nanavati, “Enabling runtime adaptation of workflows to external events in enterprise
environments,” in IEEE International Conference on Web Services (ICWS 2007), Jul. 2007, pp. 1112–1119, doi:
10.1109/ICWS.2007.85.
[29] E. Deelman et al., “Pegasus, a workflow management system for science automation,” Future Generation Computer Systems, vol.
46, pp. 17–35, May 2015, doi: 10.1016/j.future.2014.10.008.
[30] S. H. Sreedhara, V. Kumar, and S. Salma, “Efficient big data clustering using adhoc fuzzy c means and auto-encoder CNN,” in
Inventive Computation and Information Technologies, Singapore: Springer, 2023, pp. 353–368, doi: 10.1007/978-981-19-7402-
1_25.

BIOGRAPHIES OF AUTHORS

Chaya T. Doddaiah earned her bachelors of engineering (B.E.) degree in computer

science and engineering from VTU, Belagavi in 2007. She obtained her master's degree in
M.Tech. (computer science and engineering) from Siddhartha University in 2011 and currently
she is a research scholar at University BDT College of Engineering, Davangere, doing her Ph.D.
in computer science and engineering. She has more than 10+ years of experience in teaching,
worked as an assistant professor and attended many workshops and induction programs
conducted by various universities. Her areas of interest are big data analytics and cloud
computing. She can be contacted at email: [email protected], [email protected].

Dr. Mohamed Rafi from Koratagere, Tumkur, obtains B.E., M.E., and Ph.D. in
computer science and engineering. He is having more than 25 years of experience in teaching
and research. Currently he is working as professor in Department of Studies in Computer Science
and Engineering, University BDT College of Engineering, Davangere, a constituent College of
Visvesvaraya Technological University, Belagavi, India. Also serving as member of board of
studies, Bangalore University, Subject Expert of Doctoral Committee Bangalore University and
BOE of Computer Science and Information Science Board, VTU. Worked at Universities of
Ethiopia and Saudi Arabia. Published more than 170 papers in conferences, national, and
international journals. His area of interest includes digital image processing and managing big
data. He can be contacted at email: [email protected].

Int J Reconfigurable & Embedded Syst, Vol. 13, No. 2, July 2024: 372-382