Towards a Fuzzy-oriented Framework for Service Selection in Cloud

e-Marketplaces

Azubuike Ezenwoke

, Olawande Daramola

and Matthew Adigun

Covenant University, Ota, Nigeria

Center for Excellence for mobile e-services, University of Zululand, South Africa

Keywords: Cloud Service Selection, e-Marketplace, Constraint Satisfaction, Fuzzy Decision Making, Fuzzy AHP,

Visualization.

Abstract: The growing popularity of cloud services requires service selection platforms that offer enhanced user

experience in terms of handling complex user requirements, elicitation of quality of service (QoS)

requirements, and presentation of search results to aid decision making. So far, none of the existing cloud

service selection approaches has provided a framework that wholly possesses these attributes. In this paper,

we proposed a fuzzy-oriented framework that could facilitate enhanced user experience in cloud e-

marketplaces through formal composition of atomic services to satisfy complex user requirements,

elicitation and processing of subjective user QoS requirements, and presentation of search results in a

visually intuitive way that aids users’ decision making. To do this, an integration of key concepts such as

constrained-based reasoning on feature models, fuzzy pairwise comparison of QoS attributes, fuzzy decision

making, and information visualization have been used. The applicability of the framework is illustrated with

an example of Customer Relationship Management as a Service.

1 INTRODUCTION

The growing popularity of cloud services requires

service selection platforms that enable the

composition of atomic services to satisfy complex

user requirements, and highlight the quality of

services (QoS) attributes of these value-added

services under one e-marketplace structure (Akolkar

et al., 2012; Gatzioura et al., 2012). Despite their

successes, commercial cloud e-marketplaces (e.g.

AppExchange and SaaSMax) do not yet enable

dynamic composition, and employ keyword-based

search mechanisms that do not consider user’s QoS

requirements, nor support the elicitation of these

requirements in ways akin to subjective human

expressions. In addition, search results on these

platforms are presented as unordered lists of icons,

with little or no comparison apparatus that simplifies

decision making. Existing cloud selection

approaches (see Figure 1) do not currently provide

the sophistication to optimize user experience in the

e-marketplace (Akolkar et al., 2012).

For example, some approaches only allow users

to make selection from a predefined list of atomic

services, which does not address complex situations

where a user’s requirements extend beyond what

atomic services can provide (e.g. Esposito et al.,

2015). Additionally, some other methods lack the

flexibility to accommodate subjective QoS inputs,

and demand that a user specifies requirements in

exact terms, e.g. (Wittern et al., 2012; Rehman et al.,

2011).

Qu and Buyya (2014) observed that user’s QoS

requirements can indeed be specified in terms of

preferences (user’s priority for each QoS dimension)

and aspiration (user’s values for QoS dimension) as

two important considerations for determining which

cloud services to select. However, some existing

approaches that have considered subjectivity in user

requirements elicit either QoS preferences or QoS

aspirations alone from the user but rarely both (e.g.

Esposito et al., 2015; Yu and Zhang, 2014). Still,

some others (e.g. Esposito et al., 2015; Mirmotalebi

et al., 2012; Qu and Buyya, 2014; Rehman et al.,

2011) require users to assign priority weights to QoS

attributes, with the downside of being less accurate

compared to pairwise comparison of the relative

importance of QoS attributes (Millet, 1997).

Again, many approaches (e.g. Esposito et al.,

2015; Yu and Zhang, 2014; Qu and Buyya, 2014;

604

Ezenwoke, A., Daramola, O. and Adigun, M.

Towards a Fuzzy-oriented Framework for Service Selection in Cloud e-Marketplaces.

DOI: 10.5220/0006365306320637

In Proceedings of the 7th International Conference on Cloud Computing and Services Science (CLOSER 2017), pages 604-609

ISBN: 978-989-758-243-1

Wittern et al., 2012) present search results in a

textual list or table format, that do not make obvious

the trade-off among search results, which makes it

more cognitively demanding for users to make

decisions based on search results (Beets and

Wesson, 2011). Ultimately, these identified

limitations will hamper user experience in the cloud

service e-marketplace.

Figure 1: Summary of Related works.

Hence, in this paper, we present an integrated

framework for cloud service selection that supports

fuzzy-oriented decision making, and formal

composition of atomic services in response to

complex user requirements. This is proposed as an

approach to cloud service selection that caters for

observed limitations in existing cloud service

selection approaches. To do this, we employed an

integration of relevant concepts such as: 1) feature

modelling - to organize atomic services within the

cloud ecosystem; 2) constraint-based reasoning - to

guide formal service composition on the fly; 3)

Fuzzy-based prioritization and analysis methods – to

handle subjective user QoS preferences and

aspiration; and 4) information visualization – to

enable easy comparison of query results along

multiple QoS dimensions. Our framework is

proposed as an improvement to existing cloud

service selection approaches.

The rest of this paper is structured as follows:

Section 2 provides some background and Section 3

contains related works. Our framework is presented

in Section 4, while its applicability is presented in

Section 5. Section 6 discusses the implication of our

framework and Section 7 contains conclusion and

future work.

2 BACKGROUND

Cloud Services Ecosystem: A cloud services

ecosystem is an environment that host

heterogeneous cloud service offerings from different

providers, and affords the opportunity of

collaborations. A cloud service ecosystem is

analogous to a software product line (SPL).

Automated Reasoning on Feature Models:

The cornerstones of SPL endeavours are: a

knowledge model (e.g. feature model) that captures

the relationships among the components based on

variabilities, and computer-aided reasoning to derive

useful information from the model.

A feature model (FM) is a hierarchically

arranged collection of features and consists of the

inter-relationships between a parent feature and its

child features, and a set of cross–tree constraints that

define the criteria for feature inclusion or exclusion

(Berger et al., 2014). There are 3 types of FM: basic,

cardinality-based and extended feature models

(EFM) (Benavides et al., 2010). We adopted EFM

because it allows the modelling of cloud services,

their QoS attributes and relationship constraints

more naturally.

Automated reasoning is performed by mapping

the FM into logic-based encodings (e.g. description

logic, propositional logic, and constraint

programming), and encodings are inputted into

solvers to find valid compositions of atomic services

(Benavides et al., 2010). The overall QoS attributes

of the valid combinations is determined by the QoS

factors of constituent services, and are computed

using QoS aggregation functions. Types and

application of QoS aggregation functions can be

found in (Mohabbati et al., 2011).

Fuzzy Set Theory: Fuzzy set theory is effective

to capture vagueness that characterizes user QoS

requirements (Qu and Buyya, 2014; Zadeh, 1974).

Each QoS attribute can be represented as a linguistic

variable and users can express QoS preferences and

aspiration using linguistic terms. In this paper,

preference weights were derived using fuzzy

pairwise comparison in Fuzzy-AHP; while QoS

aspirations were elicited and analysed as a system of

fuzzy goals and constraints (Bellman and Zadeh,

1970).

3 RELATED WORKS

A number of approaches for selecting cloud services

exist in the literature (see Figure 1). The approach in

Towards a Fuzzy-oriented Framework for Service Selection in Cloud e-Marketplaces

605

(Esposito et al., 2015) uses fuzzy sets theory to

handle uncertainty in users’ preferences and a

TOPSIS-based method to rank services. Yu and

Zhang (2014) proposed a SaaS selection model for

group users, by eliciting vague QoS preferences

using interval numbers. The approach of Qu and

Buyya (2014) employs a hierarchical fuzzy

inference system for cloud service selection, while

Kwon and Seo (2013) described an IaaS selection

model based on Fuzzy-AHP. Wittern et al. (2012)

presented an approach to harness cloud service

capabilities using variability model and likely

alternatives are subjected to a preference-based

ranking process. Mirmotalebi et al. (2012) proposed

an approach for ranking cloud services based on

both explicit and implicit user preferences. Rehman

et al. (2011) proposed two methods for service

selection based on similarity of users’ requirements

and service’s properties.

From the summary of relevant previous efforts

shown in Figure 1, all approaches, except Wittern et

al. (2012), cannot compose atomic services to meet

complex user requirements. Furthermore, only Qu

and Buyya (2014) elicits both subjective QoS

preferences and aspirations, while only Kwon and

Seo (2012) incorporates some type of visualization

to aid service selection. In contrast, our proposed

framework will enable composition of atomic

services, elicitation of subjective QoS requirements

and result visualization to aid selection in order to

foster improved user experience during service

selection in cloud e-marketplaces.

4 PROPOSED FRAMEWORK

This paper proposes a fuzzy-oriented framework

(see Figure 2) for selecting cloud services in cloud e-

marketplace.

Figure 2: Proposed framework.

The framework comprises four modules namely:

Cloud ecosystem and service directory, GUI &

visualization, QoS requirement processing, and

Service evaluation & QoS ranking.

In step 0, the atomic services are combined to

realize the set of composite services offered in the e-

marketplace. Subjective QoS requirements are then

provided (step1), processed (step 2), optimized (step

3), and used to rank services in the directory (step

4). The ranked results are shown to the users via

bubble graph visualization (step 5). We shall discuss

each module in details subsequently.

4.1 Cloud Ecosystem and Service

Directory

The framework uses the extended feature model

notations (Benavides et al., 2010), to model the

cloud ecosystem feature model (CEFM). The CEFM

is mapped as a constraint satisfaction problem, and

the Choco-based reasoning engine (www.choco-

solver.org) reasons with a Depth-First search

algorithm to derive all valid mappings. Possible

combinations of atomic services that can be

generated from the pool of atomic services are made

available in the e-marketplace based on former

composition approaches (Akolkar et al., 2012).

4.2 GUI and Visualization

The framework integrates fuzzy-enabled web-based

widgets comprising sliders, drop-down menus and

textboxes for eliciting vague preferences and

aspirations, while bubble graph visualization is

employed to improve understanding of the

relationship among the ranked services.

Users can indicate preferences by pairwise

comparison for each QoS attribute by adjusting the

slider handle (see figure 3). The slider bar has two

colour codes that corresponds to the QoS attributes,

and indicates the level of preference for a QoS

attribute; the lengthier colour means user prefers a

QoS attributes more than the other. The positions of

the slider handle are underlined by fuzzy numbers,

from the fuzzified Saaty scale.

Since humans derive better insight from a picture

faster than mere text, the use of information

visualization to aid service selection improves user

experience (Spence, 2014). The use of bubble graph

visualization improves the understanding of how

each service in the ranking relates with others (see

Figure 4). Each service is represented as a bubble

(shape), using colours, sizes and x-y coordinates to

show services in the QoS information space. These

CLOSER 2017 - 7th International Conference on Cloud Computing and Services Science

606

dimensions (colours, size and x- and y coordinates)

represents four QoS attributes (e.g. availability, cost,

response time, security reliability etc.)

simultaneously and shows the relationship among

the top ranked cloud services; thereby, enabling

effective comparison.

4.3 QoS Requirements Processing

QoS Preference Prioritizer (QPP): The QPP

module ensures consistency in the pairwise

judgment and uses the geometric mean method to

derive priority weights. Defining comparison ratios

as fuzzy numbers are a better way to capture user’s

claim about the relative importance of criteria. To

prioritize user’s QoS preferences, the QPP employs

Fuzzy AHP-based approach. In FAHP, exact

comparison ratio a

is represented as a fuzzy

number, , based on 9 fuzzy linguistic terms

described in the fuzzified Saaty’s scale (Buckley,

1985). The user performs pairwise comparison for

all QoS criteria, which fills the fuzzy comparison

matrix. For example, a user’s degree of importance

of security criterion over availability can be

expressed by the fuzzy number “about strongly

important” . The corresponding

reciprocal from on the fuzzy comparison matrix

becomes . The fuzzy priority vector,

, is obtained by applying the geometric mean

prioritization method (Buckley, 1985).

QoS Aspiration Analyser (QAA): The QAA

module synthetizes user’s QoS values based on

fuzzy decision making, comprising membership

functions framed as fuzzy goal and constraints

(Bellman and Zadeh, 1970). Since the linguistic

terminologies describing the QoS aspiration reflect

the semantic approximations of user’s intent,

resolving the fuzzy decision results in optimal set of

QoS values that approximate user’s QoS intent.

Table 1 shows sample linguistic goals and

constraints for availability QoS attribute.

Table 1: QoS goals and constraints for Availability QoS.

QoS

Linguistic

QoS Goals

Linguistic QoS

Constraints

Availability

High

Medium

Low

Substantial greater than x

In the vicinity of x

About x

Very Close to

Examples of a fuzzy goal and constraints are “The

Availability of the service should be High”, and

“Availability should be About x”; where x is a

specific value as indicated by the user.

4.4 Service Evaluation and QoS

Ranking

QoS Requirements Optimizer (QRP): The QRP

component computes the optimal QoS values that

describe user’s requirements based on the QoS

information of all the services in the service

directory. The inputs into this component are the

priority weights and the value for QoS attributes.

The framework defines two utility functions: a

Simple Addictive Weighting (SAW) function (1)

and exponential Euclidean distance metric (eEUD)

(2), to evaluate the performance of each service

alternative w.r.t user requirements. SAW is used to

determine the QoS properties of the alternative with

the highest utility, while eEUD computes the QoS

properties closest to users’ requirements.

(1)

= j

QoS value of i

service; w

= j

QoS weight.

(2)

and y

are the values of the j

QoS properties of i

cloud service and user requirements respectively.

The results from the two functions are used to

construct the optimal QoS requirements.

QoS Ranking Engine: The output from the QRP

forms the basis for ranking the services in the

directory. The main technique used in this module is

a nearest neighbour algorithm, based on (2). The

output is the top-k services fed into the bubble graph

visualization.

5 ILLUSTRATION

A scenario of a customer relationship management

as a service (CRMaaS) ecosystem and e-marketplace

was reported in (Ezenwoke, 2016). The CRMaaS is

made up of 5 modules (contact manager, database,

marketing, social media analytics and cloud

platform) and 14 atomic services to fulfil the

modules. The modules and atomic services were

modelled using extended feature modelling notations

and a constraint-based reasoning engine is used to

derive a set of 38 valid compositions based on the

constraints guiding the relationship of the modules

and atomic services.

Based on the 38 valid composite services









,

6,7,8



,



































,





























1



Towards a Fuzzy-oriented Framework for Service Selection in Cloud e-Marketplaces

607

reported in (Ezenwoke, 2017), Table 2 shows four

QoS attributes whose ranges are represented by

linguistic variable and underlying membership

functions. Table 3 and Table 4 show a sample user’s

preferences and aspiration over the 4 QoS attributes.

Figure 3 depicts how the user’s available QoS

requirements are captured using the GUI module.

Table 5 contains the sample optimal values

representing user’s QoS requirements. Table 6

shows the ranking of service with respect to user’s

requirements and the bubble graph visualization of

the ranking is shown in Figure 4.

Table 2: QoS Attributes and linguistic variables.

QoS Fuzzy sets

Membership

Function

Availability

Very High, High,

Medium, Low

Trapezoidal

Membership

Function

Response

Time

Low, Acceptable,

Below Average

Reliability

Very High, high,

Average, Low

Cost

Premium, Standard,

Moderate, Cheap

Table 3: User’s QoS Preferences.

QoS Fuzzy Judgement QoS

Availability Extremely more important than Resp.Time

Availability Extremely less important than Reliability

Availability Somewhat Less important than Cost

Table 4: User’s QoS Aspiration.

QoS Goal Constraints

Availability Very High In the Vicinity of 98%

Resp. Time Low Very close to 400ms

Reliability Very High In the Vicinity of 75%

Cost Premium In the Vicinity of 400$

Table 5: Sample of complete user QoS requirements.

S Preference Aspiration

Availability 0.1242 98.49

Resp. Time 0.1237 489.46

Reliability 0.5798 75.43

Cost 0.1724 390.64

Table 6: Top ten services that match user requirements.

Rank ID Avail.

(%)

R. Time

(ms)

Reliab.

(%)

Cost

($/Mon)

1 S3 98.67 546.24 75.43 390.64

2 S17 99.03 546.24 75.43 386.15

3 S10 98.49 546.24 74.72 385.64

4 S35 98.62 489.46 75.72 360.98

5 S19 99.51 559.35 76 390.48

6 S4 97.16 546.24 72.48 381.15

7 S18 97.53 546.24 72.48 376.66

8 S20 98.01 559.35 73.04 380.99

9 S7 98.29 526.12 74.19 354.14

10 S32 98.02 551.35 75.62 360.46

Figure 3: Availability QoS Requirements for User’s

Requirements.

Figure 4: Bubble Graph for Ranked Services for User

Requirements: showing details for Service_ID 35 on

mouse over.

6 DISCUSSION

The proposed framework has the potential to address

some of the limitations that have been observed in

existing cloud selection techniques. These are

elaborated as follows:

Handling of Complex user Requirements that are

Beyond Capability of Individual Atomic Services:

The framework automates the composition of atomic

services to satisfy complex user requirements, and

updates the service directory by capturing scenarios

of new entrants and exists of atomic services. With

our framework, the number of potential composite

offerings can be planned a priori by e-marketplace

provider; similarly, atomic service providers can

drive the competitiveness of their offerings from

CLOSER 2017 - 7th International Conference on Cloud Computing and Services Science

608

knowing the number of composite offerings features

their services.

Enabling Flexibility to Accommodate Subjective

QoS Inputs: In addition, the framework promotes

the interface design of the e-marketplace with user

experience intended. This way, users can easily

express QoS requirements and find optimal

service(s) within the shortest time. The fuzzy-

enabled widgets integrated in our framework allow

users the flexibility of expressing subjective QoS

requirements.

Improved Presentation Format for Search

Results to Aid Decision Making: Having obtained

a ranking of cloud services with respect to user QoS

requirements, the bubble graph used in the

framework enables comparison of the top ranked

services in one single view, thus simplifying the

selection decision compared to tabular listing.

7 CONCLUSIONS

In cloud service e-marketplace, service providers

should be able to join the ecosystem easily, and user

should conveniently express subjective requirements

(Akolkar et al., 2012), and to particularly explore

services without being overwhelmed by so many

choices. The main contribution of this paper is a

cloud service selection framework that incorporates

mechanisms to: 1) compose atomic services on the

fly to satisfy complex users’ requirements; 2) allow

users the flexibility of expressing QoS requirements;

both preferences and aspirations, and do so using

subjective descriptors that is more akin to human

judgment; 3) reduce choice overload by showing

only the top best services in a manner that facilitates

easy comparison for effective decision making. In

the nearest future, we plan to fully operationalize the

framework and evaluate its effectiveness in the

context of a real cloud service e-marketplace.

REFERENCES

Akolkar, R., Chefalas, T., Laredo, J., Peng, C.-S., Sailer,

A., Schaffa, F., Silva-Lepe, I.a. and Tao, T., 2012. The

Future of Service Marketplaces in the Cloud. In IEEE

World Congress on Services, pages 262-269.

Beets, S. and Wesson, J. (2011). Using information

visualization to support web service discovery. In

SAICSIT, pages 11-20.

Bellman, R. and Zadeh, L. (1970). Decision making in

fuzzy environment. MonogemenlScience. 141-164.

Benavides, D., Segura, S. and Ruiz-Cortes, A. (2010).

Automated analysis of feature models 20 years later: A

literature review. Information Systems. 615-636.

Berger, T., Pfeiffer, R.-H., Tartler, R., Dienst, S.,

Czarnecki, K., Wasowski, A. and She, S. (2014)

Variability Mechanisms in Software Ecosystems.

Information and Software Technology, 56(11):1520-

1535.

Buckley, J. (1985) Fuzzy hierarchical analysis. Fuzzy Sets

and Systems, 233-247.

Esposito, C., Ficco, M., Palmieri, F. and Castiglione, A.

(2015). Smart Cloud Storage Service Selection Based

on Fuzzy Logic, Theory of Evidence and Game

Theory. In IEEE Transactions on Computers,

Available: 0018-9340.

Ezenwoke, A., (2017). Using constraint reasoning on

feature models to populate ecosystem-driven cloud

services e-marketplace. Covenant University

Technical Report.[Online]. Available at:

eprints.covenantuniversity.edu.ng/7624/

Gatzioura, A., Menychtas, A., Moulos, V. and Varvarigou,

T. (2012). Incorporating Business Intelligence in

Cloud Marketplaces. In IEEE 10

International

Symposium on Parallel and Distributed Processing

with Applications, pages 466-472.

Millet, I. (1997). The effectiveness of alternative

preference elicitation methods in the analytic hierarchy

process, Journal of Multi-Criteria Decision Analysis,

41-51.

Mirmotalebi, R., Ding, C. and Chi, C. (2012). Modeling

User’s Non-functional Preferences for Personalized

Service Ranking. Service-Oriented Computing.

Mohabbati, B., Gašević, D., Hatala, M., Asadi, M.,

Bagheri, E. and Bošković, M. (2011). A Quality

Aggregation Model for Service-Oriented Software

Product Lines Based on Variability and Composition

Patterns. Service-Oriented Computing, 436-451.

Qu, C. and Buyya, R. (2014). A Cloud Trust Evaluation

System using Hierarchical Fuzzy Inference System for

Service Selection. In 28

International Conference on

Advanced Information Networking and Applications,

pages 850-857.

Rehman, Z, Hussain, F. and Hussainz, O. (2011). Towards

Multi-Criteria Cloud Service Selection. In 5

International Conference on Innovative Mobile and

Internet Services in Ubiquitous Computing, pages 44-

48.

Spence, R. (2014). Information visualization: An

introduction, 3

edition, Springer.

Wittern, E., Kuhlenkamp, J. and Menzel, M. (2012).

Cloud Service Selection Based on Variability

Modeling. Service-Oriented Computing, 127-141.

Yu, Z. and Zhang, L. (2014). QoS-aware SaaS Services

Selection with Interval Numbers for Group User.

Journal of Software. 553-559.

Zadeh, L.A. (1974). The concept of a linguistic variable

and its application to Approximate Reasoning.

Learning systems and intelligent robots, 1-10.

Towards a Fuzzy-oriented Framework for Service Selection in Cloud e-Marketplaces

609