YOU ARE DOWNLOADING DOCUMENT

Please tick the box to continue:

Transcript
Page 1: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

Research ArticleIFFOAn ImprovedFruit FlyOptimizationAlgorithm forMultipleWorkflow Scheduling Minimizing Cost and Makespan in CloudComputing Environments

Ambika Aggarwal 1 Priti Dimri2 Amit Agarwal 3 Madhushi Verma 4

Hesham A Alhumyani 5 and Mehedi Masud 6

1School of Computer Science University of Petroleum and Energy Studies Dehradun India2Department of Computer Applications Govind Ballabh Pant Engineering College Pauri India3Dr APJ Abdul Kalam Institute of Technology Tanakpur India4Department of Computer Science Engineering Bennett University Uttar Pradesh India5Department of Computer Engineering College of Computers and Information Technology Taif University PO Box 11099Taif 21944 Saudi Arabia6Department of Computer Science College of Computers and Information Technology Taif University PO Box 11099Taif 21944 Saudi Arabia

Correspondence should be addressed to Mehedi Masud mmasudtuedusa

Received 25 April 2021 Accepted 14 May 2021 Published 4 June 2021

Academic Editor Vijay Kumar

Copyright copy 2021 Ambika Aggarwal et al (is is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

Cloud computing platforms have been extensively using scientific workflows to execute large-scale applications Howevermultiobjective workflow scheduling with scientific standards to optimize QoS parameters is a challenging task Various met-aheuristic scheduling techniques have been proposed to satisfy the QoS parameters like makespan cost and resource utilizationStill traditional metaheuristic approaches are incompetent to maintain agreeable equilibrium between exploration and ex-ploitation of the search space because of their limitations like getting trapped in local optimum value at later evolution stages andhigher-dimensional nonlinear optimization problem (is paper proposes an improved Fruit Fly Optimization (IFFO) algorithmto minimize makespan and cost for scheduling multiple workflows in the cloud computing environment(e proposed algorithmis evaluated using CloudSim for scheduling multiple workflows(e comparative results depict that the proposed algorithm IFFOoutperforms FFO PSO and GA

1 Introduction

Cloud is an infinite pool of configurable computing re-sources (storage network processor bandwidth etc) withsome functionalities such as an on-demand pay-per-usemodel high availability scalability and reliability [1 2] Italso supports the distributed architecture for geographicallydistributed heterogeneous resources and provisioning themto clients through virtualization for hosting large-scale ap-plications (ese applications are deployed in the form ofworkflows which are further divided into smaller tasks

Due to continuously increasing workloads and the rise intheir difficulty levels workflow scheduling has become awidely studied cloud computing problem that attracts manyresearchers As depicted in Figure 1 workflow scheduling isused to allocate the required resources to the appropriatetasks to complete the execution process During schedulingtasks on the virtual machine (VM) the clientrsquos QoS con-straints must be fulfilled Different clients may have differentQoS requests in terms of cost time security and so forth

Multiple workflow scheduling comes into considerationto maximize the cloud architecturersquos throughput when

HindawiMathematical Problems in EngineeringVolume 2021 Article ID 5205530 9 pageshttpsdoiorg10115520215205530

various client requests are received simultaneously Similartasks can be identified to be allocated on a similar set ofresources [3] Strategies must be adopted to enhance thesystemrsquos performance and ensure that all client requests arecompleted before the deadline

An efficient scheduling technique maintains a trade-offbetween user requirements and resource utilization [4]Maintaining this trade-off becomes challenging when sometasks have a parent-child relationship where a child task canonly begin executing once its parent task has finished and allthe output data from the predecessor task has been com-municated to the child task [5 6] Various list-based al-gorithms cannot be directly implemented in cloudcomputing environments because of the resource hetero-geneity varied QoS constraints and cloudrsquos dynamic nature[2] Several metaheuristic algorithms such as FFO [7] PSO[8] GA [9 10] have been explored for solving workflowscheduling problems However optimizing multiple ob-jectives is still a challenging task for the CCE [11ndash14]Optimizing one QoS parameter often results in compro-mising with the other QoS parameter (us schedulingmultiple workflows on the cloud while maintaining a trade-off among multiple QoS parameters remains a problem thatneeds to be solved

In this paper an enhanced metaheuristic optimizationtechnique IFFO has been proposed for scheduling multipleworkflows on cloud computing environments to optimizemultiple QoS parameters (e results of the proposedtechnique were generated using CloudSim and comparedwith existing FFO PSO and GA algorithms to validate theperformance of IFFO in terms of makespan and costparameters

(e rest of the paper is organized as follows Section 1describes the introductory concepts of cloud computingrelated to workflow scheduling A crisp and concise lit-erature survey on workflow scheduling for QoS param-eters is discussed in Section 2 Section 3 highlights theproblem formulation and problem definition A novel

framework using the IFFO algorithm is proposed inSection 4 (e proposed algorithmrsquos experimental resultsare given in Section 5 and the conclusion is mentioned inSection 6

2 Background

Yassa et al have presented a new multiobjective approachcalled DVFS-MODPSO [15] for scheduling workflows onthe cloud computing environment (e presented algorithmis a hybridization of PSO with Heterogeneous Earliest FinishTime (HEFT) aiming to optimize multiple objectives likemakespan cost and energy consumption Dynamic Voltageand Frequency Scaling (DVFS) is used for energy optimi-zation and the results show better Pareto optimal solutionsthan HEFT

CGA2 [16] is a technique proposed by Liu et al whichincludes an adaptive penalty function for schedulingdeadline constrained workflows in the cloud computingenvironment addressing the limitations of previouslyproposed evolutionary algorithms (e proposed algorithmprevents premature convergence unlike several existingstatic techniques by applying adaptive crossover and mu-tation probabilities and generates solutions that are able tomeet deadline constraints CGA2 is compared with tradi-tional algorithms such as PSO HEFT GA and Random todemonstrate better performance in terms of meetingdeadlines under strict constraints and reducing the overallworkflow execution cost

HSGA [17] is a GA-based hybrid workflow schedulingtechnique adopted by Delavar et al which utilizes theoptimization characteristics of Round Robin (RR) and BestFit (BF) scheduling algorithms Initially the proposedtechnique does the priority ranking of tasks based on theirdependencies and then the resource allocation is done byimplementing RR and BF for appropriate VM selection (eexperimental results depicted better performance of HSGAin terms of reducing makespan lowering failure rate and

Workflow Resource pool

Taskscheduler

T1

T2

T3 T4

T5 T6

T7

R1

R2

R3

Rn

Figure 1 Basic concept of workflow scheduling

2 Mathematical Problems in Engineering

balancing the load when compared with LAGA and NGAscheduling algorithms

CDMWS [18] is a dynamic optimization technique forscheduling multiple workflows on the cloud proposed byDelavar et al which aims at improving CPU utilizationreducing makespan and improving the makespan-deadlinemeeting ratio (e proposed technique is also divided intotwo stages (e first stage is responsible for estimating theexecution time for each task by considering workflowdeadline and task dependencies (e second stage is re-sponsible for dynamic VM allocation where VMs can reusefor tasks having similar requirements VM reusability isimplemented to lower power consumption and increaseresource utilization CDMWS is compared with two otheralgorithms EWSA and RR to verify its superiority

Another list-based heuristic MOWS [19] introduced byF Abazari et al adopts the greedy approach for prioritizingtasks and allocating appropriate resources to them (eproposed technique aims at improving the security of theoverall cloud architecture while maintaining the executiontime of workflows (e first phase of the algorithm designsthe solution based on task prioritization and assesses se-curity risk (e second part proposes an algorithm to dealwith various security threats while scheduling a workflow onthe cloud

GA-based multiobjective workflow scheduling algo-rithm MOGA [20] presented by Attiqa Rehman et al aimsat optimizing a diverse range of objectives includingmakespan budget resource utilization deadline and energyefficiency A gap search algorithmwas also introduced in thiswork that finds gaps in the schedule generated for a par-ticular workflow and fills them with independent tasks tomaximize resource utilization MOGA was compared withthree GA-based algorithms and one PSO-based algorithm(MOPSO) to validate its superiority Table 1 shows thesummary of the related works

3 Problem Formulation

Any large-scale application that needs to be deployed on acloud platform is generally represented in the form of aworkflow W (T E) A workflow can be pictorially repre-sented using a directed acyclic graph (DAG) where T T1T2 Tn denotes the set of tasks(e complete application isdivided into several dependent and independent subtasks(e tasks in a workflow are represented at different levelshaving a parent-child relationship where a child task cannotbegin execution until all its parent tasks have finished ex-ecution and all output data has been transferred to the childtask An edge Eij from Ti to Tj represents that Ti is the parentof Tj and exist a dependency between Ti and Tj

Consider a sample workflow depicted in Figure 2 (eentire application is divided into seven subtasks T1 T2 T7falling at five different levels that is Level 0 to Level 4 Atlevel 2 Tasks T3 and T4 are independent of each other sincethey are at the same level so they can be executed con-currently on different resources However their executioncan only begin once T2 at Level 1 has finished its executionand transferred all output data to T3 and T4 Similarly the

execution of T2 depends on its predecessor task T1 Since T1has no predecessor it will be the first task to be executedWorkflow execution can consider being completed once thelast task in the DAG T7 has finished its execution andgenerated its output

Resource heterogeneity will also be considered as pro-vided by any IaaS cloud provider Different types of VMs willbe available based on different configurations Any cloudservice provider who joins the cloud marketplace provides atwo-dimensional bid BVMi (PVMi CVMi) where PVMi rep-resents the processing capacity of the VMmeasured in termsof MIPS and CVMi is the cost of execution on that VM (epricing model is based on the current Amazon EC2 stan-dards where a full cycle consists of 60minutes and one extraminute will count for one complete cycle So if a resource isconsumed for 61 minutes the user will be charged for twocomplete cycles that is 120 minutes

Execution time ETlj of a task Tj on a resource VMl where

Tj T1 T2 T3 TJ forall j isin 1 2 3 J is calculatedusing equation (1) by considering the size STj

of task Tj interms of MIPS and the performance variation factor Pvarl

ofVMl introduced while adjusting the processing capacity of aVM Since a workflow involves task dependency the datatransfer time DTab from tasks Tj to Tk can be calculated usingequation (2) where DoutTj

is the amount of output datagenerated by Tjwhich is assumed to be known in advance foreach task and bw is the bandwidth between each VM Alsothe data transfer rate between two tasks scheduled on thesame resource will be zero Hence the total processing timePTl

jof each task Tj on a resource VMl is calculated usingequation (3) where e is the number of edges connected to aparent task Tj and Qe 0 if Tj and Tk are scheduled on thesame VM else 1

ETlj

STj

PVMllowast 1 minus Pvarl

1113872 11138731113872 1113873 (1)

DTjk DoutTj

bw (2)

PTlj ETl

j + 1113944e

1DTjk lowastQ⎛⎝ ⎞⎠ (3)

Similarly multiple workflow scheduling problems canalso be formulated Consider the diagram shown in Figure 3It consists of three small workflows all of which can becombined to form a single large workflow by addingDummystart and Dummyend as the starting and ending tasksrespectively (e execution time of both of these tasks will bezero as they are included only for merging the smallerworkflows

Optimal task scheduling and resource provisioning canbe done based on various objectives (is work focuses onoptimizing two scheduling objectives that is makespan andcost by finding a schedule S (Res Map Zct Zms) forscheduling workflow on cloud computing environmentwhere Res r1 r2 rc is the set of available resourcesMap depicts the task to resource mapping in the form

Mathematical Problems in Engineering 3

Mrc

tj (tj rc STtj

ETtj) for each task in the workflow which

means that a task tj is scheduled on resource rc and it willbegin execution at start time STtj

and will finish executionat the end time ETtj

Zct and Zms represent total executioncost and total execution time and can be calculated with thehelp of the following two equations respectively

Zct 1113944

|R|

c1CVMrclowast

LETrcminus LSTrc

1113872 1113873

α

⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦ (4)

Zms max ETtj1113882 1113883 (5)

where α denotes one cycle of the time unit for which theVM is charged LETrc

is the lease end time for resource rcand LSTrc

is the lease start time A sample schedule gen-erated for workflow shown in Figure 2 is depicted inFigure 4

Hence this paper works on finding an optimal scheduleS to minimize total execution cost and total execution timebased on the definitions given so far

4 Fruit Fly Optimization

Fruit Fly Optimization Algorithm [7] has been widelyadopted for solving global optimization problems because ofits simple structure and lesser number of parameters (ealgorithm is inspired by the food-finding ability of fruit flieswhere they can smell food even at a distance of 40 km Once

Table 1 (e summary of the related works

ProposedAlgorithm Objectives considered Algorithm type Workflow type Scheduling type

(SD) Tool used Year

DVFS-MODPSO [15] Makespan cost and energy HEFT+particle swarm

optimizationSimple andscientific Static CloudSim 2013

HSGA [17] Makespan failure rate and loadbalancing Genetic algorithm Complex Static + dynamic Simulator 2013

CGA2 [16] Constrained deadlines and cost Genetic algorithm Scientific mdash CloudSim 2016

CDMWS [18] CPU utilization makespandeadline List-based Scientific

(multiworkflow) Dynamic Simulator 2017

MOGA [20]Makespan budget deadlineenergy efficiency resource

utilizationGenetic algorithm Real-life Static Simulator 2018

MOWS [19] Execution time and security List-based greedyapproach

Scientific(multiworkflow) mdash WorkflowSim 2019

Level 0

Level 1

Level 2

Level 3

Level 4

T1

T2

T3 T4

T5 T6

T7

Figure 2 Sample workflow

Dummystart

Dummyend

Figure 3 Representation of multiple workflows as a singleworkflow

4 Mathematical Problems in Engineering

they get closer to the food source they use their sensitivevision for flying towards the food direction

FFO works in various phases as follows

(1) (e first phase is the initialization phase where thefruit flies are randomly distributed in the searchspace and their location (X_init Y_init) is initialized

(2) In the second phase each fruit fly is given somerandom direction and distance(X_init + RandomValue Y_init + RandomValue) tomove towards the food source

(3) Next the distance between each fruit fly and the foodlocation is estimated and the smell concentration iscalculated which is the reciprocal distance

(4) (e algorithm then goes into the fitness evaluationphase which is a function based on the smellconcentration

(5) (e maximum smell concentration of the individualfruit fly is retained and the swarm updates its po-sition to move in that direction

(6) (ese steps are iteratively repeated and the result ofeach iteration is compared with the previous one tocheck whether optimized results are obtained or not

FFO algorithm became popular because of its easy-to-implement structure and quick convergence However it isnot found suitable for complex optimization problems as itcould get trapped in the local optima at later evolution stagesand might not reach the global optima Also the conver-gence rate of the algorithm for complex optimization couldbe improved

Hence this paper presents an enhanced version of thetraditional FFO algorithm which could be implemented forcomplex optimization problems such as scheduling multipleworkflows in the cloud computing environment

41 Proposed Framework Figure 5 presents the proposedframework for IFFO Consider a cloud provider with a set ofvirtual machines VMl VM1 VM2 VM3 VML forall l isin1 2 3 L having some computational capacity In acloud computing scenario there are multiple workflowsWi W1 W2 W3 WI forall i isin 1 2 3 I thatneed to be scheduled with optimized QoS parametersKeeping this in mind multipleWi are merged and convertedinto a unified workflow Sw and Ew tasks are added at thestarting and ending position of Wi forall workflows Wi theremay be several tasks Tj T1 T2 T3 TJ forall j isin 1 2 3 J

Each resource VMl is available on-demand is accessiblefrom a shared pool of computing resources and has someQoS parameter associated with it For the present researchwork the authors have considered makespan and cost asQoS parameters Cost of running each VM isVMCm VMC1 VMC2 VMC3 VMCM forallm isin 1 23 M and makespan VMMs VMM1 VMM2 VMM3 VMMS forall s isin 1 2 3 S

Maximum cost and makespan are calculated by theaddition of cost and makespan of each task (e objectiveof the abovementioned problem is to minimize the costand makespan for executing the entire workflow (usthese multiple objective optimization problems try to findout the Pereto optimal solution in each iteration Once thecost and makespan are optimized while Tj ne TJ the IFFOalgorithm is applied on input values to find out the bestsmell function in each iteration Based on updated smellvalues the entire swarm population updates their smellconcentration and becomes ready for the next iterationWhen maximum iterations are completed or Paretooptimal solution is achieved all the tasks are sent forsimulation purposes

42 Proposed IFFO Algorithm (e proposed IFFO opti-mization algorithm is an enhanced version of the traditionalFruit Fly Optimization Algorithm (is algorithm is used tooptimize multiple objectives that is cost and makespan formultiple workflow scheduling in the cloud environment(eproposed algorithm continuously optimizes the old solutionusing the smell concentration function (is paper alsoshows an improvement in coverage rate by updating ap-propriate positions in each iteration

(e main steps of the proposed method are described asfollows

(1) Input constraints let n be the swarm size of fruit flypopulation the initial position of each fruit fly isSPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P forall p isin1 2 3 P Here each swarm particle repre-sents a possible solution moving towards a ran-dom direction with randomized distance Rv Asper traditional FFO the maximum iterationshould be 20ndash40 for current research work themaximum iteration (distance) of fruit fly move-ment is Q 20ndash40

R1

R2

R3

10 20 30 40 50 60 70 80

T1 T3 T5

T6 T7

T6 T7

Time

Reso

urce

s

VM Booting Time

VM Idle Time

VM Shutdown Time

Figure 4 Sample schedule

Mathematical Problems in Engineering 5

(2) Output constraints the target is to find out rank-1Pareto optimal solution for given input values Heref (P) S1 S2 S3 SN forall N isin 1 2 3 K is aset of solutions with lower bound LB and upperbound UB ZCtMs is the total cost and makespan ofworkflows that need to be minimized

(3) (is step defines the objective function f(obj) whereSf is the scaling factor R is a randomized functionand 1113954ℷ 1113954Ωamp 1113954Z are arbitrary constant that is1113954ℷ + 1113954Ω + 1113954Z 1

(4) For terminating condition (maximum iteration)Imax (t) started from 1 to T

(5) Calculate the initial position of each swam particlex

fft xα + Rv andy

fft yα + Rv where the initial

position (xfft y

fft )of each swarm particles isin SPloc

p SPloc 1 SPloc 2 SPloc 3 SPloc P and Rv is therandom variable ranging from 0 to 1

(6) Distance between individual swarm and food iscalculated by

(xfft )2 + (y

fft )2

1113969

and smell concen-tration by 1Distt

(7) Calculate f(SCt ) forall SSq where q 1 23 n that

is smell concentration of each individual fruit fly(8) Find out the mean of smell concentration F(Smellt)(9) Update the swarm particles position with updated

values of (xα yα) that is xα xα + xα lowastRv(0 1) +

xα lowastF(Smellt) and yα yα + yα lowastRv(0 1) + yα lowastF(Smellt) and go to step 3

(e algorithmic representation of these steps is men-tioned below (Algorithm 1)

5 Results and Discussion

51Dataset andSimulationSetup (e experimental analysiswas conducted using the CloudSim framework [21] thesimulation tool used for simulating cloud environments(eproposed algorithm was implemented for three differentdatasets and results were compared with three other met-aheuristic optimization techniques FFO GA and PSODatasets differ in terms of the number of tasks in a workflowand the number of resources available Although the cloudenvironment is considered to have an unlimited set of re-sources for arriving at an optimal solution we need to limitthe number of resources as well We have considered threesample workflows consisting of 15 25 and 35 tasks Forthese three workflows the number of resources is assumedto be 5 10 and 15 respectively

52 Performance Analysis (e proposed IFFO algorithmis compared with PSO GA and FFO based on twoscheduling objectives makespan and cost (e algo-rithms were executed for 20 iterations and the resultsdepict better performance of IFFO as compared with theother algorithms both in terms of makespan and cost(e experimental results are presented in the graphsshown in Figures 6ndash8 for datasets 1 2 and 3(e blue linerepresents the cost of execution while the orange linedepicts the makespan It is clear from these graphs thatIFFO outperforms PSO GA and FFO in bothparameters

(e percentagewise improvement of the proposed al-gorithm is depicted in Figure 9 which shows that for dataset

xtff yt

ff forall SSn

Send to simulator for mappingFind updated xa and ya

Calculate Distt and StC

Insert

Resources

PD (Imax = END)

No

Yes

Input Wi = W1 W2WI Initialize SSn SPloc p and Imax

Tj = T1 T2 TJ forall Wi

VMK = VM1 VM2 VMp

Sw and Ew in Wi

QoS parameter St Et isin π f (obj) = ℷ lowast Ct + Ω lowast Ms + ℨ lowast sf R

Figure 5 Flowchart of the proposed IFFO algorithm

6 Mathematical Problems in Engineering

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 2: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

various client requests are received simultaneously Similartasks can be identified to be allocated on a similar set ofresources [3] Strategies must be adopted to enhance thesystemrsquos performance and ensure that all client requests arecompleted before the deadline

An efficient scheduling technique maintains a trade-offbetween user requirements and resource utilization [4]Maintaining this trade-off becomes challenging when sometasks have a parent-child relationship where a child task canonly begin executing once its parent task has finished and allthe output data from the predecessor task has been com-municated to the child task [5 6] Various list-based al-gorithms cannot be directly implemented in cloudcomputing environments because of the resource hetero-geneity varied QoS constraints and cloudrsquos dynamic nature[2] Several metaheuristic algorithms such as FFO [7] PSO[8] GA [9 10] have been explored for solving workflowscheduling problems However optimizing multiple ob-jectives is still a challenging task for the CCE [11ndash14]Optimizing one QoS parameter often results in compro-mising with the other QoS parameter (us schedulingmultiple workflows on the cloud while maintaining a trade-off among multiple QoS parameters remains a problem thatneeds to be solved

In this paper an enhanced metaheuristic optimizationtechnique IFFO has been proposed for scheduling multipleworkflows on cloud computing environments to optimizemultiple QoS parameters (e results of the proposedtechnique were generated using CloudSim and comparedwith existing FFO PSO and GA algorithms to validate theperformance of IFFO in terms of makespan and costparameters

(e rest of the paper is organized as follows Section 1describes the introductory concepts of cloud computingrelated to workflow scheduling A crisp and concise lit-erature survey on workflow scheduling for QoS param-eters is discussed in Section 2 Section 3 highlights theproblem formulation and problem definition A novel

framework using the IFFO algorithm is proposed inSection 4 (e proposed algorithmrsquos experimental resultsare given in Section 5 and the conclusion is mentioned inSection 6

2 Background

Yassa et al have presented a new multiobjective approachcalled DVFS-MODPSO [15] for scheduling workflows onthe cloud computing environment (e presented algorithmis a hybridization of PSO with Heterogeneous Earliest FinishTime (HEFT) aiming to optimize multiple objectives likemakespan cost and energy consumption Dynamic Voltageand Frequency Scaling (DVFS) is used for energy optimi-zation and the results show better Pareto optimal solutionsthan HEFT

CGA2 [16] is a technique proposed by Liu et al whichincludes an adaptive penalty function for schedulingdeadline constrained workflows in the cloud computingenvironment addressing the limitations of previouslyproposed evolutionary algorithms (e proposed algorithmprevents premature convergence unlike several existingstatic techniques by applying adaptive crossover and mu-tation probabilities and generates solutions that are able tomeet deadline constraints CGA2 is compared with tradi-tional algorithms such as PSO HEFT GA and Random todemonstrate better performance in terms of meetingdeadlines under strict constraints and reducing the overallworkflow execution cost

HSGA [17] is a GA-based hybrid workflow schedulingtechnique adopted by Delavar et al which utilizes theoptimization characteristics of Round Robin (RR) and BestFit (BF) scheduling algorithms Initially the proposedtechnique does the priority ranking of tasks based on theirdependencies and then the resource allocation is done byimplementing RR and BF for appropriate VM selection (eexperimental results depicted better performance of HSGAin terms of reducing makespan lowering failure rate and

Workflow Resource pool

Taskscheduler

T1

T2

T3 T4

T5 T6

T7

R1

R2

R3

Rn

Figure 1 Basic concept of workflow scheduling

2 Mathematical Problems in Engineering

balancing the load when compared with LAGA and NGAscheduling algorithms

CDMWS [18] is a dynamic optimization technique forscheduling multiple workflows on the cloud proposed byDelavar et al which aims at improving CPU utilizationreducing makespan and improving the makespan-deadlinemeeting ratio (e proposed technique is also divided intotwo stages (e first stage is responsible for estimating theexecution time for each task by considering workflowdeadline and task dependencies (e second stage is re-sponsible for dynamic VM allocation where VMs can reusefor tasks having similar requirements VM reusability isimplemented to lower power consumption and increaseresource utilization CDMWS is compared with two otheralgorithms EWSA and RR to verify its superiority

Another list-based heuristic MOWS [19] introduced byF Abazari et al adopts the greedy approach for prioritizingtasks and allocating appropriate resources to them (eproposed technique aims at improving the security of theoverall cloud architecture while maintaining the executiontime of workflows (e first phase of the algorithm designsthe solution based on task prioritization and assesses se-curity risk (e second part proposes an algorithm to dealwith various security threats while scheduling a workflow onthe cloud

GA-based multiobjective workflow scheduling algo-rithm MOGA [20] presented by Attiqa Rehman et al aimsat optimizing a diverse range of objectives includingmakespan budget resource utilization deadline and energyefficiency A gap search algorithmwas also introduced in thiswork that finds gaps in the schedule generated for a par-ticular workflow and fills them with independent tasks tomaximize resource utilization MOGA was compared withthree GA-based algorithms and one PSO-based algorithm(MOPSO) to validate its superiority Table 1 shows thesummary of the related works

3 Problem Formulation

Any large-scale application that needs to be deployed on acloud platform is generally represented in the form of aworkflow W (T E) A workflow can be pictorially repre-sented using a directed acyclic graph (DAG) where T T1T2 Tn denotes the set of tasks(e complete application isdivided into several dependent and independent subtasks(e tasks in a workflow are represented at different levelshaving a parent-child relationship where a child task cannotbegin execution until all its parent tasks have finished ex-ecution and all output data has been transferred to the childtask An edge Eij from Ti to Tj represents that Ti is the parentof Tj and exist a dependency between Ti and Tj

Consider a sample workflow depicted in Figure 2 (eentire application is divided into seven subtasks T1 T2 T7falling at five different levels that is Level 0 to Level 4 Atlevel 2 Tasks T3 and T4 are independent of each other sincethey are at the same level so they can be executed con-currently on different resources However their executioncan only begin once T2 at Level 1 has finished its executionand transferred all output data to T3 and T4 Similarly the

execution of T2 depends on its predecessor task T1 Since T1has no predecessor it will be the first task to be executedWorkflow execution can consider being completed once thelast task in the DAG T7 has finished its execution andgenerated its output

Resource heterogeneity will also be considered as pro-vided by any IaaS cloud provider Different types of VMs willbe available based on different configurations Any cloudservice provider who joins the cloud marketplace provides atwo-dimensional bid BVMi (PVMi CVMi) where PVMi rep-resents the processing capacity of the VMmeasured in termsof MIPS and CVMi is the cost of execution on that VM (epricing model is based on the current Amazon EC2 stan-dards where a full cycle consists of 60minutes and one extraminute will count for one complete cycle So if a resource isconsumed for 61 minutes the user will be charged for twocomplete cycles that is 120 minutes

Execution time ETlj of a task Tj on a resource VMl where

Tj T1 T2 T3 TJ forall j isin 1 2 3 J is calculatedusing equation (1) by considering the size STj

of task Tj interms of MIPS and the performance variation factor Pvarl

ofVMl introduced while adjusting the processing capacity of aVM Since a workflow involves task dependency the datatransfer time DTab from tasks Tj to Tk can be calculated usingequation (2) where DoutTj

is the amount of output datagenerated by Tjwhich is assumed to be known in advance foreach task and bw is the bandwidth between each VM Alsothe data transfer rate between two tasks scheduled on thesame resource will be zero Hence the total processing timePTl

jof each task Tj on a resource VMl is calculated usingequation (3) where e is the number of edges connected to aparent task Tj and Qe 0 if Tj and Tk are scheduled on thesame VM else 1

ETlj

STj

PVMllowast 1 minus Pvarl

1113872 11138731113872 1113873 (1)

DTjk DoutTj

bw (2)

PTlj ETl

j + 1113944e

1DTjk lowastQ⎛⎝ ⎞⎠ (3)

Similarly multiple workflow scheduling problems canalso be formulated Consider the diagram shown in Figure 3It consists of three small workflows all of which can becombined to form a single large workflow by addingDummystart and Dummyend as the starting and ending tasksrespectively (e execution time of both of these tasks will bezero as they are included only for merging the smallerworkflows

Optimal task scheduling and resource provisioning canbe done based on various objectives (is work focuses onoptimizing two scheduling objectives that is makespan andcost by finding a schedule S (Res Map Zct Zms) forscheduling workflow on cloud computing environmentwhere Res r1 r2 rc is the set of available resourcesMap depicts the task to resource mapping in the form

Mathematical Problems in Engineering 3

Mrc

tj (tj rc STtj

ETtj) for each task in the workflow which

means that a task tj is scheduled on resource rc and it willbegin execution at start time STtj

and will finish executionat the end time ETtj

Zct and Zms represent total executioncost and total execution time and can be calculated with thehelp of the following two equations respectively

Zct 1113944

|R|

c1CVMrclowast

LETrcminus LSTrc

1113872 1113873

α

⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦ (4)

Zms max ETtj1113882 1113883 (5)

where α denotes one cycle of the time unit for which theVM is charged LETrc

is the lease end time for resource rcand LSTrc

is the lease start time A sample schedule gen-erated for workflow shown in Figure 2 is depicted inFigure 4

Hence this paper works on finding an optimal scheduleS to minimize total execution cost and total execution timebased on the definitions given so far

4 Fruit Fly Optimization

Fruit Fly Optimization Algorithm [7] has been widelyadopted for solving global optimization problems because ofits simple structure and lesser number of parameters (ealgorithm is inspired by the food-finding ability of fruit flieswhere they can smell food even at a distance of 40 km Once

Table 1 (e summary of the related works

ProposedAlgorithm Objectives considered Algorithm type Workflow type Scheduling type

(SD) Tool used Year

DVFS-MODPSO [15] Makespan cost and energy HEFT+particle swarm

optimizationSimple andscientific Static CloudSim 2013

HSGA [17] Makespan failure rate and loadbalancing Genetic algorithm Complex Static + dynamic Simulator 2013

CGA2 [16] Constrained deadlines and cost Genetic algorithm Scientific mdash CloudSim 2016

CDMWS [18] CPU utilization makespandeadline List-based Scientific

(multiworkflow) Dynamic Simulator 2017

MOGA [20]Makespan budget deadlineenergy efficiency resource

utilizationGenetic algorithm Real-life Static Simulator 2018

MOWS [19] Execution time and security List-based greedyapproach

Scientific(multiworkflow) mdash WorkflowSim 2019

Level 0

Level 1

Level 2

Level 3

Level 4

T1

T2

T3 T4

T5 T6

T7

Figure 2 Sample workflow

Dummystart

Dummyend

Figure 3 Representation of multiple workflows as a singleworkflow

4 Mathematical Problems in Engineering

they get closer to the food source they use their sensitivevision for flying towards the food direction

FFO works in various phases as follows

(1) (e first phase is the initialization phase where thefruit flies are randomly distributed in the searchspace and their location (X_init Y_init) is initialized

(2) In the second phase each fruit fly is given somerandom direction and distance(X_init + RandomValue Y_init + RandomValue) tomove towards the food source

(3) Next the distance between each fruit fly and the foodlocation is estimated and the smell concentration iscalculated which is the reciprocal distance

(4) (e algorithm then goes into the fitness evaluationphase which is a function based on the smellconcentration

(5) (e maximum smell concentration of the individualfruit fly is retained and the swarm updates its po-sition to move in that direction

(6) (ese steps are iteratively repeated and the result ofeach iteration is compared with the previous one tocheck whether optimized results are obtained or not

FFO algorithm became popular because of its easy-to-implement structure and quick convergence However it isnot found suitable for complex optimization problems as itcould get trapped in the local optima at later evolution stagesand might not reach the global optima Also the conver-gence rate of the algorithm for complex optimization couldbe improved

Hence this paper presents an enhanced version of thetraditional FFO algorithm which could be implemented forcomplex optimization problems such as scheduling multipleworkflows in the cloud computing environment

41 Proposed Framework Figure 5 presents the proposedframework for IFFO Consider a cloud provider with a set ofvirtual machines VMl VM1 VM2 VM3 VML forall l isin1 2 3 L having some computational capacity In acloud computing scenario there are multiple workflowsWi W1 W2 W3 WI forall i isin 1 2 3 I thatneed to be scheduled with optimized QoS parametersKeeping this in mind multipleWi are merged and convertedinto a unified workflow Sw and Ew tasks are added at thestarting and ending position of Wi forall workflows Wi theremay be several tasks Tj T1 T2 T3 TJ forall j isin 1 2 3 J

Each resource VMl is available on-demand is accessiblefrom a shared pool of computing resources and has someQoS parameter associated with it For the present researchwork the authors have considered makespan and cost asQoS parameters Cost of running each VM isVMCm VMC1 VMC2 VMC3 VMCM forallm isin 1 23 M and makespan VMMs VMM1 VMM2 VMM3 VMMS forall s isin 1 2 3 S

Maximum cost and makespan are calculated by theaddition of cost and makespan of each task (e objectiveof the abovementioned problem is to minimize the costand makespan for executing the entire workflow (usthese multiple objective optimization problems try to findout the Pereto optimal solution in each iteration Once thecost and makespan are optimized while Tj ne TJ the IFFOalgorithm is applied on input values to find out the bestsmell function in each iteration Based on updated smellvalues the entire swarm population updates their smellconcentration and becomes ready for the next iterationWhen maximum iterations are completed or Paretooptimal solution is achieved all the tasks are sent forsimulation purposes

42 Proposed IFFO Algorithm (e proposed IFFO opti-mization algorithm is an enhanced version of the traditionalFruit Fly Optimization Algorithm (is algorithm is used tooptimize multiple objectives that is cost and makespan formultiple workflow scheduling in the cloud environment(eproposed algorithm continuously optimizes the old solutionusing the smell concentration function (is paper alsoshows an improvement in coverage rate by updating ap-propriate positions in each iteration

(e main steps of the proposed method are described asfollows

(1) Input constraints let n be the swarm size of fruit flypopulation the initial position of each fruit fly isSPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P forall p isin1 2 3 P Here each swarm particle repre-sents a possible solution moving towards a ran-dom direction with randomized distance Rv Asper traditional FFO the maximum iterationshould be 20ndash40 for current research work themaximum iteration (distance) of fruit fly move-ment is Q 20ndash40

R1

R2

R3

10 20 30 40 50 60 70 80

T1 T3 T5

T6 T7

T6 T7

Time

Reso

urce

s

VM Booting Time

VM Idle Time

VM Shutdown Time

Figure 4 Sample schedule

Mathematical Problems in Engineering 5

(2) Output constraints the target is to find out rank-1Pareto optimal solution for given input values Heref (P) S1 S2 S3 SN forall N isin 1 2 3 K is aset of solutions with lower bound LB and upperbound UB ZCtMs is the total cost and makespan ofworkflows that need to be minimized

(3) (is step defines the objective function f(obj) whereSf is the scaling factor R is a randomized functionand 1113954ℷ 1113954Ωamp 1113954Z are arbitrary constant that is1113954ℷ + 1113954Ω + 1113954Z 1

(4) For terminating condition (maximum iteration)Imax (t) started from 1 to T

(5) Calculate the initial position of each swam particlex

fft xα + Rv andy

fft yα + Rv where the initial

position (xfft y

fft )of each swarm particles isin SPloc

p SPloc 1 SPloc 2 SPloc 3 SPloc P and Rv is therandom variable ranging from 0 to 1

(6) Distance between individual swarm and food iscalculated by

(xfft )2 + (y

fft )2

1113969

and smell concen-tration by 1Distt

(7) Calculate f(SCt ) forall SSq where q 1 23 n that

is smell concentration of each individual fruit fly(8) Find out the mean of smell concentration F(Smellt)(9) Update the swarm particles position with updated

values of (xα yα) that is xα xα + xα lowastRv(0 1) +

xα lowastF(Smellt) and yα yα + yα lowastRv(0 1) + yα lowastF(Smellt) and go to step 3

(e algorithmic representation of these steps is men-tioned below (Algorithm 1)

5 Results and Discussion

51Dataset andSimulationSetup (e experimental analysiswas conducted using the CloudSim framework [21] thesimulation tool used for simulating cloud environments(eproposed algorithm was implemented for three differentdatasets and results were compared with three other met-aheuristic optimization techniques FFO GA and PSODatasets differ in terms of the number of tasks in a workflowand the number of resources available Although the cloudenvironment is considered to have an unlimited set of re-sources for arriving at an optimal solution we need to limitthe number of resources as well We have considered threesample workflows consisting of 15 25 and 35 tasks Forthese three workflows the number of resources is assumedto be 5 10 and 15 respectively

52 Performance Analysis (e proposed IFFO algorithmis compared with PSO GA and FFO based on twoscheduling objectives makespan and cost (e algo-rithms were executed for 20 iterations and the resultsdepict better performance of IFFO as compared with theother algorithms both in terms of makespan and cost(e experimental results are presented in the graphsshown in Figures 6ndash8 for datasets 1 2 and 3(e blue linerepresents the cost of execution while the orange linedepicts the makespan It is clear from these graphs thatIFFO outperforms PSO GA and FFO in bothparameters

(e percentagewise improvement of the proposed al-gorithm is depicted in Figure 9 which shows that for dataset

xtff yt

ff forall SSn

Send to simulator for mappingFind updated xa and ya

Calculate Distt and StC

Insert

Resources

PD (Imax = END)

No

Yes

Input Wi = W1 W2WI Initialize SSn SPloc p and Imax

Tj = T1 T2 TJ forall Wi

VMK = VM1 VM2 VMp

Sw and Ew in Wi

QoS parameter St Et isin π f (obj) = ℷ lowast Ct + Ω lowast Ms + ℨ lowast sf R

Figure 5 Flowchart of the proposed IFFO algorithm

6 Mathematical Problems in Engineering

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 3: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

balancing the load when compared with LAGA and NGAscheduling algorithms

CDMWS [18] is a dynamic optimization technique forscheduling multiple workflows on the cloud proposed byDelavar et al which aims at improving CPU utilizationreducing makespan and improving the makespan-deadlinemeeting ratio (e proposed technique is also divided intotwo stages (e first stage is responsible for estimating theexecution time for each task by considering workflowdeadline and task dependencies (e second stage is re-sponsible for dynamic VM allocation where VMs can reusefor tasks having similar requirements VM reusability isimplemented to lower power consumption and increaseresource utilization CDMWS is compared with two otheralgorithms EWSA and RR to verify its superiority

Another list-based heuristic MOWS [19] introduced byF Abazari et al adopts the greedy approach for prioritizingtasks and allocating appropriate resources to them (eproposed technique aims at improving the security of theoverall cloud architecture while maintaining the executiontime of workflows (e first phase of the algorithm designsthe solution based on task prioritization and assesses se-curity risk (e second part proposes an algorithm to dealwith various security threats while scheduling a workflow onthe cloud

GA-based multiobjective workflow scheduling algo-rithm MOGA [20] presented by Attiqa Rehman et al aimsat optimizing a diverse range of objectives includingmakespan budget resource utilization deadline and energyefficiency A gap search algorithmwas also introduced in thiswork that finds gaps in the schedule generated for a par-ticular workflow and fills them with independent tasks tomaximize resource utilization MOGA was compared withthree GA-based algorithms and one PSO-based algorithm(MOPSO) to validate its superiority Table 1 shows thesummary of the related works

3 Problem Formulation

Any large-scale application that needs to be deployed on acloud platform is generally represented in the form of aworkflow W (T E) A workflow can be pictorially repre-sented using a directed acyclic graph (DAG) where T T1T2 Tn denotes the set of tasks(e complete application isdivided into several dependent and independent subtasks(e tasks in a workflow are represented at different levelshaving a parent-child relationship where a child task cannotbegin execution until all its parent tasks have finished ex-ecution and all output data has been transferred to the childtask An edge Eij from Ti to Tj represents that Ti is the parentof Tj and exist a dependency between Ti and Tj

Consider a sample workflow depicted in Figure 2 (eentire application is divided into seven subtasks T1 T2 T7falling at five different levels that is Level 0 to Level 4 Atlevel 2 Tasks T3 and T4 are independent of each other sincethey are at the same level so they can be executed con-currently on different resources However their executioncan only begin once T2 at Level 1 has finished its executionand transferred all output data to T3 and T4 Similarly the

execution of T2 depends on its predecessor task T1 Since T1has no predecessor it will be the first task to be executedWorkflow execution can consider being completed once thelast task in the DAG T7 has finished its execution andgenerated its output

Resource heterogeneity will also be considered as pro-vided by any IaaS cloud provider Different types of VMs willbe available based on different configurations Any cloudservice provider who joins the cloud marketplace provides atwo-dimensional bid BVMi (PVMi CVMi) where PVMi rep-resents the processing capacity of the VMmeasured in termsof MIPS and CVMi is the cost of execution on that VM (epricing model is based on the current Amazon EC2 stan-dards where a full cycle consists of 60minutes and one extraminute will count for one complete cycle So if a resource isconsumed for 61 minutes the user will be charged for twocomplete cycles that is 120 minutes

Execution time ETlj of a task Tj on a resource VMl where

Tj T1 T2 T3 TJ forall j isin 1 2 3 J is calculatedusing equation (1) by considering the size STj

of task Tj interms of MIPS and the performance variation factor Pvarl

ofVMl introduced while adjusting the processing capacity of aVM Since a workflow involves task dependency the datatransfer time DTab from tasks Tj to Tk can be calculated usingequation (2) where DoutTj

is the amount of output datagenerated by Tjwhich is assumed to be known in advance foreach task and bw is the bandwidth between each VM Alsothe data transfer rate between two tasks scheduled on thesame resource will be zero Hence the total processing timePTl

jof each task Tj on a resource VMl is calculated usingequation (3) where e is the number of edges connected to aparent task Tj and Qe 0 if Tj and Tk are scheduled on thesame VM else 1

ETlj

STj

PVMllowast 1 minus Pvarl

1113872 11138731113872 1113873 (1)

DTjk DoutTj

bw (2)

PTlj ETl

j + 1113944e

1DTjk lowastQ⎛⎝ ⎞⎠ (3)

Similarly multiple workflow scheduling problems canalso be formulated Consider the diagram shown in Figure 3It consists of three small workflows all of which can becombined to form a single large workflow by addingDummystart and Dummyend as the starting and ending tasksrespectively (e execution time of both of these tasks will bezero as they are included only for merging the smallerworkflows

Optimal task scheduling and resource provisioning canbe done based on various objectives (is work focuses onoptimizing two scheduling objectives that is makespan andcost by finding a schedule S (Res Map Zct Zms) forscheduling workflow on cloud computing environmentwhere Res r1 r2 rc is the set of available resourcesMap depicts the task to resource mapping in the form

Mathematical Problems in Engineering 3

Mrc

tj (tj rc STtj

ETtj) for each task in the workflow which

means that a task tj is scheduled on resource rc and it willbegin execution at start time STtj

and will finish executionat the end time ETtj

Zct and Zms represent total executioncost and total execution time and can be calculated with thehelp of the following two equations respectively

Zct 1113944

|R|

c1CVMrclowast

LETrcminus LSTrc

1113872 1113873

α

⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦ (4)

Zms max ETtj1113882 1113883 (5)

where α denotes one cycle of the time unit for which theVM is charged LETrc

is the lease end time for resource rcand LSTrc

is the lease start time A sample schedule gen-erated for workflow shown in Figure 2 is depicted inFigure 4

Hence this paper works on finding an optimal scheduleS to minimize total execution cost and total execution timebased on the definitions given so far

4 Fruit Fly Optimization

Fruit Fly Optimization Algorithm [7] has been widelyadopted for solving global optimization problems because ofits simple structure and lesser number of parameters (ealgorithm is inspired by the food-finding ability of fruit flieswhere they can smell food even at a distance of 40 km Once

Table 1 (e summary of the related works

ProposedAlgorithm Objectives considered Algorithm type Workflow type Scheduling type

(SD) Tool used Year

DVFS-MODPSO [15] Makespan cost and energy HEFT+particle swarm

optimizationSimple andscientific Static CloudSim 2013

HSGA [17] Makespan failure rate and loadbalancing Genetic algorithm Complex Static + dynamic Simulator 2013

CGA2 [16] Constrained deadlines and cost Genetic algorithm Scientific mdash CloudSim 2016

CDMWS [18] CPU utilization makespandeadline List-based Scientific

(multiworkflow) Dynamic Simulator 2017

MOGA [20]Makespan budget deadlineenergy efficiency resource

utilizationGenetic algorithm Real-life Static Simulator 2018

MOWS [19] Execution time and security List-based greedyapproach

Scientific(multiworkflow) mdash WorkflowSim 2019

Level 0

Level 1

Level 2

Level 3

Level 4

T1

T2

T3 T4

T5 T6

T7

Figure 2 Sample workflow

Dummystart

Dummyend

Figure 3 Representation of multiple workflows as a singleworkflow

4 Mathematical Problems in Engineering

they get closer to the food source they use their sensitivevision for flying towards the food direction

FFO works in various phases as follows

(1) (e first phase is the initialization phase where thefruit flies are randomly distributed in the searchspace and their location (X_init Y_init) is initialized

(2) In the second phase each fruit fly is given somerandom direction and distance(X_init + RandomValue Y_init + RandomValue) tomove towards the food source

(3) Next the distance between each fruit fly and the foodlocation is estimated and the smell concentration iscalculated which is the reciprocal distance

(4) (e algorithm then goes into the fitness evaluationphase which is a function based on the smellconcentration

(5) (e maximum smell concentration of the individualfruit fly is retained and the swarm updates its po-sition to move in that direction

(6) (ese steps are iteratively repeated and the result ofeach iteration is compared with the previous one tocheck whether optimized results are obtained or not

FFO algorithm became popular because of its easy-to-implement structure and quick convergence However it isnot found suitable for complex optimization problems as itcould get trapped in the local optima at later evolution stagesand might not reach the global optima Also the conver-gence rate of the algorithm for complex optimization couldbe improved

Hence this paper presents an enhanced version of thetraditional FFO algorithm which could be implemented forcomplex optimization problems such as scheduling multipleworkflows in the cloud computing environment

41 Proposed Framework Figure 5 presents the proposedframework for IFFO Consider a cloud provider with a set ofvirtual machines VMl VM1 VM2 VM3 VML forall l isin1 2 3 L having some computational capacity In acloud computing scenario there are multiple workflowsWi W1 W2 W3 WI forall i isin 1 2 3 I thatneed to be scheduled with optimized QoS parametersKeeping this in mind multipleWi are merged and convertedinto a unified workflow Sw and Ew tasks are added at thestarting and ending position of Wi forall workflows Wi theremay be several tasks Tj T1 T2 T3 TJ forall j isin 1 2 3 J

Each resource VMl is available on-demand is accessiblefrom a shared pool of computing resources and has someQoS parameter associated with it For the present researchwork the authors have considered makespan and cost asQoS parameters Cost of running each VM isVMCm VMC1 VMC2 VMC3 VMCM forallm isin 1 23 M and makespan VMMs VMM1 VMM2 VMM3 VMMS forall s isin 1 2 3 S

Maximum cost and makespan are calculated by theaddition of cost and makespan of each task (e objectiveof the abovementioned problem is to minimize the costand makespan for executing the entire workflow (usthese multiple objective optimization problems try to findout the Pereto optimal solution in each iteration Once thecost and makespan are optimized while Tj ne TJ the IFFOalgorithm is applied on input values to find out the bestsmell function in each iteration Based on updated smellvalues the entire swarm population updates their smellconcentration and becomes ready for the next iterationWhen maximum iterations are completed or Paretooptimal solution is achieved all the tasks are sent forsimulation purposes

42 Proposed IFFO Algorithm (e proposed IFFO opti-mization algorithm is an enhanced version of the traditionalFruit Fly Optimization Algorithm (is algorithm is used tooptimize multiple objectives that is cost and makespan formultiple workflow scheduling in the cloud environment(eproposed algorithm continuously optimizes the old solutionusing the smell concentration function (is paper alsoshows an improvement in coverage rate by updating ap-propriate positions in each iteration

(e main steps of the proposed method are described asfollows

(1) Input constraints let n be the swarm size of fruit flypopulation the initial position of each fruit fly isSPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P forall p isin1 2 3 P Here each swarm particle repre-sents a possible solution moving towards a ran-dom direction with randomized distance Rv Asper traditional FFO the maximum iterationshould be 20ndash40 for current research work themaximum iteration (distance) of fruit fly move-ment is Q 20ndash40

R1

R2

R3

10 20 30 40 50 60 70 80

T1 T3 T5

T6 T7

T6 T7

Time

Reso

urce

s

VM Booting Time

VM Idle Time

VM Shutdown Time

Figure 4 Sample schedule

Mathematical Problems in Engineering 5

(2) Output constraints the target is to find out rank-1Pareto optimal solution for given input values Heref (P) S1 S2 S3 SN forall N isin 1 2 3 K is aset of solutions with lower bound LB and upperbound UB ZCtMs is the total cost and makespan ofworkflows that need to be minimized

(3) (is step defines the objective function f(obj) whereSf is the scaling factor R is a randomized functionand 1113954ℷ 1113954Ωamp 1113954Z are arbitrary constant that is1113954ℷ + 1113954Ω + 1113954Z 1

(4) For terminating condition (maximum iteration)Imax (t) started from 1 to T

(5) Calculate the initial position of each swam particlex

fft xα + Rv andy

fft yα + Rv where the initial

position (xfft y

fft )of each swarm particles isin SPloc

p SPloc 1 SPloc 2 SPloc 3 SPloc P and Rv is therandom variable ranging from 0 to 1

(6) Distance between individual swarm and food iscalculated by

(xfft )2 + (y

fft )2

1113969

and smell concen-tration by 1Distt

(7) Calculate f(SCt ) forall SSq where q 1 23 n that

is smell concentration of each individual fruit fly(8) Find out the mean of smell concentration F(Smellt)(9) Update the swarm particles position with updated

values of (xα yα) that is xα xα + xα lowastRv(0 1) +

xα lowastF(Smellt) and yα yα + yα lowastRv(0 1) + yα lowastF(Smellt) and go to step 3

(e algorithmic representation of these steps is men-tioned below (Algorithm 1)

5 Results and Discussion

51Dataset andSimulationSetup (e experimental analysiswas conducted using the CloudSim framework [21] thesimulation tool used for simulating cloud environments(eproposed algorithm was implemented for three differentdatasets and results were compared with three other met-aheuristic optimization techniques FFO GA and PSODatasets differ in terms of the number of tasks in a workflowand the number of resources available Although the cloudenvironment is considered to have an unlimited set of re-sources for arriving at an optimal solution we need to limitthe number of resources as well We have considered threesample workflows consisting of 15 25 and 35 tasks Forthese three workflows the number of resources is assumedto be 5 10 and 15 respectively

52 Performance Analysis (e proposed IFFO algorithmis compared with PSO GA and FFO based on twoscheduling objectives makespan and cost (e algo-rithms were executed for 20 iterations and the resultsdepict better performance of IFFO as compared with theother algorithms both in terms of makespan and cost(e experimental results are presented in the graphsshown in Figures 6ndash8 for datasets 1 2 and 3(e blue linerepresents the cost of execution while the orange linedepicts the makespan It is clear from these graphs thatIFFO outperforms PSO GA and FFO in bothparameters

(e percentagewise improvement of the proposed al-gorithm is depicted in Figure 9 which shows that for dataset

xtff yt

ff forall SSn

Send to simulator for mappingFind updated xa and ya

Calculate Distt and StC

Insert

Resources

PD (Imax = END)

No

Yes

Input Wi = W1 W2WI Initialize SSn SPloc p and Imax

Tj = T1 T2 TJ forall Wi

VMK = VM1 VM2 VMp

Sw and Ew in Wi

QoS parameter St Et isin π f (obj) = ℷ lowast Ct + Ω lowast Ms + ℨ lowast sf R

Figure 5 Flowchart of the proposed IFFO algorithm

6 Mathematical Problems in Engineering

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 4: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

Mrc

tj (tj rc STtj

ETtj) for each task in the workflow which

means that a task tj is scheduled on resource rc and it willbegin execution at start time STtj

and will finish executionat the end time ETtj

Zct and Zms represent total executioncost and total execution time and can be calculated with thehelp of the following two equations respectively

Zct 1113944

|R|

c1CVMrclowast

LETrcminus LSTrc

1113872 1113873

α

⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦ (4)

Zms max ETtj1113882 1113883 (5)

where α denotes one cycle of the time unit for which theVM is charged LETrc

is the lease end time for resource rcand LSTrc

is the lease start time A sample schedule gen-erated for workflow shown in Figure 2 is depicted inFigure 4

Hence this paper works on finding an optimal scheduleS to minimize total execution cost and total execution timebased on the definitions given so far

4 Fruit Fly Optimization

Fruit Fly Optimization Algorithm [7] has been widelyadopted for solving global optimization problems because ofits simple structure and lesser number of parameters (ealgorithm is inspired by the food-finding ability of fruit flieswhere they can smell food even at a distance of 40 km Once

Table 1 (e summary of the related works

ProposedAlgorithm Objectives considered Algorithm type Workflow type Scheduling type

(SD) Tool used Year

DVFS-MODPSO [15] Makespan cost and energy HEFT+particle swarm

optimizationSimple andscientific Static CloudSim 2013

HSGA [17] Makespan failure rate and loadbalancing Genetic algorithm Complex Static + dynamic Simulator 2013

CGA2 [16] Constrained deadlines and cost Genetic algorithm Scientific mdash CloudSim 2016

CDMWS [18] CPU utilization makespandeadline List-based Scientific

(multiworkflow) Dynamic Simulator 2017

MOGA [20]Makespan budget deadlineenergy efficiency resource

utilizationGenetic algorithm Real-life Static Simulator 2018

MOWS [19] Execution time and security List-based greedyapproach

Scientific(multiworkflow) mdash WorkflowSim 2019

Level 0

Level 1

Level 2

Level 3

Level 4

T1

T2

T3 T4

T5 T6

T7

Figure 2 Sample workflow

Dummystart

Dummyend

Figure 3 Representation of multiple workflows as a singleworkflow

4 Mathematical Problems in Engineering

they get closer to the food source they use their sensitivevision for flying towards the food direction

FFO works in various phases as follows

(1) (e first phase is the initialization phase where thefruit flies are randomly distributed in the searchspace and their location (X_init Y_init) is initialized

(2) In the second phase each fruit fly is given somerandom direction and distance(X_init + RandomValue Y_init + RandomValue) tomove towards the food source

(3) Next the distance between each fruit fly and the foodlocation is estimated and the smell concentration iscalculated which is the reciprocal distance

(4) (e algorithm then goes into the fitness evaluationphase which is a function based on the smellconcentration

(5) (e maximum smell concentration of the individualfruit fly is retained and the swarm updates its po-sition to move in that direction

(6) (ese steps are iteratively repeated and the result ofeach iteration is compared with the previous one tocheck whether optimized results are obtained or not

FFO algorithm became popular because of its easy-to-implement structure and quick convergence However it isnot found suitable for complex optimization problems as itcould get trapped in the local optima at later evolution stagesand might not reach the global optima Also the conver-gence rate of the algorithm for complex optimization couldbe improved

Hence this paper presents an enhanced version of thetraditional FFO algorithm which could be implemented forcomplex optimization problems such as scheduling multipleworkflows in the cloud computing environment

41 Proposed Framework Figure 5 presents the proposedframework for IFFO Consider a cloud provider with a set ofvirtual machines VMl VM1 VM2 VM3 VML forall l isin1 2 3 L having some computational capacity In acloud computing scenario there are multiple workflowsWi W1 W2 W3 WI forall i isin 1 2 3 I thatneed to be scheduled with optimized QoS parametersKeeping this in mind multipleWi are merged and convertedinto a unified workflow Sw and Ew tasks are added at thestarting and ending position of Wi forall workflows Wi theremay be several tasks Tj T1 T2 T3 TJ forall j isin 1 2 3 J

Each resource VMl is available on-demand is accessiblefrom a shared pool of computing resources and has someQoS parameter associated with it For the present researchwork the authors have considered makespan and cost asQoS parameters Cost of running each VM isVMCm VMC1 VMC2 VMC3 VMCM forallm isin 1 23 M and makespan VMMs VMM1 VMM2 VMM3 VMMS forall s isin 1 2 3 S

Maximum cost and makespan are calculated by theaddition of cost and makespan of each task (e objectiveof the abovementioned problem is to minimize the costand makespan for executing the entire workflow (usthese multiple objective optimization problems try to findout the Pereto optimal solution in each iteration Once thecost and makespan are optimized while Tj ne TJ the IFFOalgorithm is applied on input values to find out the bestsmell function in each iteration Based on updated smellvalues the entire swarm population updates their smellconcentration and becomes ready for the next iterationWhen maximum iterations are completed or Paretooptimal solution is achieved all the tasks are sent forsimulation purposes

42 Proposed IFFO Algorithm (e proposed IFFO opti-mization algorithm is an enhanced version of the traditionalFruit Fly Optimization Algorithm (is algorithm is used tooptimize multiple objectives that is cost and makespan formultiple workflow scheduling in the cloud environment(eproposed algorithm continuously optimizes the old solutionusing the smell concentration function (is paper alsoshows an improvement in coverage rate by updating ap-propriate positions in each iteration

(e main steps of the proposed method are described asfollows

(1) Input constraints let n be the swarm size of fruit flypopulation the initial position of each fruit fly isSPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P forall p isin1 2 3 P Here each swarm particle repre-sents a possible solution moving towards a ran-dom direction with randomized distance Rv Asper traditional FFO the maximum iterationshould be 20ndash40 for current research work themaximum iteration (distance) of fruit fly move-ment is Q 20ndash40

R1

R2

R3

10 20 30 40 50 60 70 80

T1 T3 T5

T6 T7

T6 T7

Time

Reso

urce

s

VM Booting Time

VM Idle Time

VM Shutdown Time

Figure 4 Sample schedule

Mathematical Problems in Engineering 5

(2) Output constraints the target is to find out rank-1Pareto optimal solution for given input values Heref (P) S1 S2 S3 SN forall N isin 1 2 3 K is aset of solutions with lower bound LB and upperbound UB ZCtMs is the total cost and makespan ofworkflows that need to be minimized

(3) (is step defines the objective function f(obj) whereSf is the scaling factor R is a randomized functionand 1113954ℷ 1113954Ωamp 1113954Z are arbitrary constant that is1113954ℷ + 1113954Ω + 1113954Z 1

(4) For terminating condition (maximum iteration)Imax (t) started from 1 to T

(5) Calculate the initial position of each swam particlex

fft xα + Rv andy

fft yα + Rv where the initial

position (xfft y

fft )of each swarm particles isin SPloc

p SPloc 1 SPloc 2 SPloc 3 SPloc P and Rv is therandom variable ranging from 0 to 1

(6) Distance between individual swarm and food iscalculated by

(xfft )2 + (y

fft )2

1113969

and smell concen-tration by 1Distt

(7) Calculate f(SCt ) forall SSq where q 1 23 n that

is smell concentration of each individual fruit fly(8) Find out the mean of smell concentration F(Smellt)(9) Update the swarm particles position with updated

values of (xα yα) that is xα xα + xα lowastRv(0 1) +

xα lowastF(Smellt) and yα yα + yα lowastRv(0 1) + yα lowastF(Smellt) and go to step 3

(e algorithmic representation of these steps is men-tioned below (Algorithm 1)

5 Results and Discussion

51Dataset andSimulationSetup (e experimental analysiswas conducted using the CloudSim framework [21] thesimulation tool used for simulating cloud environments(eproposed algorithm was implemented for three differentdatasets and results were compared with three other met-aheuristic optimization techniques FFO GA and PSODatasets differ in terms of the number of tasks in a workflowand the number of resources available Although the cloudenvironment is considered to have an unlimited set of re-sources for arriving at an optimal solution we need to limitthe number of resources as well We have considered threesample workflows consisting of 15 25 and 35 tasks Forthese three workflows the number of resources is assumedto be 5 10 and 15 respectively

52 Performance Analysis (e proposed IFFO algorithmis compared with PSO GA and FFO based on twoscheduling objectives makespan and cost (e algo-rithms were executed for 20 iterations and the resultsdepict better performance of IFFO as compared with theother algorithms both in terms of makespan and cost(e experimental results are presented in the graphsshown in Figures 6ndash8 for datasets 1 2 and 3(e blue linerepresents the cost of execution while the orange linedepicts the makespan It is clear from these graphs thatIFFO outperforms PSO GA and FFO in bothparameters

(e percentagewise improvement of the proposed al-gorithm is depicted in Figure 9 which shows that for dataset

xtff yt

ff forall SSn

Send to simulator for mappingFind updated xa and ya

Calculate Distt and StC

Insert

Resources

PD (Imax = END)

No

Yes

Input Wi = W1 W2WI Initialize SSn SPloc p and Imax

Tj = T1 T2 TJ forall Wi

VMK = VM1 VM2 VMp

Sw and Ew in Wi

QoS parameter St Et isin π f (obj) = ℷ lowast Ct + Ω lowast Ms + ℨ lowast sf R

Figure 5 Flowchart of the proposed IFFO algorithm

6 Mathematical Problems in Engineering

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 5: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

they get closer to the food source they use their sensitivevision for flying towards the food direction

FFO works in various phases as follows

(1) (e first phase is the initialization phase where thefruit flies are randomly distributed in the searchspace and their location (X_init Y_init) is initialized

(2) In the second phase each fruit fly is given somerandom direction and distance(X_init + RandomValue Y_init + RandomValue) tomove towards the food source

(3) Next the distance between each fruit fly and the foodlocation is estimated and the smell concentration iscalculated which is the reciprocal distance

(4) (e algorithm then goes into the fitness evaluationphase which is a function based on the smellconcentration

(5) (e maximum smell concentration of the individualfruit fly is retained and the swarm updates its po-sition to move in that direction

(6) (ese steps are iteratively repeated and the result ofeach iteration is compared with the previous one tocheck whether optimized results are obtained or not

FFO algorithm became popular because of its easy-to-implement structure and quick convergence However it isnot found suitable for complex optimization problems as itcould get trapped in the local optima at later evolution stagesand might not reach the global optima Also the conver-gence rate of the algorithm for complex optimization couldbe improved

Hence this paper presents an enhanced version of thetraditional FFO algorithm which could be implemented forcomplex optimization problems such as scheduling multipleworkflows in the cloud computing environment

41 Proposed Framework Figure 5 presents the proposedframework for IFFO Consider a cloud provider with a set ofvirtual machines VMl VM1 VM2 VM3 VML forall l isin1 2 3 L having some computational capacity In acloud computing scenario there are multiple workflowsWi W1 W2 W3 WI forall i isin 1 2 3 I thatneed to be scheduled with optimized QoS parametersKeeping this in mind multipleWi are merged and convertedinto a unified workflow Sw and Ew tasks are added at thestarting and ending position of Wi forall workflows Wi theremay be several tasks Tj T1 T2 T3 TJ forall j isin 1 2 3 J

Each resource VMl is available on-demand is accessiblefrom a shared pool of computing resources and has someQoS parameter associated with it For the present researchwork the authors have considered makespan and cost asQoS parameters Cost of running each VM isVMCm VMC1 VMC2 VMC3 VMCM forallm isin 1 23 M and makespan VMMs VMM1 VMM2 VMM3 VMMS forall s isin 1 2 3 S

Maximum cost and makespan are calculated by theaddition of cost and makespan of each task (e objectiveof the abovementioned problem is to minimize the costand makespan for executing the entire workflow (usthese multiple objective optimization problems try to findout the Pereto optimal solution in each iteration Once thecost and makespan are optimized while Tj ne TJ the IFFOalgorithm is applied on input values to find out the bestsmell function in each iteration Based on updated smellvalues the entire swarm population updates their smellconcentration and becomes ready for the next iterationWhen maximum iterations are completed or Paretooptimal solution is achieved all the tasks are sent forsimulation purposes

42 Proposed IFFO Algorithm (e proposed IFFO opti-mization algorithm is an enhanced version of the traditionalFruit Fly Optimization Algorithm (is algorithm is used tooptimize multiple objectives that is cost and makespan formultiple workflow scheduling in the cloud environment(eproposed algorithm continuously optimizes the old solutionusing the smell concentration function (is paper alsoshows an improvement in coverage rate by updating ap-propriate positions in each iteration

(e main steps of the proposed method are described asfollows

(1) Input constraints let n be the swarm size of fruit flypopulation the initial position of each fruit fly isSPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P forall p isin1 2 3 P Here each swarm particle repre-sents a possible solution moving towards a ran-dom direction with randomized distance Rv Asper traditional FFO the maximum iterationshould be 20ndash40 for current research work themaximum iteration (distance) of fruit fly move-ment is Q 20ndash40

R1

R2

R3

10 20 30 40 50 60 70 80

T1 T3 T5

T6 T7

T6 T7

Time

Reso

urce

s

VM Booting Time

VM Idle Time

VM Shutdown Time

Figure 4 Sample schedule

Mathematical Problems in Engineering 5

(2) Output constraints the target is to find out rank-1Pareto optimal solution for given input values Heref (P) S1 S2 S3 SN forall N isin 1 2 3 K is aset of solutions with lower bound LB and upperbound UB ZCtMs is the total cost and makespan ofworkflows that need to be minimized

(3) (is step defines the objective function f(obj) whereSf is the scaling factor R is a randomized functionand 1113954ℷ 1113954Ωamp 1113954Z are arbitrary constant that is1113954ℷ + 1113954Ω + 1113954Z 1

(4) For terminating condition (maximum iteration)Imax (t) started from 1 to T

(5) Calculate the initial position of each swam particlex

fft xα + Rv andy

fft yα + Rv where the initial

position (xfft y

fft )of each swarm particles isin SPloc

p SPloc 1 SPloc 2 SPloc 3 SPloc P and Rv is therandom variable ranging from 0 to 1

(6) Distance between individual swarm and food iscalculated by

(xfft )2 + (y

fft )2

1113969

and smell concen-tration by 1Distt

(7) Calculate f(SCt ) forall SSq where q 1 23 n that

is smell concentration of each individual fruit fly(8) Find out the mean of smell concentration F(Smellt)(9) Update the swarm particles position with updated

values of (xα yα) that is xα xα + xα lowastRv(0 1) +

xα lowastF(Smellt) and yα yα + yα lowastRv(0 1) + yα lowastF(Smellt) and go to step 3

(e algorithmic representation of these steps is men-tioned below (Algorithm 1)

5 Results and Discussion

51Dataset andSimulationSetup (e experimental analysiswas conducted using the CloudSim framework [21] thesimulation tool used for simulating cloud environments(eproposed algorithm was implemented for three differentdatasets and results were compared with three other met-aheuristic optimization techniques FFO GA and PSODatasets differ in terms of the number of tasks in a workflowand the number of resources available Although the cloudenvironment is considered to have an unlimited set of re-sources for arriving at an optimal solution we need to limitthe number of resources as well We have considered threesample workflows consisting of 15 25 and 35 tasks Forthese three workflows the number of resources is assumedto be 5 10 and 15 respectively

52 Performance Analysis (e proposed IFFO algorithmis compared with PSO GA and FFO based on twoscheduling objectives makespan and cost (e algo-rithms were executed for 20 iterations and the resultsdepict better performance of IFFO as compared with theother algorithms both in terms of makespan and cost(e experimental results are presented in the graphsshown in Figures 6ndash8 for datasets 1 2 and 3(e blue linerepresents the cost of execution while the orange linedepicts the makespan It is clear from these graphs thatIFFO outperforms PSO GA and FFO in bothparameters

(e percentagewise improvement of the proposed al-gorithm is depicted in Figure 9 which shows that for dataset

xtff yt

ff forall SSn

Send to simulator for mappingFind updated xa and ya

Calculate Distt and StC

Insert

Resources

PD (Imax = END)

No

Yes

Input Wi = W1 W2WI Initialize SSn SPloc p and Imax

Tj = T1 T2 TJ forall Wi

VMK = VM1 VM2 VMp

Sw and Ew in Wi

QoS parameter St Et isin π f (obj) = ℷ lowast Ct + Ω lowast Ms + ℨ lowast sf R

Figure 5 Flowchart of the proposed IFFO algorithm

6 Mathematical Problems in Engineering

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 6: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

(2) Output constraints the target is to find out rank-1Pareto optimal solution for given input values Heref (P) S1 S2 S3 SN forall N isin 1 2 3 K is aset of solutions with lower bound LB and upperbound UB ZCtMs is the total cost and makespan ofworkflows that need to be minimized

(3) (is step defines the objective function f(obj) whereSf is the scaling factor R is a randomized functionand 1113954ℷ 1113954Ωamp 1113954Z are arbitrary constant that is1113954ℷ + 1113954Ω + 1113954Z 1

(4) For terminating condition (maximum iteration)Imax (t) started from 1 to T

(5) Calculate the initial position of each swam particlex

fft xα + Rv andy

fft yα + Rv where the initial

position (xfft y

fft )of each swarm particles isin SPloc

p SPloc 1 SPloc 2 SPloc 3 SPloc P and Rv is therandom variable ranging from 0 to 1

(6) Distance between individual swarm and food iscalculated by

(xfft )2 + (y

fft )2

1113969

and smell concen-tration by 1Distt

(7) Calculate f(SCt ) forall SSq where q 1 23 n that

is smell concentration of each individual fruit fly(8) Find out the mean of smell concentration F(Smellt)(9) Update the swarm particles position with updated

values of (xα yα) that is xα xα + xα lowastRv(0 1) +

xα lowastF(Smellt) and yα yα + yα lowastRv(0 1) + yα lowastF(Smellt) and go to step 3

(e algorithmic representation of these steps is men-tioned below (Algorithm 1)

5 Results and Discussion

51Dataset andSimulationSetup (e experimental analysiswas conducted using the CloudSim framework [21] thesimulation tool used for simulating cloud environments(eproposed algorithm was implemented for three differentdatasets and results were compared with three other met-aheuristic optimization techniques FFO GA and PSODatasets differ in terms of the number of tasks in a workflowand the number of resources available Although the cloudenvironment is considered to have an unlimited set of re-sources for arriving at an optimal solution we need to limitthe number of resources as well We have considered threesample workflows consisting of 15 25 and 35 tasks Forthese three workflows the number of resources is assumedto be 5 10 and 15 respectively

52 Performance Analysis (e proposed IFFO algorithmis compared with PSO GA and FFO based on twoscheduling objectives makespan and cost (e algo-rithms were executed for 20 iterations and the resultsdepict better performance of IFFO as compared with theother algorithms both in terms of makespan and cost(e experimental results are presented in the graphsshown in Figures 6ndash8 for datasets 1 2 and 3(e blue linerepresents the cost of execution while the orange linedepicts the makespan It is clear from these graphs thatIFFO outperforms PSO GA and FFO in bothparameters

(e percentagewise improvement of the proposed al-gorithm is depicted in Figure 9 which shows that for dataset

xtff yt

ff forall SSn

Send to simulator for mappingFind updated xa and ya

Calculate Distt and StC

Insert

Resources

PD (Imax = END)

No

Yes

Input Wi = W1 W2WI Initialize SSn SPloc p and Imax

Tj = T1 T2 TJ forall Wi

VMK = VM1 VM2 VMp

Sw and Ew in Wi

QoS parameter St Et isin π f (obj) = ℷ lowast Ct + Ω lowast Ms + ℨ lowast sf R

Figure 5 Flowchart of the proposed IFFO algorithm

6 Mathematical Problems in Engineering

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 7: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

(1) Input SSn SPloc p SPloc 1 SPloc 2 SPloc 3 SPloc P and Imax 20-40 forall p isin 1 2 3 PSSn Swarn Size SPloc initial location of individual swarm particles and Imax Maximum number of iteration

(2) Output Pareto optimal solutionminSNisin[LBNUBN ]N1 23K

f(P) S1 S2 S3 SN1113864 1113865

there4QoS 1113936KC1 1113936

JM1 ZCtMs and Outmin min(QoS)

SN are existing solutions ZCtMs is total cost ampmakespan of multiple workflows andOutmin is expected QoS optimized solution(3) f(obj) 1113954ℷ lowastCt + 1113954ΩlowastMs + 1113954Zlowast sfR

sf is scaling factor 1113954ℷ + 1113954Ω + 1113954Z 1 and R is a randomized function(4) for Imax (t) larr 1 to T do

(5) xfft xα + Rv andy

fft yα + Rv

(xfft y

fft )initial position of each swarm particle and Rv (01)

(6) Distt

(xfft )2 + (y

fft )2

1113969

and SCt 1Distt

Distt is distance between individual fruit fly and food and SCt is smell concentration

(7) Smellt f(SCt ) for each individual fruit fly

(8) F(Smellt) 1ω 1113936ωt1 ft(Smellt)

(9) Update swarm particles location(xα yα)

91 xα xα + xα lowastRv(0 1) + xα lowastF(Smellt)

92 yα yα + yα lowastRv(0 1) + yα lowastF(Smellt)

93 Go to step 3(10) End for

ALGORITHM 1 IFFOndashQoS optimization for multiple workflow scheduling

5034999505 5584805905 51429995054604803295

51157764

49004644

48277764

47782322

0

2000

4000

6000

8000

10000

12000

PSO GA FFO IFFO

CostMakespan

Dataset 1 (15lowast5)

Figure 6 Cost and makespan analysis for dataset 1

CostMakespan

5098304804 5631287557 5223133874

4628862224

1559817612 1596881181 15881933251446707195

020000400006000080000

100000120000140000160000180000

PSO GA FFO IFFO

Dataset 2 (25lowast10)

Figure 7 Cost and makespan analysis for dataset 2

Mathematical Problems in Engineering 7

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 8: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

1 IFFO is 854 1755 and 1046 better than PSO GAand FFO respectively in terms of cost and 66 249 and103 better than PSO GA and FFO respectively in termsof makespan For dataset 2 the improvement percentage is921 178 and 1138 in terms of cost and 725 94and 891 in terms of makespan when compared with PSOGA and FFO resp Similarly for dataset 3 IFFO showed animprovement of 1124 1934 and 1498 in terms ofcost and 961 1368 and 1935 in terms of makespanwhen compared with PSO GA and FFO respectively

(e proposed algorithm is capable of optimizing boththe parameters simultaneously unlike many other optimi-zation algorithms where the client has to compromise withone objective while trying to optimize the other In suchcases a decision has to be made regarding which objective isto be given preference over the other

6 Conclusion

Scientific workflows play a significant role in large-scalecloud-based applications In workflow scheduling nature-inspired algorithms elucidate the promising optimized

results for multiobjective problems in the cloud environ-ment But to avoid local optima trapping problems inmultiobjective optimization traditional nature-inspiredtechniques continuously try to maintain a balance betweenexploration and exploitation In this paper multipleworkflows are considered andmerged with dummy start andend nodes to represent it as a single monolithic workflow(e proposed IFFO enhanced the traditional FFO algorithmto minimize the ldquostuck at the local optimardquo problem by usingan enhanced swarm smell function (e activation functionused the mean smell function for the generation of newpositions of the swarm particles (e IFFO is used forscheduling multiple workflows to minimize cost andmakespan parameters while providing a Pareto optimalsolution (e proposed algorithm is implemented on theCloudSim platform and the result for dataset 1 shows thatIFFO is better than PSO GA and FFO by 1514 2004and 1147 respectively in terms of cost and makespanconjointly Similarly for dataset 2 the proposed algorithmshows 1646 272 and 2029 improvement And fordataset 3 the improvement is 2085 3302 and 3433 ascompared with PSO GA and FFO

5298911635

5831028756 5532031635

4703179592

1858600563 1946361435 2083080563 1680080439

0100000200000300000400000500000600000700000

PSO GA FFO IFFO

Dataset 3 (35lowast15)

CostMakespan

Figure 8 Cost and makespan analysis for dataset 3

85466

921725

1124961

1755

249

178

94

1934

1368

1046

103

1138891

1498

1935

0

5

10

15

20

25

Cost () Makespan ()Dataset 1

Cost () Makespan ()Dataset 2

Cost () Makespan ()Dataset 3

Performance analysis IFFO

PSOGAFFO

Figure 9 Comparative performance analysis of IFFO with respect to PSO GA and FFO

8 Mathematical Problems in Engineering

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9

Page 9: IFFO:AnImprovedFruitFlyOptimizationAlgorithmforMultiple ... · 2021. 4. 25. · technique IFFO has been proposed for scheduling multiple workflows on cloud computing environments

(e future scope is to implement the proposed IFFOtechnique with more QoS parameters such as energy effi-ciency and load balancing to enhance the overall systemperformance (e IFFO can be applied in various state-of-the-art research areas like sensor networks IoT decision-making system smart agriculture and ecological engi-neering problem

Data Availability

All data are included within this manuscript

Conflicts of Interest

(e authors declare that they have no conflicts of interest toreport regarding the present study

Acknowledgments

(e authors would like to acknowledge the support from TaifUniversity Researchers Supporting Project (no TURSP-2020216) Taif University Taif Saudi Arabia

References

[1] B P Rimal A Jukan D Katsaros and Y Goeleven ldquoAr-chitectural requirements for cloud computing systems anenterprise cloud approachrdquo Journal of Grid Computing vol 9no 1 pp 3ndash26 2010

[2] A Aggarwal P Dimri and A Agarwal ldquoSurvey on schedulingalgorithms for multiple workflows in cloud computing en-vironmentrdquo International Journal of Computer Sciences andEngineering vol 7 no 6 pp 565ndash570 2019

[3] X Zhou G Zhang J Sun J Zhou T Wei and S HuldquoMinimizing cost and makespan for workflow scheduling incloud using fuzzy dominance sort based HEFTrdquo FutureGeneration Computer Systems vol 93 pp 278ndash289 2019

[4] A Aggarwal P Dimri A Agarwal and A Bhatt ldquoSelfadaptive fruit fly algorithm for multiple workflow schedulingin cloud computing environmentrdquo Kybernetes vol 24 2020

[5] M A Rodriguez and R Buyya ldquoDeadline based resurceprovisioning and scheduling algorithm for scientific work-flows on cloudsrdquo IEEE Transactions on Cloud Computingvol 2 no 2 pp 222ndash235 2014

[6] Z Zhu G Zhang M Li and X Liu ldquoEvolutionary multi-objective workflow scheduling in cloudrdquo IEEE Transactionson Parallel and Distributed Systems vol 27 no 5pp 1344ndash1357 2016

[7] W-T Pan ldquoA new fruit fly optimization algorithm taking thefinancial distress model as an examplerdquo Knowledge-BasedSystems vol 26 pp 69ndash74 2011

[8] I Trelea ldquo(e particle swarm optimization algorithm con-vergence analysis and parameter selectionrdquo InformationProcessing Letters vol 85 pp 317ndash325 2002

[9] J McCall ldquoGenetic algorithms for modelling and optimisa-tionrdquo Journal of Computational and Applied Mathematicsvol 184 no 1 pp 205ndash222 2005

[10] A Gupta D Singh and M Kaur ldquoAn efficient image en-cryption using non-dominated sorting genetic algorithm-IIIbased 4-D chaotic mapsrdquo Journal of Ambient Intelligence andHumanized Computing vol 11 no 3 pp 1309ndash1324 2020

[11] M Kaur and D Singh ldquoMulti-modality medical image fusiontechnique using multi-objective differential evolution based

deep neural networksrdquo Journal of Ambient Intelligence andHumanized Computing vol 12 no 2 pp 2483ndash2493 2021

[12] D Rani and R Ranjan ldquoA comparative study of SaaS PaaSand IaaS in cloud computingrdquo International Journal of Ad-vanced Research in Computer Science and Software Engi-neering vol 4 no 6 pp 158ndash161 2014

[13] A Bhatt P Dimri and A Aggarwal ldquoSelf-adaptive brain-storming for jobshop scheduling in multicloud environmentrdquoSoftware Praactice and Experience vol 54 pp 1ndash18 2020

[14] H Hua X Guangquan P Shanchen and Z ZenghualdquoAdaptive multi-objective task scheduling strategy in cloudcomputingrdquo Strategies and Schemes vol 13 pp 162ndash1712016

[15] S Yassa R Chelouah H Kadima and B Granado ldquoMulti-objective approach for energy-aware workflow scheduling incloud computing environmentsrdquo Ee Scientific World Jour-nal vol 2013 Article ID 350934 13 pages 2013

[16] L Liu M Zhang R Buyya and Q Fan ldquoDeadline-con-strained coevolutionary genetic algorithm for scientificworkflow scheduling in cloud computingrdquo Concurrency andComputation Practice and Experience vol 29 pp 1ndash12 2016

[17] A G Delavar and Y Aryan ldquoA hybrid heuristic algorithm forworkflow scheduling in cloud systemsrdquo Cluster Computingvol 17 pp 129ndash137 2013

[18] M Adhikari and S Koley ldquoCloud Computing A multi-workflow scheduling algorithm with dynamic reusabilityrdquoArabian Journal for Science and Engineering (AJSE) vol 43pp 645ndash660 2017

[19] F Abazari M Analoui H Takabi and S Fu ldquoMulti-objectiveworkflow scheduling in cloud computing based on heuristicalgorithmrdquo Simulation Modelling Practice andEeory vol 93pp 119ndash132 2018

[20] A Rehman S S Hussain Z U Rehman S Zia andS Shamshirband ldquoMulti-objective approach of energy effi-cient workflow scheduling in cloud environmentsrdquo Concur-rency Computat Pract Exper vol 32 pp 1ndash20 2018

[21] R N Calheiros R Ranjan A Beloglazov C A F De Roseand R Buyya ldquoA toolkit for modeling and simulation of cloudcomputing environments and evaluation of resource provi-sioning algorithmsrdquo Software Practice and Experience vol 41no 1 pp 23ndash50 2011

Mathematical Problems in Engineering 9


Related Documents