Research Article Modeling, Design, and Implementation of a ...

Research ArticleModeling Design and Implementation of a Cloud WorkflowEngine Based on Aneka

Jiantao Zhou1 Chaoxin Sun1 Weina Fu1 Jing Liu1 Lei Jia1 and Hongyan Tan2

1 College of Computer Science Inner Mongolia University Hohhot 010021 China2High Performance Network Lab Institute of Acoustics Chinese Academy of Sciences Beijing 100190 China

Correspondence should be addressed to Jiantao Zhou zhoujiantaotsinghuaorgcn

Received 28 January 2014 Accepted 11 March 2014 Published 29 April 2014

Academic Editor X Song

Copyright copy 2014 Jiantao Zhou et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

This paper presents a Petri net-based model for cloud workflow which plays a key role in industry Three kinds of parallelisms incloud workflow are characterized and modeled Based on the analysis of the modeling a cloud workflow engine is designed andimplemented in Aneka cloud environmentThe experimental results validate the effectiveness of our approach of modeling designand implementation of cloud workflow

1 Introduction

With the successful cases of the worldrsquos leading compa-nies for example Amazon and Google cloud computinghas become a hot topic in both industrial and academicareas It embraces WEB 20 middleware virtualization andother technologies and also develops upon grid comput-ing distributed computing parallel computing and utilitycomputing and so forth [1] Comparing with classic com-puting paradigms cloud computing provides ldquoa pool ofabstracted virtualized dynamically scalable managed com-puting power storage platforms and services are deliveredon demand to external customers over the Internetrdquo [2] Itcan provide scalable resources conveniently on demand todifferent system requirements According to features of ser-vices mainly delivered by the established cloud infrastruc-tures researchers separated the services into three levelswhich are infrastructure as a service (IaaS) platform asa service (PaaS) and software as a service (SaaS) IaaSoffers hardware resources and computing power such asAmazon S3 for storage and EC2 for computing power PaaStargets providing entire facilities including hardware andthe application development environment such as MicrosoftAzure Services platform and Google App Engine SaaS refersto those software applications offered as services in cloudenvironments

However along with the development of cloud comput-ing corresponding issues are also arising in both theoreticaland technical aspects One of the most prominent problemsis how to minimize running costs and maximize revenueson the premise of maintaining or even improving the qualityof service (QoS) [3] Workflow technology can be regardedas one of the solutions [2ndash4] A workflow is defined as ldquotheautomation of a business process in whole or part duringwhich documents information or tasks are passed from oneparticipant to another for action according to a set of proce-dural rulesrdquo by the workflowmanagement coalition (WFMC)[5] Workflow management system (WFMS) is a system fordefining implementing and managing the workflows inwhich workflow engine is themost significant component fortask scheduling data movement and exception handling

Cloud workflow also called cloud-based workflow [1] orcloud computing-oriented workflow [3] is a new applicationmode of workflow management system in the cloud envi-ronment whose goal is to optimize the system performanceguarantee the QoS and reduce running cost It integratesworkflowwith cloud computing which combined the advan-tages of both sides Cloud workflow technology can be usedin two levels [3] from the cloud usersrsquo view it supportsprocess definitions of cloud applications and enables flexibleconfiguration and automated operation of the processesThiskind of cloudworkflow is regarded as ldquoabove-the-cloudrdquo such

Hindawi Publishing CorporationJournal of Applied MathematicsVolume 2014 Article ID 512476 9 pageshttpdxdoiorg1011552014512476

2 Journal of Applied Mathematics

as Microsoft Biztalk workflow service and IBM LotusliveFrom the cloud service providersrsquo view cloud workflow offersautomatic task scheduling and resourcemanagement of cloudcomputing environment This category is referred to as ldquoin-the-cloudrdquo for example Cordys Process Factory (for GoogleAPPs)

There is still much timely and worthwhile work todo in the field of cloud workflow for example scalabilityand load balance of above-the-cloud workflow optimizationand integration of in-the-cloud workflow and so on Thegoal of our work in this paper is to design an above-the-cloud workflow engine based on Aneka cloud platform [6]considering scalability and load balanceThe remainder of thepaper is organized as follows Related work about workflowgrid workflow and cloud workflow is given in Section 2 Thecore part is given in Section 3 including formal modelingof workflow processes and design of the Aneka-based cloudworkflow engineThen the implementation and experimentsare presented in Section 4 Effectiveness and correctness ofthe novel workflow engine are shown by analysis resultsFinally in Section 5 the main results of the paper aresummarized

2 Related Work

From the mid-1990s to around 2000 business process man-agement and workflow technology were developed rapidlyand many valuable results and products were gained Thedetailed surveys can be found in [7ndash9] Besides flexibilityadaptability dynamic changes and exception handling ofworkflow have also been focused on and taken progress in thenext few years [10ndash13]Thus from the modeling and verifica-tion of workflow process to the design and implementationof workflow engine to the building and practice of workflowmanagement system a series of results have made workflowtechnology play amore andmore important role in social life

In the next ten years or so with the emergence ofnew computer technologies and computing paradigms forexample web services P2P and grid computing workflowcan be developed in two aspects On the one hand theexisting workflows can be transferred to new computingparadigms On the other hand adapting to new features oftechnologies and paradigms workflow technology shouldmake a progress Grid workflow can be classed into twocategories above-the-grid and in-the-grid as the discussionof cloud workflow in the above section The goal of in-the-grid workflow is integration composition and orchestrationof grid services in the grid environment considering peer-to-peer service interaction and complicated lifecycle man-agement of grid services [14] In regard to above-the-gridworkflow transformation of traditional business workflowsystems is the part of the work and many researches andpractices have been done concerning distributed and parallelfeatures of grid [15 16] covering modeling verificationscheduling problems and so on [17ndash21]

Comparing cloud computing emerging in 2007 with gridcomputing [2] it can be seen that their visions are thesame meanwhile there are both similarities and differences

between them from architecture security model businessmodel and computing model to provenance and applica-tions Based on the analysis of workflows running in thesetwo kinds of infrastructure similarities and differences arealso found [3 22] There are many workflow studies ondifferent levels of cloud services Research of workflowsbuilding on IaaS focuses on dynamical deployment andmonitoring in cloud nodes which is used in large-scale dataintensive computing [23 24] Some researchers use cloudworkflows in community team working among multipleprocesses [25] Otherwise many workflow studies in PaaSpay attention to the integration of cloud and workflow Itis concerned with recognition and execution of workflowtasks in cloud environment [26 27] Today there are alsomany studies on scientific and commercial cloud workflowsYuan et al studied data dependency and storage on scientificcloud workflows [28 29] Wu et al carried out researcheson hierarchical scheduling strategy in commercial cloudworkflows [30]

According to our analysis three kinds of differences orimprovements can be concluded Firstly cloud workflowtechnology research is always carried out joint with multipletechnologies and computing paradigms such as web servicesP2P and grid computing Secondly cloud workflow concen-trates more on data and resources and not just the controlflow Thirdly performance of cloud workflow is paid moreattention than functionalities [31ndash33]

3 Modeling and Design of Aneka-BasedCloud Workflow Engine

Our Aneka-based cloud workflow engine will be given in thissection To improve scalability the cloud workflow enginewill be designed to support different parallelism levels ofworkflow processes And to clarify these parallelisms theformal modeling technique for processes is given firstly

31 Preliminaries Workflowmanagement system completelyor partly supports automation of workflow schema Andworkflow schema models business processes it is character-ized by the decomposition into subflows and atomic activitiesthe control flow between activities (subflows) data flow anddata and the assignment of resources (including humanresources and equipment resources) to each activity [5]That is workflow schema is a combination of three essentialdimensions control flow data flow and resource flow

Petri net is a simple graphical yet rigorousmathematicalformalism which has been used tomodel workflowprocesses[34] However the usage of Petri net is always limited tomodel control flow which is not enough for describing theabove three dimensions of workflow An advanced modelfounded on the basic Petri net has been developed in ourprevious paper [35] called 3DWFN Its definition is given asfollows which is suitable for describing cloud workflow

Definition 1 Three-Dimension WorkFlow Net 3DWFNLet Σ Γ and Ψ be finite alphabet sets of activ-

ity names data names and resource names respectively

Journal of Applied Mathematics 3

A three-dimension workflow net over them is a system3119863119882119865119873 = ⟨119878 119879 119865 119862 119871119886119887 119864119909119901119872

0⟩ where we have the

following

(i) finite sets 119878 of places and 119879 of transitions 119878 cap 119879 = 0119878 cup 119879 = 0 and 119879 = 119879119886 cup 119879119901 cup 119879120591 where 119879119886 is a setof atomic transitions 119879119901 is a set of subnet transitionsand 119879120591 is a set of internal transitions

(ii) 119865 sube (119878times119879)cup(119879times119878) is a set of arcs especially includingthe set of inhibitor arcs 119865∘ sube 119865

(iii) 119862 is a finite and nonempty color set including types oftokens and the default is denoted by a black dot ldquo∙rdquo

(iv) Lab119879 rarr Σ119862 rarr ΓcupΨcupΓcupΨ is a labeling function(v) Exp 119865 rarr 119873times119862times119862119900119899 is an arc expression function

where119873 denotes the set of natural numbers and 119862119900119899is a set of expressions of numeric computation or logiccomputation

(vi) 1198720 S rarr 0 120583C is the initial marking of the net

where 120583C is the multiset over 119862

In view of defining transition rule of 3DWFN some notationsare introduced beforehand

Definition 2 (i) Projection Suppose 119865119909 119876 rarr 1198751times 1198752times

1198753times sdot sdot sdot exist119902 isin 119876 119901

119894isin 119875119894(119894 = 1 2 3 ) 119865119909(119902) =

(1199011 1199012 1199013 ) 119865119909(119902) uarr 119875

1times 1198752represents 119865119909(119902)rsquos projection

on 1198751times 1198752 and 119865119909(119902) uarr 119875

1times 1198752= (1199011 1199012) or 119901

1 1199012

(ii) Variables Replacement Let ldquo119909 119860rdquo represent 119909 is avariable of set 119860 Variable 119909 may be replaced by any elementin 119860

Then the transition rule is given

Definition 3 Transition Rule of 3DWFN

(i) Preconditions about a transition 119905 are denoted as forall 119904 isin

∙119905 (119904 119905) notin 119865

∘ Pre(119904 119905) = Exp(119904 119905)

Pre(119905) = ⋃119904isin∙119905Exp(119904 119905) and for all 119904 isin ∙119905 (119904 119905) isin

119865∘ Prev(119904 119905) = Exp(119904 119905)

(ii) Postcondition about a transition 119905 is denoted as for all119904 isin 119905∙ Post(119905 119904) = Exp(119905 119904)Post(119905) = ⋃

119904isin119905∙ Exp(119905 119904)

(iii) A transition 119905 isin 119879 is enabled under a marking 119872if and only if (for all 119904 isin ∙119905 (119904 119905) notin 119865

∘ 119872(119904) ge

(Pre(119904 119905) uarr NtimesC))and(for all 119904 isin ∙119905 (119904 119905) isin 119865∘ 119872(119904) ≺(Prev(119904 119905) uarr N times C))

(iv) A new marking 1198721015840 is produced after an enabledtransition fires1198721015840 = 119872minus(Pre(119905) uarr NtimesC)+(Post(119905) uarrN times C) denoting as119872[119905 gt 1198721015840 or119872 119905997888rarr 119905119872

1015840

32 Modeling and Analysis of Workflow Process Withdetailed analysis of generalized cloud workflow systems itis really indispensable to discriminate different parallelismsof workflow processes for adopting cloud technologies topromote its execution efficiencyTherefore we divide the exe-cution of multiple tasks in a cloud workflow system intothree levels according to different parallelisms that is processlevel execution task level execution and application levelexecution

U = ui | i = 1 2 3 middot middot middotR = ri | i = 1 2 3 middot middot middot

S11 S1t2S12 S13S1t1 S1t3 S14

S21 S2t3

S22

S26S2t1 S2t5 S28S23

S24 S2t4

S2t2 S25

S27

S29S2t6

S210

x U y R

⟨x y⟩ ⟨x y⟩⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩⟨x y⟩

⟨x y⟩ ⟨x y⟩

⟨x y⟩⟨x y⟩ ⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

Figure 1 Parallel processes modeled by 3DWFN

As shown in Figure 1 by 3DWFN there are two parallelprocesses named S1 and S2 S1 shows a sequential processand S2 shows a process including parallel tasks 119880 and 119877 arecolored sets 119880 represents the set of users while 119877 representsthe set of resources As a result execution of a 3DWFN lookslike a user acting as some role carries some kind of data oruses some kind of resources to ldquowalkrdquo through a certain pathof the net

These two processes could be performed in parallelat different computing node in cloud environment whichillustrates the parallelism of process level execution in thecloud workflow system

When the workflow system enter state S22 S23 and S24after transition S2t1 fired three tasks that are represented byS2t2 S2t3 and S2t4 are enabled to be carried out in parallelat different computing node in cloud environment controlledby a single user or multiple users Then the joint task S2t5could be triggered to be executed only if resources in S25S26 S27 and S210 are all available This scenario illustratesthe parallelism of task level execution in the cloud workflowsystem

Finally if a single task is intensive computing such asthe task execution between S12 and S13 it could be dividedinto more fine-grained subtasks and carried out in parallelat different computing node in cloud environment controlledby a single user This scenario shows the parallelism inapplication level

33 Architecture of CloudWorkflow According to the above-mentioned analysis the integrated framework of Aneka-based cloud workflow engine is presented There are threeparts environment applications and control parts whichcan solve the analyzed parallelisms problems in the threelevels and achieve extensibility and reusability of workflowDetails are presented in Figure 2

331 Cloud Workflow Environment There are three execut-ing models in the Aneka cloud environment Task ModelMapReduce Model and Thread Model In the remainder ofthis paper all experiments are run with Task Model


Aneka cloud platform

Task model Mapreduce model Thread model

Workflow engine Host application

Start Cloud application EndWorkflow

instance 1

Start EndWorkflow instance 2

Parallel processes

Task 2

middot middot middot


Figure 2 Integrated framework of Aneka-based cloud workflow engine

332 Cloud Workflow Applications The cloud workflowapplications contain all instances of workflow processes

333 Cloud Workflow Control The cloud workflow controlpart contains the workflow engine and host application Thecloud workflow engine is in charge of running workflowinstances and the host application is in charge of definingworkflow processes andmonitoring workflow process execu-tions

34 Design of Cloud Workflow

341 Design of Workflow Runtime Environment A light-weight cloud workflow engine is designed whose functionsinclude starting process execution scheduling processes andtasks based on preestablished rules The workflow runtimeenvironment is shown in Figure 3 which is composed of threeparts host application workflow instances and runtimeengine

Then Figure 4 shows all service classes design of theworkflow engineThe class ldquoWorkflowRuntimeServicerdquo is thebase class and the others are the derived classes

342 Design of Cloud Workflow Execution Process After aprocess is started each task will be executed following thecontrol flow when requirements of data flow and resourceflow are met When a task is enabled the task executorwill send its execution request to workflow engine Thenthe workflow engine will instantiate the ready task If thetask is not intensive computing then it will be executedlocally Otherwise if the task is intensive computing it willbe submitted to the cloud The execution process of cloudworkflow is depicted as in Figure 5

343 Design of Task Model Submission Process Next thesubmission process of task is designed As mentioned aboveTask Model is chosen In Aneka cloud environment TaskModel is used not only to solve the distributed applicationswhich are composed of single tasks but also to execute

AnekaWorkflow instance

Host application

Tasks

Rules

Runtime engine

Execution

Scheduling

Algorithm

(cloud application)

Figure 3 Workflow runtime environment

the correlating tasks of these applications Once users submittheir sequence of tasks with rules the results will be returnedby Aneka after a while

Aneka Task Model is composed of the class AnekaTaskthe interface ITask and the class AnekaApplication The tasksubmission process in Aneka Task Model is as follows firstlyto define a class ldquoUserTaskrdquo which inherits class ldquoAnekaTaskrdquoin Aneka Task Model secondly to create an instance ofUserTask for the application program and thirdly to packageclass ldquoUserTaskrdquo instance to class ldquoAnekaTaskrdquo and submit itto Aneka cloud by class ldquoAnekaApplicationrdquo The sequencediagram of the above process is shown in Figure 6

Then the implementation of workflow tasks is presentedin Figure 6 Firstly the implementation manager submitstasks to the workflow runtime engine Then the workflowruntime engine selects a custom made operation flow andcreates its workflow instance and then runs it at the sametime

4 Implementation and Experiment

41 Building Aneka Cloud Environment Our experimentalcloud environment includes a server as a master nodesome common PCs as the worker nodes and a manager


WorkflowRuntimeService-Runtime-State

+Start()+Stop()

ExternalDataExchangeService WorkflowPresistenceService

SqlWorkflowPresistenceService

TrackingService

+OnStopped()+OnStarted()

+RaiseServicesExceptionNotHandleEvent()

Figure 4 Class diagramWF service

Task execution request

Task executor

Runtime workflow engine

Processes completion

Flow1

Flow2

Task3 Task4Flow3

Aneka

Workflow instancesSubmitting to

execution Schedulingmonitoring

Figure 5 Process of cloud workflow execution

node In this paper the detailed hardware configuration ispresented in Table 1

Then the cloud nodes are pretreated according to thefollowing steps before installing Aneka

(a) Install FTP server on the master node and workernodes

(b) Offer access authority for the manager node to themaster node and worker nodes

After pretreatment Aneka cloudmanagement platform isinstalled in the manager node Then we use remote access toinstall and configure the master node and worker nodes

42Workflow Process of an Example Our example is to com-pute definite integral by a probabilisticmethodThe workflow

Table 1 Aneka cloud environment configuration

Type of node Operating system Quantity of computersManager Windows 7 1Master Windows Server 2008 1Worker Windows 7 3

process is composed of four steps ldquogenerations of randomnumberrdquo ldquocomputing119883-axisrdquo ldquocomputing119884-axisrdquo and ldquocom-putation of final resultrdquo

In this processing there are three correlating taskswhich are ldquogeneration of random numbersrdquo ldquocomputationof 119883-axis 119909rdquo and ldquocomputation of 119884-axis 119910rdquo We use thetask ldquogeneration of random numbersrdquo to compute real and


WF Engine AnekaTask Wrap

CreateWorkFlow Instance TaskService

Submit task

Submission response

New

TaskScheduleService TaskExecutionService

Task startWorkUnitSubmit

message

The process of creating App

WorktUnitSubmitstate

Figure 6 Process of submitting tasks

imaginary axes Then we match and combine values of 119909 119910to a 119901119900119894119899119905(119909 119910) Admittedly the combinational step is nextto the computational step of 119891(119909 119910) and the steps are in onetask In this case we divide the position relations of all pointsinto two casesWhen a119891(119909 119910) is between119909 = 119886 and119909 = 119887 welet the count increase by 1 In contrary count is not changedThen letting the number of all points be equal to number 119899we use 119901 = 119877119899 to estimate int119887

119886119891(119909) The operation flow is

present in Figure 7

43 Implementation Methods The task of computing axes isintensive computing andwill be submitted to theAneka cloudin Task Model The execution function under the interfaceITask is presented in Algorithm 1 Then the task is packagedand submitted to the cloud by the above mentioned classAnekaApplicationThe core implementation is configured inAlgorithm 2

44 Results Analysis Based on the above experiments run-ning results and workflow logs are analyzed Three analysisconclusions are given as follows Firstly as shown in Table 2the results of definite integral workflow are presented withdifferent number of points It shows that the degree ofaccuracy increases along with the increase of taskrsquos numberIt accords with the mathematical regular rule It is indicatedthat the functionality of our cloudworkflow engine is normal

Secondly the intensive computing tasks submitted to theAneka cloud are executed in parallel by different workers Asshown in Figure 8 the screenshot of workflow log tasks Aand B represent respectively the two intensive computingtasks ldquocomputing X-axisrdquo and ldquocomputing Y-axisrdquo It isindicated that our cloud workflow engine is effective andefficient

Task start

Get random numbers

Compute Y-axis Y Compute X-axis X

Combine point

Divide position relation of (X Y

(X Y)

) and

If x gt a and x lt b compute f(x y)

While n = 0 n

(n = 70)

Y = X2

fcount = fcount + 1

f = fcountn

Figure 7 Process diagram of task

Table 2 Result of different number of the spots

Number of the spots 70 100 150 1000Computing result 0286 0310 0393 0342

Thirdly the running time of the whole workflow comple-tion is not simple linear growth with the number of pointsused in definite integral The screenshot of workflow logshown in Table 3 presents the time table of 10 tasks for their


public void Execute ()

thisresult = (this119910 lowast this119910) lowast thisrd


thisresult = this119909+ (this119910minus this119909) lowast thisrd

Algorithm 1 Code of execute function

private static AnekaApplication lt AnekaTask TaskManager gt Setup (string [] args)

Configuration conf = ConfigurationGetConfiguration (ldquoconfiguration filerdquo) ensure that SingleSubmission is set to false and that ResubmitMode to MANUALconfSingleSubmission = falseconfResubmitMode = ResubmitModeMANUALconfUserCredential = new AnekaSecurityUserCredentials(ldquoadministratorrdquo ldquordquo)AnekaApplication lt AnekaTask TaskManager gt app =

new AnekaApplication lt AnekaTask TaskManager gt (ldquoWorkflow1rdquo conf) ensure that SingleSubmission is set to falseif (argsLength = = 1)

bLogOnly = (args [0] == ldquoLogOnlyrdquo true false)

return app

|

Algorithm 2 Aneka application configuration

Figure 8 Process of task execution

execution time waiting time and total time Mean executiontime is about 26 seconds and mean waiting time is about 1second except the last task The waiting time of the last taskis so long that it is nearly equal to the sum time of the other9 tasks In this case if the number of used points is increasedin the next experiment the total time might not be increasedbut decreased because the cloud might assign more workers(resources) to execute them to save time It is indicated thatthe cloud workflow engine is scalable which will not spend

an impossible large amount of time when a lot of tasks andprocesses run in parallel

5 Conclusion

The work in this paper can be concluded as follows A Petrinet-based model called 3DWFN is given firstly which candescribe three dimensions of a workflow that is control flowdata flow and resource flow According to the analysis of theexisting workflow systems it is found that cloud workflowpays more attention to data resources and performancethan control flow and functionality researched commonly intraditional workflow Thus 3DWFN is suitable for modelingof cloud workflow processesThrough analysis of the featuresof workflow processes executed potentially in cloud environ-ment three kinds of parallelisms are recognized as processlevel task level and application level and then modeledspecifically by 3DWFN The goal of the following design ofcloud workflow engine is to support these parallelisms toimprove scalability by using resources as far as in parallel

Then the architecture of the Aneka cloud basedworkflowengine is designed The workflow runtime environment andexecution process are stated and the process of packagingand submitting an intensive computing task to Aneka is


Table 3 Timetable of task

Total execution time Total waiting time Total time spent000002 0000005730000 0000025730000000006 0000012300000 0000072300000000002 0000006670000 0000026670000000002 0000013100000 0000033100000000003 0000008700000 0000038700000000004 0000000470000 0000040470000000003 0000004500000 0000034500000000001 0000029370000 0000039370000000001 0000007000000 0000017000000000002 00060195058631 0006159415449

explained After the engine is implemented a definite integralprocess is used as an example By different numbers of pointsused to definite integral the functionality and effectivity ofthe cloud workflow engine are shown Based on the analysisof running results and workflow logs the scalability andefficiency of the cloud workflow engine are given

In the future the research can be improved in followingdirections First more practical workflow processes will bedesigned using powerful expression of 3DWFN Secondnovel cloudworkflow engine will be built which has dynamicscheduling and handling functions with complicated process

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by Grants of the National NaturalScience Foundation of China [nos 61262082 and 61261019]Key Project of Chinese Ministry of Education [no 212025]and Inner Mongolia Science Foundation for DistinguishedYoung Scholars [2012JQ03] The authors wish to thank theanonymous reviewers for their helpful comments in review-ing this paper

References

[1] M A Vouk ldquoCloud computingmdashissues research and imple-mentationsrdquo Journal of Computing and Information Technologyvol 16 no 4 pp 235ndash246 2008

[2] I Foster Y Zhao I Raicu and S Lu ldquoCloud computing andgrid computing 360-degree comparedrdquo in Proceedings of theGrid Computing Environments Workshop (GCE rsquo08) pp 1ndash10November 2008

[3] X Z Chai and J Cao ldquoCloud computing oriented workflowtechnologyrdquo Journal of Chinese Computer Systems vol 33 no1 pp 90ndash95 2012

[4] R Buyya J Broberg and AM Goscinski EdsCloud Comput-ing Principles and Paradigms vol 87 John Wiley amp Sons 2010

[5] D Hollingsworth ldquoWorkflow management coalition the work-flow reference modelrdquo Tech Rep 68 Workflow ManagementCoalition Hampshire UK 1993

[6] RN Calheiros C Vecchiola D Karunamoorthy andR BuyyaldquoThe Aneka platform and QoS-driven resource provisioningfor elastic applications on hybrid cloudsrdquo Future GenerationComputer Systems vol 28 no 6 pp 861ndash870 2012

[7] W M P Van Der Aalst A H M Ter Hofstede and M WeskeldquoBusiness process management a surveyrdquo in Business ProcessManagement vol 2678 of Lecture Notes in Computer Sciencepp 1ndash12 Springer Berlin Germany 2003

[8] R Lu and S Sadiq ldquoA survey of comparative business processmodeling approachesrdquo in Business Information Systems pp 82ndash94 Springer Berlin Germany 2007

[9] E M Bahsi E Ceyhan and T Kosar ldquoConditional workflowmanagement a survey and analysisrdquo Scientific Programmingvol 15 no 4 pp 283ndash297 2007

[10] H Schonenberg R Mans N Russell N Mulyar and WVan Der Aalst ldquoProcess flexibility a survey of contemporaryapproachesrdquo in Advances in Enterprise Engineering I LectureNotes in Business Information Processing pp 16ndash30 SpringerBerlin Germany 2008

[11] S Smanchat S Ling and M Indrawan ldquoA survey on context-aware workflow adaptationsrdquo in Proceedings of the 6th Inter-national Conference on Advances in Mobile Computing andMultimedia (MoMM rsquo08) pp 414ndash417 ACM November 2008

[12] S Rinderle M Reichert and P Dadam ldquoCorrectness criteriafor dynamic changes in workflow systemsmdasha surveyrdquoData andKnowledge Engineering vol 50 no 1 pp 9ndash34 2004

[13] F Casati S Ceri S Paraboschi and G Pozzi ldquoSpecificationand implementation of exceptions in workflow managementsystemsrdquo ACM Transactions on Database Systems vol 24 no3 pp 405ndash451 1999

[14] S Krishnan P Wagstrom and G Von Laszewski ldquoGSFL aworkflow framework for grid servicesrdquo Tech Rep ANLMCS-P980-0802 Argonne National Laboratory 2002

[15] J Yu and R Buyya ldquoA taxonomy of workflow managementsystems for Grid computingrdquo Journal of Grid Computing vol3 no 3-4 pp 171ndash200 2005

[16] G C Fox and D Gannon ldquoSpecial issue workflow in gridsystemsrdquo Concurrency Computation Practice and Experiencevol 18 no 10 pp 1009ndash1019 2006

[17] C Pautasso and G Alonso ldquoParallel computing patterns forgrid workflowsrdquo in Proceedings of the Workshop on Workflowsin Support of Large-Scale Science (WORKS rsquo06) pp 1ndash10 IEEEJune 2006

[18] F Lautenbacher and B Bauer ldquoA survey on workflow annota-tion amp composition approachesrdquo in Proceedings of theWorkshop


on Semantic Business Process and Product Lifecycle Management(SemBPM rsquo07) 2007

[19] M Wieczorek A Hoheisel and R Prodan ldquoTaxonomies ofthe multi-criteria grid workflow scheduling problemrdquo in GridMiddleware and Services pp 237ndash264 Springer New York NYUSA 2008

[20] J Yu R Buyya and K Ramamohanarao ldquoWorkflow schedul-ing algorithms for grid computingrdquo Studies in ComputationalIntelligence vol 146 pp 173ndash214 2008

[21] J Chen and Y Yang ldquoA taxonomy of grid workflow verificationand validationrdquo Concurrency Computation Practice and Experi-ence vol 20 no 4 pp 347ndash360 2008

[22] E Deelman ldquoGrids and clouds making workflow applicationswork in heterogeneous distributed environmentsrdquo InternationalJournal ofHigh PerformanceComputingApplications vol 24 no3 pp 284ndash298 2010

[23] S Pandey D Karunamoorthy K K Gupta and R BuyyaldquoMegha workflow management system for application work-flowsrdquo in Proceedings of the IEEE Science amp Engineering Gradu-ate Research Expo 2009

[24] Q Chen L Wang and Z Shang ldquoMRGIS a MapReduce-enabled high performance workflow system for GISrdquo in Pro-ceedings of the 4th IEEE International Conference on eScience(eScience rsquo08) pp 646ndash651 December 2008

[25] W Li ldquoA Community Cloud oriented workflow system frame-work and its Scheduling Strategyrdquo inProceedings of the 2nd IEEESymposium onWeb Society (SWS rsquo10) pp 316ndash325 August 2010

[26] X Liu J Chen and Y Yang Temporal QOS Management inScientific Cloud Workflow Systems Elsevier 2012

[27] S Pandey D Karunamoorthy and R Buyya ldquoWorkflow enginefor cloudsrdquo in Cloud Computing Principles and Paradigms RBuyya J Broberg and A Goscinski Eds pp 321ndash344 JohnWiley amp Sons New York NY USA 2011

[28] D Yuan Y Yang X Liu G Zhang and J Chen ldquoA datadependency based strategy for intermediate data storage in sci-entific cloudworkflow systemsrdquoConcurrency andComputationPractice and Experience vol 24 no 9 pp 956ndash976 2012

[29] D Yuan Y Yang X Liu and J Chen ldquoOn-demand minimumcost benchmarking for intermediate dataset storage in scientificcloud workflow systemsrdquo Journal of Parallel and DistributedComputing vol 71 no 2 pp 316ndash332 2011

[30] Z Wu X Liu Z Ni D Yuan and Y Yang ldquoA market-orientedhierarchical scheduling strategy in cloud workflow systemsrdquoJournal of Supercomputing vol 63 no 1 pp 256ndash293 2013

[31] G Juve E Deelman G B Berriman B P Berman and PMaechling ldquoAn evaluation of the cost and performance of sci-entific workflows on amazon EC

2rdquo Journal of Grid Computing

vol 10 no 1 pp 5ndash21 2012[32] A Verma and S Kaushal ldquoDeadline and budget distribution

based cost-time optimization workflow scheduling algorithmfor cloudrdquo inProceedings of the IJCAon International Conferenceon Recent Advances and Future Trends in Information Technol-ogy (iRAFIT rsquo12) pp 1ndash4 2012

[33] J Behl T Distler F Heisig R Kapitza and M Schunter ldquoPro-viding fault-tolerant execution of web-service-based workflowswithin cloudsrdquo in Proceedings of the 2nd InternationalWorkshopon Cloud Computing Platforms (CloudCP rsquo12) article 7 April2012

[34] W M P Van Der Aalst ldquoThe application of Petri nets to work-flow managementrdquo Journal of Circuits Systems and Computersvol 8 no 1 pp 21ndash66 1998

[35] J Zhou and X Ye ldquoA flexible control strategy on workflowmodeling and enactingrdquo in Proceedings of the 8th InternationalConference Advanced Communication Technology (ICACT rsquo06)pp 1712ndash1716 Republic of Korea February 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Discrete MathematicsJournal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


as Microsoft Biztalk workflow service and IBM LotusliveFrom the cloud service providersrsquo view cloud workflow offersautomatic task scheduling and resourcemanagement of cloudcomputing environment This category is referred to as ldquoin-the-cloudrdquo for example Cordys Process Factory (for GoogleAPPs)

There is still much timely and worthwhile work todo in the field of cloud workflow for example scalabilityand load balance of above-the-cloud workflow optimizationand integration of in-the-cloud workflow and so on Thegoal of our work in this paper is to design an above-the-cloud workflow engine based on Aneka cloud platform [6]considering scalability and load balanceThe remainder of thepaper is organized as follows Related work about workflowgrid workflow and cloud workflow is given in Section 2 Thecore part is given in Section 3 including formal modelingof workflow processes and design of the Aneka-based cloudworkflow engineThen the implementation and experimentsare presented in Section 4 Effectiveness and correctness ofthe novel workflow engine are shown by analysis resultsFinally in Section 5 the main results of the paper aresummarized

2 Related Work

From the mid-1990s to around 2000 business process man-agement and workflow technology were developed rapidlyand many valuable results and products were gained Thedetailed surveys can be found in [7ndash9] Besides flexibilityadaptability dynamic changes and exception handling ofworkflow have also been focused on and taken progress in thenext few years [10ndash13]Thus from the modeling and verifica-tion of workflow process to the design and implementationof workflow engine to the building and practice of workflowmanagement system a series of results have made workflowtechnology play amore andmore important role in social life

In the next ten years or so with the emergence ofnew computer technologies and computing paradigms forexample web services P2P and grid computing workflowcan be developed in two aspects On the one hand theexisting workflows can be transferred to new computingparadigms On the other hand adapting to new features oftechnologies and paradigms workflow technology shouldmake a progress Grid workflow can be classed into twocategories above-the-grid and in-the-grid as the discussionof cloud workflow in the above section The goal of in-the-grid workflow is integration composition and orchestrationof grid services in the grid environment considering peer-to-peer service interaction and complicated lifecycle man-agement of grid services [14] In regard to above-the-gridworkflow transformation of traditional business workflowsystems is the part of the work and many researches andpractices have been done concerning distributed and parallelfeatures of grid [15 16] covering modeling verificationscheduling problems and so on [17ndash21]

Comparing cloud computing emerging in 2007 with gridcomputing [2] it can be seen that their visions are thesame meanwhile there are both similarities and differences

between them from architecture security model businessmodel and computing model to provenance and applica-tions Based on the analysis of workflows running in thesetwo kinds of infrastructure similarities and differences arealso found [3 22] There are many workflow studies ondifferent levels of cloud services Research of workflowsbuilding on IaaS focuses on dynamical deployment andmonitoring in cloud nodes which is used in large-scale dataintensive computing [23 24] Some researchers use cloudworkflows in community team working among multipleprocesses [25] Otherwise many workflow studies in PaaSpay attention to the integration of cloud and workflow Itis concerned with recognition and execution of workflowtasks in cloud environment [26 27] Today there are alsomany studies on scientific and commercial cloud workflowsYuan et al studied data dependency and storage on scientificcloud workflows [28 29] Wu et al carried out researcheson hierarchical scheduling strategy in commercial cloudworkflows [30]

According to our analysis three kinds of differences orimprovements can be concluded Firstly cloud workflowtechnology research is always carried out joint with multipletechnologies and computing paradigms such as web servicesP2P and grid computing Secondly cloud workflow concen-trates more on data and resources and not just the controlflow Thirdly performance of cloud workflow is paid moreattention than functionalities [31ndash33]

3 Modeling and Design of Aneka-BasedCloud Workflow Engine

Our Aneka-based cloud workflow engine will be given in thissection To improve scalability the cloud workflow enginewill be designed to support different parallelism levels ofworkflow processes And to clarify these parallelisms theformal modeling technique for processes is given firstly

31 Preliminaries Workflowmanagement system completelyor partly supports automation of workflow schema Andworkflow schema models business processes it is character-ized by the decomposition into subflows and atomic activitiesthe control flow between activities (subflows) data flow anddata and the assignment of resources (including humanresources and equipment resources) to each activity [5]That is workflow schema is a combination of three essentialdimensions control flow data flow and resource flow

Petri net is a simple graphical yet rigorousmathematicalformalism which has been used tomodel workflowprocesses[34] However the usage of Petri net is always limited tomodel control flow which is not enough for describing theabove three dimensions of workflow An advanced modelfounded on the basic Petri net has been developed in ourprevious paper [35] called 3DWFN Its definition is given asfollows which is suitable for describing cloud workflow

Definition 1 Three-Dimension WorkFlow Net 3DWFNLet Σ Γ and Ψ be finite alphabet sets of activ-

ity names data names and resource names respectively




following











119894isin 119875119894(119894 = 1 2 3 ) 119865119909(119902) =

(1199011 1199012 1199013 ) 119865119909(119902) uarr 119875


on 1198751times 1198752 and 119865119909(119902) uarr 119875

1times 1198752= (1199011 1199012) or 119901

1 1199012





∙119905 (119904 119905) notin 119865

∘ Pre(119904 119905) = Exp(119904 119905)


119865∘ Prev(119904 119905) = Exp(119904 119905)


119904isin119905∙ Exp(119905 119904)


∘ 119872(119904) ge



1015840



S11 S1t2S12 S13S1t1 S1t3 S14

S21 S2t3

S22

S26S2t1 S2t5 S28S23

S24 S2t4

S2t2 S25

S27

S29S2t6

S210

x U y R

⟨x y⟩ ⟨x y⟩⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩⟨x y⟩

⟨x y⟩ ⟨x y⟩

⟨x y⟩⟨x y⟩ ⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩













instance 1


Parallel processes

Task 2












Host application

Tasks

Rules

Runtime engine

Execution

Scheduling

Algorithm

(cloud application)









+Start()+Stop()



TrackingService





Task executor



Flow1

Flow2

Task3 Task4Flow3

Aneka

















Submit task

Submission response

New



message






present in Figure 7




Task start

Get random numbers


Combine point


(X Y)

) and


While n = 0 n

(n = 70)

Y = X2

fcount = fcount + 1

f = fcountn















return app

|





5 Conclusion










Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra












following











119894isin 119875119894(119894 = 1 2 3 ) 119865119909(119902) =

(1199011 1199012 1199013 ) 119865119909(119902) uarr 119875


on 1198751times 1198752 and 119865119909(119902) uarr 119875

1times 1198752= (1199011 1199012) or 119901

1 1199012





∙119905 (119904 119905) notin 119865

∘ Pre(119904 119905) = Exp(119904 119905)


119865∘ Prev(119904 119905) = Exp(119904 119905)


119904isin119905∙ Exp(119905 119904)


∘ 119872(119904) ge



1015840



S11 S1t2S12 S13S1t1 S1t3 S14

S21 S2t3

S22

S26S2t1 S2t5 S28S23

S24 S2t4

S2t2 S25

S27

S29S2t6

S210

x U y R

⟨x y⟩ ⟨x y⟩⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩⟨x y⟩

⟨x y⟩ ⟨x y⟩

⟨x y⟩⟨x y⟩ ⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩

⟨x y⟩













instance 1


Parallel processes

Task 2












Host application

Tasks

Rules

Runtime engine

Execution

Scheduling

Algorithm

(cloud application)









+Start()+Stop()



TrackingService





Task executor



Flow1

Flow2

Task3 Task4Flow3

Aneka

















Submit task

Submission response

New



message






present in Figure 7




Task start

Get random numbers


Combine point


(X Y)

) and


While n = 0 n

(n = 70)

Y = X2

fcount = fcount + 1

f = fcountn















return app

|





5 Conclusion










Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra














instance 1


Parallel processes

Task 2












Host application

Tasks

Rules

Runtime engine

Execution

Scheduling

Algorithm

(cloud application)









+Start()+Stop()



TrackingService





Task executor



Flow1

Flow2

Task3 Task4Flow3

Aneka

















Submit task

Submission response

New



message






present in Figure 7




Task start

Get random numbers


Combine point


(X Y)

) and


While n = 0 n

(n = 70)

Y = X2

fcount = fcount + 1

f = fcountn















return app

|





5 Conclusion










Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra











+Start()+Stop()



TrackingService





Task executor



Flow1

Flow2

Task3 Task4Flow3

Aneka

















Submit task

Submission response

New



message






present in Figure 7




Task start

Get random numbers


Combine point


(X Y)

) and


While n = 0 n

(n = 70)

Y = X2

fcount = fcount + 1

f = fcountn















return app

|





5 Conclusion










Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra












Submit task

Submission response

New



message






present in Figure 7




Task start

Get random numbers


Combine point


(X Y)

) and


While n = 0 n

(n = 70)

Y = X2

fcount = fcount + 1

f = fcountn















return app

|





5 Conclusion










Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra



















return app

|





5 Conclusion










Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















Acknowledgments


References















































Volume 2014




Journal of











Journal of


Function Spaces






Algebra





































Volume 2014




Journal of











Journal of


Function Spaces






Algebra









Research Article Modeling, Design, and Implementation of a ...

Documents

Transcript of Research Article Modeling, Design, and Implementation of a ...