Simulation as a Service

A Case Study of Provisioning Scientific Simulation Software via a Cloud Service

Morgan Eld

red, Alice Good and Carl Adams

School of Computing, University of Portsmouth, Portsmouth, U.K.

Keywords: Cloud Computing, Simulation as a Service, Software as a Service (SaaS).

Abstract: This paper reports on a case study that was conducted on a large scale cloud service project that moved

scientific simulation software to the cloud, one that used sensitive data. The study aimed to explore the

challenges and practicalities of initiating and evaluating simulation as a cloud service. Action research was

used to examine the nuances throughout the project as the service was moved from on-premise into a public

cloud, lasting over one year from start to finish. The study was able to identify some emergent issues

affecting initiation, technical security challenges and the evaluation of a significant change in a critical

applications provisioning model.

1 INTRODUCTION

During the last 20 years there has been a continuing

trend towards IT industrialisation. This has resulted

in IT services becoming repeatable and usable by a

wide range of customers and service providers. This

is because of the increasing commoditization of

technologies, virtualization and the rise of service-

oriented software architectures, along with the

dramatic growth in use of the Internet. These factors

constitute the basis of a discontinuity that offers

opportunities to shape the relationship between those

who consume and those who provide IT services.

The discontinuity implies that the ability to deliver

specialized services in IT can now be paired with the

ability to deliver those services in an industrialized

and pervasive way. The reality of this implication is

that users of IT services can focus on the business

capability of what the services provide, rather than

how the services are implemented or hosted. Similar

in nature to how utility companies sell power on

demand to subscribers, IT services can now easily be

delivered an provisioned as a contractual service.

This is not a new concept, but it does represent a

different model from the licensed-based, on-

premises models that have traditionally dominated

the IT industry.

Cloud services provide a new way of delivering

computing resources. Several types of cloud

computing platforms exist, of which the main types

are public, private and hybrid. Public clouds are

normally offered by commercial organisations that

provide access for a fee. Private clouds exist within

are contained within a specific organisation and

typically are not available for outside use. Hybrid

clouds are a mixture of private and public clouds

with the typical setup being that of a private cloud

that has the ability to call upon additional resources

from a public cloud (Chang, 2014).

The main advantage of cloud computing is the

ability of equilibrating the access to computing

resources for all types of businesses, regardless their

dimensions and investment capabilities. These

advantages include cost efficiency, scalability,

concentration, security and accessibility with a

further list below.

This paper outlines the overview, key issues and

themes that emerged in a study of a large scale

project within a mid-sized multinational company

that ran a pilot to provision a scientific simulation

software package via a public cloud.

2 RESEARCH METHODOLOGY

The research was conducted via a case study, taking

an action research approach which used an iterative

approach to collecting and analysing data. The

benefits of an action research approach are that it

focuses on generating solutions to practical

problems and empowers the researcher to engage

with the research and subsequent implementation

Eldred M., Good A. and Adams C..

Simulation as a Service - A Case Study of Provisioning Scientiﬁc Simulation Software via a Cloud Service.

DOI: 10.5220/0005539100370044

In Proceedings of the 2nd International Workshop on Emerging Software as a Service and Analytics (ESaaSA-2015), pages 37-44

ISBN: 978-989-758-110-6

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

activities (Mayer, 2000). A typical action research

methodology takes a five step approach, as follows:

 Step 1: Identify the Problem

 Step 2: Devise a Plan

 Step 3: Act to Implement a Plan

 Step 4: Observe

 Step 5: Reflect and Share

Using this methodology, the approach starts with

identifying the problem, which in this case was to

determine if the simulation software was able to run

via a cloud service. The second step was then to

devise a plan around the migration of the service to

the cloud and then test the success criteria. The next

step was to execute the plan and implement the

service, via a cloud provisioning model. This is the

part of the approach where the action research is

taking place via an iterative approach. After the plan

was implemented, the researcher would observe how

the service was or was not working. Once the

researcher has had time to observe the situation then

the entire process of action research was reflected

upon, and at times the whole research approach may

start over again (McCallister, 2011).

For this research, the researcher was a participant

observer who was present for: top management

meetings; from the inception of project to start-up;

to designing the service; all the way throughout the

whole project, till the end state of deciding if the

service would be provisioned via the cloud. This

access provided rare insight into what goes on in a

multinational organisation during a large scale cloud

service project. Along with access to top

management meetings, the researcher had access to

the critical role of the Head of IT strategy &

projects, which was the primary role for

orchestrating one of the biggest cloud pilots within

the industry. The research itself used an academic

approach to a real-world case study.

During the observation and reflection stages of

the action research approach, mixed methods were

used to evaluate the success of the project. These

included quantitative methods that were used to

determine the technical success criteria. These

methods looked at indicating whether the data would

be consistent before migrating to the cloud and then

in determining the run-time performance of the

simulations within a cloud environment.

The researcher used a mixed method of both

qualitative and quantitative data, in the form of

surveys. These were distributed to twenty four

employees of both technical and business staff to

find insight into trends that occur and organisational

challenges. The use of a mixed method helped back

up one set of findings from one method of data

collection underpinned by one methodology, with

another different method underpinned by another

methodology. The researcher designed a series of

survey questionnaires that included both boolean

and open-ended questions, so that the resulting data

would be both qualitative and quantitative.

Qualitative data was used and analysed in the

following approach. Questionnaires with open-ended

questions were sent to twenty four pre-selected

participants, coming from a wide range of both

technical and business staff. The Questionnaires

were distributed electronically via an online survey

tool, with replies sorted and trends were identified to

find commonalities. Upon the initial analysis another

set of quantitative questionnaires was distributed to

further investigate the findings and commonalities.

The decision on the selection of interviewees was

determined via a deductive approach to responses.

Examples of the specific questions that were

used in the survey are listed below.

 Was the project a success

 Is cloud a viable provision model for

scientific applications

 Is cloud scalable to run simulations

 Was the organization ready for cloud

 Does the organization need to introduce

new processes for the adoption of cloud

Examples of the specific open-ended questions that

were asked are listed below.

 What emerging themes were identified

 What key issues that were identified

 What was the impact of key issues

3 CASE STUDY

The case study is based on a midsized international

company with a headquarters in Europe with

operations in Europe, the Middle-east and Africa.

The company has approximately 5,000 employees

located over seven countries with revenues

consistently averaging between $8-10 billion dollars

and has a corporate culture that promotes

innovation.

The company was exploring the possibility of

migrating scientific simulation software that has

significant computing and storage requirements into

a cloud based HPC environment service delivery

model. The benefits of moving from on-premise to a

cloud provisioning model were that it would enable

the scientific community within the company to

flexibly increase compute via a cost effective, on-

demand, pay-per use model (Jackson et al, 2012).

ESaaSA2015-WorkshoponEmergingSoftwareasaServiceandAnalytics

If successful, this new capability would enable

the company to compete with much larger

companies who had the capital to invest in the

development and maintenance of large scale on-

premise super-computing environments. A project

was conducted to do a formal evaluation of

migrating the service to the cloud and to determine if

the concept was feasible from a technical and

economic perspective, before a decision to invest

further into a cloud provisioning service model was

decided. The project was a multimillion dollar

project, lasting over a year that consisted of a five

person project team, with twelve other stakeholders

from IT and outside IT whom were involved in the

project.

3.1 Problem Statement

Scientists within the organisation within this case

study were being challenged with a need for superior

simulation modelling, as both the supply of

information and the sophistication of quantitative

techniques increases. The organisation invested

heavily in technology that providing a vastly higher

resolution of raw data, generating unprecedented

volumes of data. All this additional data enables

finer-scale simulation, as the geo-cellular models

they simulated burst through the 10 and 100 million

cell thresholds. As impressive as these advances

were, they only represent more granular approaches

to traditional modeling methods.

As the models increase in size, the organisation

requires significantly more computing resources to

run, given the increasing complexity with detailed

models needing to run hundreds of times to quantify

uncertainties and define the risks. The growth in

demand for high performance computing was

exceeding the supply from vendors. This results in

the organisations’ science community needing to

limit simulations due to computing capacity, and

was the driving force in running a proof of concept

to explore new sources of high-performance

computing capacity via a cloud provisioning model.

With the organisation investing in the project, it

would need to determine real practical questions in

relation to simulation as a service such as:

 Will the project be successful

 Is cloud a viable model for scientific

applications

 Is cloud scalable for simulations

 Is the organisation ready for cloud

 Will new processes need to be implemented

for cloud

 What themes would emerge

 What key issues would be faced

3.2 Plan

The project plan was drafted during the development

of the business case. The project was broken down

into six milestones, with each milestone having an

estimated time. The milestones consisted of having:

a signed off business case; a contract agreement with

vendors; a high and low level design; the

implementation; testing and finally an analysis

conducted on the results leading to the project

findings. Overall the initial estimation of the project

plan was that it would take around five months to

complete, however the actual time for the project

was fifteen months. Significant challenges were

faced in almost every step of the project plan.

Business Case: This stage took four months

instead of an expected one month, as the biggest

delay was in getting approval and buy-in for the

business case, with the critical element revolving

around the way that the sensitive data used within

the simulations would be protected.

Contract: Similar to attaining business case

approval, getting a contract signed with the software

vendor took four months instead of the expected one

month. This was because neither the organisation

nor the software vendor could come to an agreement

around intellectual property rights. In the end the

situation was resolved as both parties agreed to

waive rights.

Design & Implementation: The design took twice

as long as expected, due to stringent internal

information and data security requirements. This

impacted the implementation as the vendor software

required using a physical license dongle. As the

project took an action research approach, the design

and implementation phases took an iterative

approach and required reworking of the designs

during the implementation phase.

Testing of the service was approximately three

Table I: Milestones.

Milestone Timelines

Estimated Actual

Business Case

1 Month 4 Months

Contract

Agreement

1 Month 4 Months

Design

6 Weeks 3 Month

Implementation

1 Week 1 Month

Testing

3 Months 3 Months

Findings

1 Month 2 Months

SimulationasaService-ACaseStudyofProvisioningScientificSimulationSoftwareviaaCloudService

months the same time as initially estimated.

Findings: The findings took two months instead

of the estimated one month, as the key themes and

issues, led to some insightful findings, which

required further investigation into finding

commonalities and in conducting in depth interviews

to validate the findings.

3.3 Design

The design was developed with the guiding principle

that the architecture would need to be secure, lean

and agile. This is because it was hypothesised that it

would drive efficiencies and reliability through an

elastic architecture that could dynamically scale up

or down compute clusters as needed. The objective

of the design was to simulate a real-life corporate

network within a cloud scenario, so that the cloud

service would be almost identical to the on-premise

service, so that data could easily be moved in and

out. As such a Virtual Private Cloud-VPC in the

Amazon European datacentre was setup to act as the

“corporate network”. The next step was to create a

VPC in an Amazon US datacentre to act as the

“cloud network”. The connections between the

installations were facilitated through the use of

OpenVPN which was installed on standard Linux

Amazon Machine Images.

A major design requirement for running

simulations in the cloud involves how to transfer

large datasets between the corporate network and the

cloud environment. To achieve this, a cloud network

attached storage-NAS server was provisioned in the

cloud, with a virtual device in the Amazon cloud

configured to acts as a NAS front end to Amazon’s

object based data cloud, simple storage service-S3.

Due to the large storage requirements; there was a

need for a common internet file system, which is a

standard way that computer users share files across

corporate intranets and the internet, with a network

file system interface. This design is commonly

known as a cloud storage security gateway system

and is considered a secure way for encrypting and

decrypting data as it is either uploaded or

downloaded via Amazon’s S3 by examining the

consistency of the contents and preventing data

tampering (Wang et al 2013).

The next design step was to create the Simulator

software head node in the cloud. This node would be

static and would be were the simulation jobs would

be submitted to and then run in a dynamically-

created compute cluster. The head node rand on a

red hat enterprise linux node running on an Amazon

C3.xlarge size server. This server was selected as it

had the required compute capacity need for the

simulations along with having a solid state drive and

a 10GB network interface. The design choice for

using a solid state drive is that they offer higher

performance compared to traditional storage devices

and are needed in HPC systems, especially those

with a growing demand of supporting big data

applications (Chen et al, 2013).

A major security challenge was in the need to

connect the simulation software’s physical Universal

Serial Bus-USB license dongle to a virtual server.

This was resolved via the use of a USB network

device server being placed within a de-militarised

zone-DMZ within the corporate network. This

enabled the mapping of a USB port to a virtual

server over the network. The USB port on the device

server was then mapped to the simulator license

server in the DMZ and was configured with a public

IP. The DMZ was also configured to allow traffic

between Amazon and the license server.

Figure 1 depicts the conceptual design which

indicates the three main networks within the setup,

the Amazon cloud, the virtual office and the DMZ

and the major components within each of these

networks.

Figure 1: Conceptual Design.

A key component in the design was the NAS secure

storage gateway (which also acted as a cache).

Figure 2 depicts the dataflow from the simulation

cloud which was within the amazon cloud. The data

was de/encrypted as it passed through the cloud

NAS and resided in the amazon storage, until it was

to the NAS in the virtual office cloud, where it was

then de/encrypted and passed into the main data

source.

Along with the data flow there was a need to

have a connection into the corporate DMZ as the

simulation license server needing to be on the same

subnet as the USB device server, thus not enabling

the license server to be placed in the Amazon Cloud.

The detailed design for this is depicted in Figure 3.

ESaaSA2015-WorkshoponEmergingSoftwareasaServiceandAnalytics

Figure 2: Data Flow.

Figure 3: DMZ Design.

3.4 Technical Validation

Before the simulation cloud service was provisioned,

an on-premise validation test was needed to ensure

that simulations would run with the same accuracy

regardless of the technology provisioning model.

The approach taken for testing was to test out the

characteristics with different stakeholders, to ensure

requirements from users and those supporting the

simulation software were properly gathered. The

guidance provided was that the validation would

need to ensure that the business characteristics, such

as the need to ensure consistency when running

simulation on the scalable dataset. The technical

characteristics were that cpu performance scalability

would need to be performed.

To achieve this, four cases were run on a

workstation and then moved to the companies on-

premise cluster which had a maximum of 8 core

cpu’s. The test ran the cases on the on-premise

cluster using one, four and 8 cpu’s to ensure

consistency. The results from this indicated that

moving the simulation jobs, did not have an impact

on the jobs and once this was successful the

simulation jobs could be migrated to the cloud

service. Figure 4 indicates that the four cases

demonstrated identical results for production rate

and cumulative production for the duration of the

field history as expected:

Figure 4: Simulation Validation.

The case with a single CPU on a workstation was

completed in 20.9 hours, while the case on the on-

premise cluster was completed in 19.6 hrs. The run

time for the on-premise cluster with four and eight 8

CPUs ran at 7.7 and 5.7 hours respectively. The in-

house cluster setup at the time of this study did not

allow job executions on more than 8 CPUs. Figure 5

demonstrates the relationship between number of

CPUs and wall clock time.

Figure 5: CPU Performance Validation.

With the run simulations being validated and with

the performance of the on-premise service being

measured, the next step was to compare this against

the results of the performance within the cloud

provisioned serviced.

Figure 6 shows the results, and indicates that the

wall clock time decreases as more CPUs were

added, for both calculations for the on-premise and

cloud service. It was observed that on-premise

calculations stagnated at more than 4 CPUs,

SimulationasaService-ACaseStudyofProvisioningScientificSimulationSoftwareviaaCloudService

resulting in sublinear scaling. This is contrary to the

performance via the cloud service, which observed

close to linear scaling. Assuming that this linear

scaling persists when adding more than 8 CPUs,

extrapolating from this observation, it was

hypothesised that for larger jobs, the performance of

the cloud provisioned service would significantly

increase compared to what can be achieved on-

premise. This analysis is not exhaustive, but was

severely limited by the size of the on-premise

service having only 8 CPUs.

Figure 6: On-premise & Cloud Performance Validation.

3.5 Data Collection & Analysis

Once the system was implemented and had passed

the technical validation aspect, twenty four

individuals involved in the project completed a

questionnaire, with 15 coming from IT and 9 from

outside.

The 24 interviewees were asked: whether they

thought the project was a success; whether cloud

was a viable service model for scientific applications

and if cloud was a scalable to run simulations; if the

organisation was ready for cloud and if the

organisation needed to introduce new processes for

cloud. For data analysis purposes yes equating to a

score of 1, while a no equated to a score of 0. The

response and standard deviation were calculated as

indicated in Table 1. Not all questions had an input,

as respondents preferred not to say and the

breakdown is as follows:

 Question 2: 4 participants declined

 Question 3: 1 participant declined

 Question 4: 1 participant declined

The following key insights were indicated

 80% indicated that the project was a success.

 71% indicating that cloud was a viable service

model for scientific applications.

 79% indicating that cloud is a scalable.

 Surprisingly only 25% indicated that the

organisation was ready for cloud.

 71% indicating that the organisation needed to

implement new processes for cloud.

Table 1: Success Criteria.

Question Response

MEAN STANDARD

DEVIATION

Yes

& (Total

1. Was the

project a

success

(80%)

0.833 0.381

2. Is cloud

viable for

scientific

applications

(71%)

0.850 0.366

3. Is cloud

scalable

(79%)

0.826 0.388

4. Is the

organisation

ready for

cloud

(25%)

0.261 0.449

5. The

organisation

needs new

processes

for cloud

adoption

(71%)

0.708 0.464

The same interviewees were then asked their opinion

of the three major themes which emerged during the

life of the project.

The following key insights were indicated

 62.5% indicated that politics were a prevalent

theme, 38% indicated that politics was the first

theme, 25% indicated it was the second theme.

 Innovation was second with 45.8%, 25% for the

first theme, 8% for the second and 13% for the

third.

 Security at 33.3%, 13% for the first, 17% for the

second and 4% for the third.

Other key themes included vendor solutions,

intellectual property rights, a lack of the required

skills, internal processes, business value and change.

Interviewees were then asked their opinion of the

three major issues which emerged during the life of

the project.

The following key insights were indicated

 Politics was again the highest at 66.7% which

was aligned with the responses from the

emerging themes, with 42% of respondents

indicating it was the first issues while 25%

indicated it was the second issues and similar to

themes zero respondents indicated that it was

the third issue.

1248

Wall Clock Time [Hours]

Wall Clock Time versus Number of CPU's

Internal

Amazon

ESaaSA2015-WorkshoponEmergingSoftwareasaServiceandAnalytics

Table 2: Emerging Themes.

Theme First Second Third Inclusion

Total %

Business

Value

1 2 0 12.5%

Change 0 2 1 12.5%

Innovation 6 2 3 45.8%

Intellectual

Property

Rights

2 2 1 20.8%

Politics 9 6 0 62.5%

Processes 0 2 1 12.5%

Security 3 4 1 33.3%

Skills 0 1 2 12.5%

Vendor

Solutions

3 1 1 20.8%

 Project Management was second with 37.5%,

21% for the first theme and 16.5% for the

second.

 Contracts and Processes were tied for the third

issue both with a response of 20.8% with

contracts had a response of 12.5% for being the

first issue and 4.15% for being the second and

third issues.

 4.15% identified processes as being the first

issue and 16.65% for being the third issue.

Other key issues included capability of staff, lack of

clear KPI’s to measure and information Security.

Table 3: Key Issues.

Key Issue First Second Third Total Total

Capability 2 2 0 4 16.7%

Contracts 3 1 1 5 20.8%

KPI’s 0 1 1 2 8.3%

Politics 10 6 0 16 66.7%

Processes 1 0 4 5 20.8%

Project

Management

5 4 0 9 37.5%

Security 0 0 1 1 4.2%

4 DISCUSSION

The pilot was initially resisted by internal members

of the IT department that were responsible for

supporting the simulation software. This delayed the

approval of the business case. Design challenges

arose during the design and implementation stages

due to stringent internal security requirements.

This was a surprising finding as the respondents

did not indicate information security to be an issue,

this along with the data captured from the surveys

indicated that politics was the pervasive theme and

key issue of moving to the cloud. Considering that

the project was determined a success, but that the

organisation was not ready and would need to

introduce new processes to support cloud. This leads

to a finding of how organisation behavior and the

perception of trust in security pose a real threat to

the adoption of cloud. This is indicated by 2009

Gartner survey with indicates that politics is a

challenge of cloud service adoption (Gartner, 2009).

Resistance to change is a normal human response

as employees seek to translate the change to a

personal context, which can be greatly magnified by

fear of the unknown (Berube, 2012). If internal

policies and security concerns are a significant

challenge in cloud service adoption, then building a

maturity assessment at the start of the project to

understand the organisations culture, internal

processes would clearly assist in migration to a

cloud service, and the delivery of a cloud service

project could then plan accordingly to further train

and develop staff on the impact of cloud, before any

implementation would occur. This activity was not

included in the project, but the insights received

after the fact, could help in any future cloud project

by being able to measure and mitigate the risks of

the cloud.

The data which was collected using an action

research approach indicates that a lot is still

unknown about dealing with challenges during the

initiation stages of a cloud project were the

realization that the change from one modal of

working to another different modal has a significant

impact on the success of a project. Though the

project was validated as being a success, several

emergent themes impacted the adoption. One

significant emergent theme from the research was

that the organisation did not have the appropriate

internal policies.

The research shows that the evaluation and

adoption of simulation as a service project, which is

a considerable change to business practices, will

likely involve more than technical performance and

business improvements: It will also need to consider

the wider political fault-lines of issues that would

impact the acceptance from various stakeholders.

5 CONCLUSIONS

Cloud computing is maturing, but there is still a lot

that remains uncertain for its adoption within

SimulationasaService-ACaseStudyofProvisioningScientificSimulationSoftwareviaaCloudService

enterprises, such as the organizational changes

brought about by cloud computing. Cloud services

that support simulation via a HPC environment are

attracting more attention in literature, in big business

and in governments.

This paper has reported on research exploring the

practicalities of conducting a significant simulation

as a service project within a large company. This

paper further explores the practicalities and contexts

the issues of applying cloud to larger compute

processing needs

This is one of the few works that covers

simulation as a service in a real life project.

The research involved an iterative methodology

based upon an action research methodology and

covered all the stages of the project from creation to

evaluation. The pilot project and research focused on

evaluating the possibility of running simulation as a

service which leverage a cloud infrastructure to

address the HPC needs of the multinational company

using a range of criteria, including technical

capability and wider business case.

It was a successful project and the insights taken

from this work can further be used to make informed

decisions about moving simulations to the cloud.

Lessons learned from this would be that doing a

proof of concept is a good method.

The data which was collected using an action

research approach indicates that a lot is still

unknown about dealing with challenges during the

initiation stages of a cloud project were the

realization that the change from one modal of

working to another different modal has a significant

impact on the success of a project. Though the

project was validated as being a success, several

emergent themes impacted the adoption. One

significant emergent theme from the research was

that the organisation did not have the appropriate

internal policies

The research shows that the evaluation and

adoption of simulation as a service project, which is

a considerable change to business practices, will

likely involve more than technical performance and

business improvements: It will also need to consider

the wider political fault-lines of issues that would

impact the acceptance from various stakeholders.

Developers and project managers can take

practical guidelines from this paper that can be used

to make informed decisions about moving

simulations to the cloud. These examples are in the

form of design, validation steps but more

importantly the need to get feedback from different

stakeholders before starting a project and the need to

have an understanding of the potential political

impact may occur similar to this project in terms of

project delays and in design requirements. Key

contributions to knowledge are that even if the

project is successful, the organisation may not be

ready for cloud and that new processes would need

to be developed to operate via a cloud provisioning

model. For considerably sized projects of this type

the recommendation is to run a pilot first and to plan

and execute the development of internal processes

that are required to enable the organisation to be

cloud ready.

REFERENCES

H. F. Wang, L. J. Wang, Pingjian Institute of Information

Engineering, Beijing, China, Chinese Academy of

Sciences, 2013.

N. Agrawal, V. Prabhakaran, T. Wobber, J.D. Davis, M.

Manasse, R. Panigrahy, Design Tradeoffs for SSD

Performance, Microsoft Research, Silicon Valley.

J. Chen, Using pattern-models to guide SSD deployment

for Big Data applications in HPC systems, 2013 IEEE

International Conference on Big Data, 2013.

V. Chang, The Business Intelligence As a Service in the

Cloud, 37, 512-534 ed. , Future Generation Computer

Systems, 2014.

C. Vecchiola, S. Pandey, R. Buyya, High-Performance

Cloud Computing: A View of Scientific Applications,

Pervasive Systems, Algorithms, and Networks, 10th

International Symposium on Pervasive Systems, and

Networks, 2009.

V. Chang C S. Li, D. De Roure, G. Wills, R. Walters, C.

Chee, The Financial Clouds Review, 1 (2). pp. 41-63.

ISSN 2156-1834, eISSN 2156-1826. ed. ,

International Journal of Cloud Applications and

Computing, 2011.

M. Armbrust, A. Fox, R. Griffith, A. Joseph, R. Katz, A.

Konwinski, G. Lee, D. Patterson, A. Rabkin, I. Stoica,

M. Zaharia, A View of Cloud Computing, Vol. 53 No.

4 ed. , ACM, 2010.

J. Meyer, Using qualitative methods in health related

action research, 320: 178-181 ed. , British Medical

Journal, 2000.

J. McCallister, Contemporary Social Work Issues, MSW,

2011.

K. R. Jackson, L. Ramakrishnan, K. Muriki, S. Canon,

Performance Analysis of High Performance

Computing Applications on the Amazon Web Services

Cloud, IEEE Second International Conference on

Cloud Computing Technology and Science, 2010.

Gartner, Managing and Deploying Clouds, Gartner 2009

Datacenter Conference, 2009.

D. Berube, Resistance to Change is a Good Thing, Life

Cycle Engineering, 2012.

A. Khajeh-Hosseini, I. Sommerville, I. Sriram, Research

Challenges for Enterprise Cloud Computing, 1st ACM

Symposium on Cloud Computing, 2010.

ESaaSA2015-WorkshoponEmergingSoftwareasaServiceandAnalytics