QUALITY OF SERVICE AGGREGATION IN E-BUSINESS

APPLICATIONS

Nabil Fakhfakh, Fr´ed´eric Pourraz and Herv´e Verjus

LISTIC Laboratory, University of Savoie, Chemin de Bellevue, Annecy-Le-Vieux, France

Keywords:

QoS aggregation, e-Business applications, Satisfaction degree measurement, The Choquet Integral.

Abstract:

In e-business applications, enterprises build their processes to achieve their business goals. One of the archi-

tectural models of e-business applications is a service-based approach. This approach consists in orchestrating

the e-services offered by one or several enterprises partners in order to build the desired business processes.

It is important for the enterprises to ensure client satisfaction in order to be more attractive and more com-

petitive. Quality of Service has a signiﬁcant impact on client satisfaction. Therefore, clients need e-business

applications with high Quality of Service to be satisﬁed. In this context, we propose in this paper an approach

that allows clients to measure the satisfaction degree of the services orchestration. This approach takes into

account client’s preferences on QoS attributes and their related dependencies in the measurement of the satis-

faction degree. We treat two examples of services orchestration and show how does the measured satisfaction

allow the client to choose the best one.

1 INTRODUCTION

Electronic business or e-business can be deﬁned as

the use of the technology of the Web to do business.

There is a variety of e-business models. Among them,

we found the Business to Client (B2C) and Business

to Business to Client (B2B2C) (Qi and Huang, 2005).

Enterprises based on these models put much impor-

tance to client’s satisfaction in the development of

their e-business applications. For this purpose, en-

terprises have to provide e-business applications with

high Quality of Service (QoS) to be more competitive

and to reach client’s satisfaction. QoS has a signif-

icant impact on client’s satisfaction and it is closely

related to this latter. In this paper, we propose a

method to measure client’s satisfaction related to QoS

of the e-business application. We are interesting in e-

business applications supported through orchestrated

e-services.

When developing their e-business applications,

enterprises aim to respect client’s QoS expectations

speciﬁed on the services orchestration. QoS expecta-

tions are deﬁned by the upper and lower bounds of

the QoS levels that the services orchestration must

meet to guarantee client’s satisfaction. The upper QoS

level’s bound is denoted desired QoS level (Parasura-

man et al., 1994), and represents the QoS level that

satisfy the best the client. Whereas, the lower QoS le-

vel’s bound is denoted adequate QoS level (Parasur-

aman et al., 1994), and represents the minimum QoS

level that satisfy the client. We denote also the mea-

sured QoS level at run-time and on client’s side by

the perceived QoS level (Parasuraman et al., 1994).

However, during services’ life cycle, perceived QoS

attributes values of the services may change. This

leads to a variation of the perceived QoS level of the

services orchestration into the range of QoS expec-

tations. Besides, the client’s satisfaction of the e-

business application supported through this services

orchestration will be impacted. Therefore, clients ex-

ploiting the e-business application need to know how

much they are satisﬁed in terms of QoS.

Various works dealing with evaluation of ser-

vices orchestration’s QoS attributes exist in the litera-

ture (Cardoso et al., 2002; Jaeger et al., 2005; Rosen-

berg, 2009). Currently, none approach supports dif-

ferent QoS attributes simultaneously in order to pro-

vide a high level information about the QoS of the

overall services orchestration. The satisfaction degree

of services orchestration is a such high level infor-

mation, which facilitates the interpretation of QoS at-

tributes values of the services orchestration.

To measure the satisfaction degree of services or-

chestrations, we will use a Multi-Criteria Decision

Making (MCDM) method that takes only client’s

preferences on the QoS attributes as inputs, neces-

100

Fakhfakh N., Pourraz F. and Verjus H..

QUALITY OF SERVICE AGGREGATION IN E-BUSINESS APPLICATIONS.

DOI: 10.5220/0003519401000110

In Proceedings of the International Conference on e-Business (ICE-B-2011), pages 100-110

ISBN: 978-989-8425-70-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

sary to construct the QoS aggregation model. More-

over, we consider that there are dependencies be-

tween clients’ preferences over QoS attributes. There-

fore, we use an aggregation operator, which is able

to support preferential dependencies. The method we

choose is the MACBETH method (Costa et al., 2005)

extended to the 2-additive Choquet Integral (Cliville

et al., 2007; Mayag et al., 2010).

The remainder of this paper is organized as fol-

lows. Section 2 discusses related work. In Section 3,

we present our approach for the measurement of ser-

vices orchestration satisfaction degree. Section 4 de-

tails a use case of the satisfaction degree measure,

while Section 5 concludes the paper.

2 RELATED WORK

In this Section, we compare our approach to the ex-

isting ones in 2.1, and we overview related works on

QoS evaluation for services orchestrations in 2.2.

2.1 QoS Aggregation Efforts

(Menasc´e, 2003) presents an automatic QoS con-

troller for e-commerce sites. The QoS controller

monitors the site’s workload and determine the best

conﬁguration that meet the site’s QoS requirements.

It executes an algorithm that takes into account the

observed workload, the desired QoS levels, to deter-

mine conﬁguration parameters. The algorithm tries

to ﬁnd the conﬁguration parameters that maximise a

QoS metric deﬁned as:

QoS =

∑

i=1

× f

(∆

)

where n is the number of QoS attributes being aggre-

gated, w

is a relative importance weight assigned to

QoS attribute k (

∑

i=1

= 1), ∆

is a relative devia-

tion of the QoS attribute k deﬁned in a way that the

relative deviation is positive when the QoS attribute

exceeds its requirement and negative otherwise, and

() is an increased function of ∆

. The authors as-

sume that the relative weights (w

) are assigned by

site management. In our mind, assigning directly

weights to QoS attribute is not a trivial task, espe-

cially when we have several QoS attributes. In our

approach,we need only decision makerpreferencesto

determine aggregation model parameters. Moreover,

the authors use linear transformation to normalize ∆

(i.e., to have a value in the interval [0,1]). The bet-

ter the normalized value is (tends to 1), the better the

QoS requirement is met. Therefore, the normalized

value represents the satisfactory value to the related

QoS attribute. Using linear transformation is not very

accurate to model the real satisfaction of the client.

For example, the model of the availability’s satisfac-

tion can be a curve. In our approach (see Section 3),

we do not use linear transformation to normalize QoS

attributes but we compute normalized values based on

informations given by the client (see Figure 6).

In (Szydlo and Zielinski, 2008), authors present

a method for adaptive quality control of services or-

chestrations. The goal of the method is to satisfy

client requirements and to preserve his budget by

changingSLA during execution. The method is based

on a QoS controller that monitors deviation of per-

ceived QoS attributes values from the agreed in the

SLA, and on this basis, services to be invoked are

selected. The client choose a set of SLAs he is in-

terested in, and the system selects the SLA with the

best ﬁtness function and price. The ﬁtness function

is deﬁned as a weighted mean of QoS attributes val-

ues. Authors uses the Analytical Hierarchy Process

(AHP) (Forman and Selly, 2001) method to build the

aggregation model. However, the AHP method does

not take into account dependencies between pref-

erences on QoS attributes since it is based on the

weighted mean operator.

(Herssens et al., 2008) presents an approach for

services selection. The approach is based on a QoS

model that enables users to express their requirements

and providers to represent their services’ QoS. The

model also allows to represent priorities (preferences)

and dependencies between QoS attributes. The selec-

tion mechanism relies on a MCDM method that takes

into account relationships and dependencies between

QoS attributes. The MCDM used is the Choquet

Integral. However, authors assume that the service

provider speciﬁes the dependencies and their effects

(i.e., positive or negative) that can exist between QoS

attributes. In general, dependencies between QoS at-

tributes are due to the techniques used to improve one

or several QoS attributes. In other words, depend-

ing on the technique used, the improvement of a QoS

attribute can affect other QoS attributes in a positive

or negative way. However, we can obtain the desired

QoS (that ﬁt the best the expectations) by using ap-

propriate techniques. In our approach, we assume

that dependencies exist between client’s preferences

over the set of QoS attributes and not between QoS

attributes variables. Authors also use linear transfor-

mation to normalize QoS attributes. As we discussed

above, this assume that the satisfaction of the client is

linear which is not necessarily the case. Our approach

takes only client’s preferences as inputs and allows to

generate normalized QoS attributes values and aggre-

gation operator parameters.

QUALITY OF SERVICE AGGREGATION IN E-BUSINESS APPLICATIONS

101

In our previous work (Fakhfakh et al., 2011),

we have used Measuring Attractiveness by a

Categorical-Based Evaluation TecHnique (MAC-

BETH) method (Costa et al., 2005) based on the

weighted mean operator to aggregate QoS attributes

values. The MACBETH method allows us to ex-

press QoS attributes in the same scale and determine

the weights of the weighted mean operator. The ad-

vantage of MACBETH method is that its inputs are

only restricted to the informations provided by the

client (i.e., preferences and their intensity over QoS

attributes) and does not make any other assumption.

We used this kind of method because we consider

that the satisfaction is client speciﬁc (i.e., varies from

one client to another) and depends on client prefer-

ences. However, the major limitation of the MAC-

BETH method is that it assumes that client’s prefer-

ences on QoS attributes are independent. For exam-

ple, let us consider three QoS attributes : response

time, reliability and availability. One client may ex-

press that he prefers more the availability than the re-

sponse time when the reliability is good. However, he

prefers more the response time than the availability

when the reliability is bad. These kinds of informa-

tion is not supported by MCDM methods based on a

weighted mean operator due to its independence as-

sumption. For this purpose, we use a MCDM method

based on the Choquet Integral operator.

In this paper, we will use an extension of the

MACBETH method to the 2-additive Choquet Inte-

gral (Cliville et al., 2007; Mayag et al., 2010) that

permits to take into account dependencies between

client’s preferences.

2.2 QoS Evaluation for Services

Orchestrations Approaches

Various approaches have been proposed to compute

each QoS attribute value independently for services

orchestration (e.g., giving response time values for all

services composing the orchestration, how to com-

pute the response time of the overall orchestration ?).

These approaches can be classiﬁed in two categories:

probabilistic model-based approaches (Gallotti et al.,

2008)(Satoand Trivedi, 2007)(Cortellessaand Grassi,

2007)(Zhong and Qi, 2006) and workﬂow pattern-

based approaches (Jaeger et al., 2005) (Rosenberg,

2009)(Cardoso et al., 2002) (Coppolino et al., 2007).

The ﬁrst category of approaches allowing to com-

pute QoS attributes values of services orchestration

is based on probabilistic models. It consists in

transforming the services orchestration model into a

probabilistic model (e.g., Continuous Time Markov

Chain (CTMC) model (Gallotti et al., 2008)(Sato

and Trivedi, 2007) or Discrete Time Markov Chain

(DTMC) (Cortellessa and Grassi, 2007) or Stochas-

tic Petri Nets (SPN)(Zhong and Qi, 2006)). Then,

the probabilistic model is annotated with QoS at-

tributes values. Finally, these approaches use tools

like PRISM (Gallotti et al., 2008) or SPNP (Zhong

and Qi, 2006) to compute each QoS attribute value of

services orchestration. The major drawback of these

approaches is that they only support reliability and/or

response time (see Table 1).

The second category consists in deﬁning aggrega-

tion rules of QoS attributes for each pair of workﬂow

patterns (van der Aalst et al., 2003). A pair of work-

ﬂow pattern is composed of ”one split pattern” (e.g.,

AND-split) and ”one join (synchronisation) pattern”

(e.g., XOR-join) except the sequence and the loop

patterns, which are individually considered. The ad-

vantage of workﬂow pattern-based approaches is that

they support larger set of QoS attributes (see Table 1).

Moreover, they are extensible: (i) more workﬂow

patterns could be added and (ii) new QoS attributes

could be integrated. For that reason, in our approach,

we will exploit a workﬂow pattern-based approach.

However, when changes affect positively or neg-

atively some QoS attributes values, it becomes dif-

ﬁcult to estimate how much the whole orchestration

ﬁts client’s expectations and satisfaction ; especially

when we have several QoS attributes. Thus, it would

be useful to have a high-level information. We de-

ﬁne this information as the services orchestration sat-

isfaction degree. This could be done by aggregating

the QoS attributes values to provide a sole aggregated

value that measures the satisfaction degree of the ser-

vices orchestration. To this end, we propose to use a

MCDM method.

3 AGGREGATION PROPOSAL

In this section, we introduce our proposal for aggre-

gating QoS attributes values in order to provide a

measure of the satisfaction degree of services orches-

tration according to client’s QoS expectations. We op-

erationalize the satisfaction degree as the score out

of 1 that represents how much the perceived QoS

level, for the overall services orchestration, respects

QoS expectations (i.e., client satisfaction). It takes

the value of 0 if the perceived QoS level is less than

or equal to the adequate QoS level, and the value of

1 if the perceived QoS level is greater than or equal

to the desired QoS level. As we said above, QoS ex-

pectations are deﬁned by means of desired and ade-

quate QoS levels. We deﬁne a QoS level as a vector

of QoS attributes values denoted (q

, ..., q

) , where

ICE-B 2011 - International Conference on e-Business

102

Table 1: Aggregation Categories vs Supported QoS Attributes.

Category Research work Supported QoS attributes

Probabilistic model-based approaches

(Gallotti et al., 2008) Execution time, reliability

(Cortellessa and Grassi,

2007)

Reliability

(Sato and Trivedi, 2007) Response time, reliability

(Zhong and Qi, 2006) Reliability

Workﬂow pattern-based approaches

(Jaeger et al., 2005) Throughput, response time,

cost, availability, reputation,

security

(Rosenberg, 2009) Throughput, response time,

cost, availability, reputation,

security , scalability, accu-

racy, robustness

(Cardoso et al., 2002) Response time, cost, relia-

bility, ﬁdelity

(Coppolino et al., 2007) Reliability

Figure 1: Principe of the Aggregation Approach.

j, 1≤ j≤n

is the j

QoS attribute value. Therefore, the

desired QoS level is a vector of the desired QoS at-

tributes values, while the adequate QoS level is a vec-

tor of the adequate QoS attributes values. If the client

knows the services involved in the services orchestra-

tion, he may deﬁne his individual QoS expectations

on each service. In this case, QoS expectations on the

services orchestration can be computed by applying a

workﬂow pattern-based aggregation approach.

Our approach is composed of two phases (see Fig-

ure 1): in the ﬁrst phase, we use workﬂow patterns

aggregation rules, while in the second phase we use

a MCDM method based on the 2-additive Choquet

Integral. At execution time, each service S

has a

perceived QoS level (q

, ..., q

)

(obtained from per-

ceived QoS attributes values). Given these perceived

QoS levels of all orchestrated services, they are ﬁrstly

aggregated in phase 1 using workﬂow patterns aggre-

gation rules. This results in one perceived QoS level

of the services orchestration (q

, ..., q

)

orch

(see Fig-

ure 1). Then, the QoS attributes values of this per-

Figure 2: B2C Services Orchestration.

ceived QoS level will be aggregated in phase 2 us-

ing a MCDM method. This provides us a sole and

consolidated value, which is the satisfaction degree

of the services orchestration. Before detailing these

two phases, we present a car insurance e-business ap-

plication supported through a services orchestration

described in Figure 2. We will use this B2C services

orchestration model to illustrate our approach here-

after.

The process starts by asking some informations to

the user (user age, driving license, car typeand model,

etc.). Such informations are ﬁrstly analysed for a de-

cision: either the quotation request is accepted, either

QUALITY OF SERVICE AGGREGATION IN E-BUSINESS APPLICATIONS

103

Table 2: Aggregation Rules (Rosenberg, 2009)(Cardoso et al., 2002).

Response Time Reliability Availability

Sequence

∑

i=1

)

∏

i=1

rel

)

∏

i=1

)

Loop q

) ∗ c q

rel

)

AND-AND max(q

), ..., q

))

∏

i=1

rel

)

∏

i=1

)

XOR-XOR

∑

i=1

)

∑

i=1

rel

)

∑

i=1

)

c denote the number of occurring loops

the probabilities of the outgoing branches for XOR-XOR

it is immediately rejected. When accepted, a quota-

tion is established taken into account user informa-

tions provided. Then, a commercial offer is pack-

aged comprising the car insurancequotationand some

commercial and promotional offers (life insurance,

house insurance, etc.). The commercial offer is sent

to the user.

Hereafter, we assume that QoS expectations (i.e.,

desired and adequate QoS levels) and measurements

(i.e., perceived QoS attributes values) are respectively

given by the client and a monitoring system (QoS at-

tributes measurement is out of the scope of this pa-

per).

3.1 Phase 1: Aggregation based on

Workﬂow Patterns Rules

In the ﬁrst phase (see Figure 1), we use aggregation

rules based on workﬂow patterns (see section 2.2)

to compute each QoS attribute value of the services

orchestration. This consists in applying step-by-step

rules in order to aggregate QoS attributes values. The

applied rules are those corresponding to the pairs of

workﬂow patterns used in the services orchestration

model. Beginning from the most nested pair of work-

ﬂow pattern, the orchestration modelis parsed and ag-

gregation rules for each QoS attribute in the perceived

QoS level are progressively applied. This terminates

when the whole services orchestration is reduced to

a single node (Figure 3). The resulted QoS attributes

values of the resulting node compose the perceived

QoS level of the services orchestration. This approach

is relevant for each QoS attribute that has aggregation

rules for the pairs of workﬂow patterns. We will de-

tail this phase through the illustrative example of the

services orchestration described in Figure 2.

For simpliﬁcation purpose, we will consider a re-

stricted set of three QoS attributes values: response

time (q

), reliability (q

rel

) and availability (q

). The

aggregation rules for each pair of workﬂow patterns

and for each QoS attribute are summarized in Table 2.

The ﬁrst step consists in checking the most nested

workﬂow pattern which is the sequence pattern be-

tween S

, S

and S

in Figure 3a. Then, we apply the

respective aggregation rule from Table 2. The QoS

attributes values computation of this composition pat-

tern gives :

4,5,6

) = q

) + q

)

rel

4,5,6

) = q

rel

).q

rel

).q

rel

)

4,5,6

) = q

).q

)

Thus, the orchestration model is reduced to that

given in Figure 3b. Then, taking into account the

reduced orchestration model, the next workﬂow pat-

tern to be considered is the sequence pattern of S

and

4,5,6

. The QoS attributes values computation of this

composition pattern gives :

3,4,5,6

) = p

4,5,6

) + p

)

rel

3,4,5,6

) = p

rel

4,5,6

) + p

rel

)

3,4,5,6

) = p

4,5,6

) + p

)

The obtained orchestration model from this step is

composed of three nodes structured in sequence (Fig-

ure 3c). By aggregatingQoS attributes values of these

three nodes in sequence, we obtain:

1,2,3,4,5,6

) = q

) + q

3,4,5,6

)

rel

1,2,3,4,5,6

) = q

rel

).q

rel

).q

rel

3,4,5,6

)

1,2,3,4,5,6

) = q

).q

3,4,5,6

)

This resulting values are those that compose the

perceived QoS level of the whole services orches-

tration (q

, q

rel

, q

)

orch

and will be the input of the

phase 2 (Figure 1).

3.2 Phase 2: Aggregation using the

2-Additive Choquet Integral

The goal of this phase is to aggregate different values

in the perceived QoS level of the services orchestra-

tion (i.e., (q

, q

rel

, q

)

orch

) in order to obtain a mea-

sure of the satisfaction degree of the services orches-

tration (q

orch

) (see Figure 1). This measure allows us

ICE-B 2011 - International Conference on e-Business

104

(a) Step 1. (b) Step 2. (c) Step 3. (d) Step 4.

Figure 3: Workﬂow Pattern-Based Aggregation Steps.

Table 3: Example of QoS Levels Illustrating Dependencies.

Response Time Reliability Availability

A 0.7 0.8 0.9

B 0.8 0.75 0.9

C 0.8 0.75 0.5

D 0.7 0.8 0.5

to detect positive or negative deviations that affect the

perceivedQoS levelof the services orchestrationfrom

one execution to another. It is also useful to compare

several services orchestrations having the same busi-

ness goal.

As we discussed above (see Section 2.1), the

weighted mean operator considers that the client’s

preferences on the QoS attributes are independent.

However, they could be some dependencies between

them. As an example, let us consider four QoS lev-

els denoted A, B, C and D presented in Table 3. One

client may express that he prefers the reliability to the

response time when the availability is good (i.e., he

prefers A to B). On the other hand, the same client

may say that he prefers the response time to the relia-

bility when the availability is bad (i.e., he prefers C to

D). These two expressions of preference leads to the

following inequalities:

orch

((0.7, 0.8, 0.9)) > q

orch

((0.8, 0.75, 0.9)

orch

((0.7, 0.8, 0.5)) < q

orch

((0.8, 0.75, 0.5)

These preferences cannot be modelled by a

weighted mean operator. Indeed, the ﬁrst inequal-

ity implies that reliability is more important than re-

sponse time, whereas the second inequality implies

exactly the opposite. This means that the importance

between the reliability and the response time depends

on the satisfaction of the availability. This case is an

example of preferential dependencies between crite-

ria and is not supported by the weighted mean oper-

ator (see (Grabisch and Labreuche, 2005; Grabisch

and Labreuche, 2008) for more details about criteria

dependencies).

For this reason, we have choose the 2-additive

Choquet Integral, which takes into account preferen-

tial dependencies between criteria.

The 2-additive Choquet Integral operator is de-

ﬁned by the following aggregation formula:

orch

∑

i=1

−

∑

j=1

.|q

− q

| (1)

and involves 2 types of parameters :

• Shapley parameters ν

, which are the weights of

each QoS attribute, with

∑

i=1

= 1,

• Interaction parameters I

that quantify mutual in-

teraction between criteria i and j, with I

∈ [−1, 1]

and



−

∑

j=1



≥ 0 ∀i ∈ [1, n] and j 6= i.

These parameters (I

) may be:

– positive, which implies that there is a contradic-

tion between the pairs of criteria. So the aggre-

gated value of QoS attributes (q

orch

) decreases,

– negative, which implies that there is a positive

synergy between the pairs of criteria. Thus, the

aggregated value of QoS attributes (q

orch

) in-

creases,

QUALITY OF SERVICE AGGREGATION IN E-BUSINESS APPLICATIONS

105

– null, which implies that the pairs of criteria are

independent. Therefore, the 2-additive Choquet

Integral becomes equivalent to the weighted

mean operator.

We use an extension of the MACBETH

method (Costa et al., 2005) to construct the model of

the 2-additive Choquet Integral (Cliville et al., 2007;

Mayag et al., 2010), as it takes as inputs only client’s

preferences. It is based on pairwise comparison of

situations made by the decision maker, who is the

client in our case. The MACBETH method comprises

four main steps (Figure 4). As MACBETH method is

based on the weighted mean operator, only the two

latter steps differ since the aggregation operator and

its related parameters are different.

3.2.1 Context Deﬁnition Step

The ﬁrst step consists in identifying the criteria that

the comparison will be based on. In our context, crite-

ria are QoS attributes (e.g., response time, reliability,

availability). Secondly, situations that will be com-

pared are deﬁned. In our case, situations are repre-

sented by QoS levels (i.e., vectors of QoS attributes

values).

If we consider the example of the e-business pro-

cess cited above, the situations to be compared are the

perceived QoS level of the overall services orches-

tration (q

, q

rel

, q

)

orch

(i.e., resulting from work-

ﬂow patterns aggregation rules) in addition to the

desired and adequate QoS levels of the overall ser-

vices orchestration. The desired and adequate QoS

levels are called reference situations in MACBETH

method. They are denoted respectively as good situa-

tion and neutral situation. After normalization, these

two situations correspond respectively to the situa-

tions (1, 1,1) and (0, 0, 0), for which the associated

satisfaction degrees are respectively 1 and 0. So, the

client (the decision maker) has to compare the follow-

ing situations:

good

orch

= (q

good

, q

good

rel

, q

good

)

perceived

orch

= (q

, q

rel

, q

)

orch

neutral

orch

= (q

neutral

, q

neutral

rel

, q

neutral

)

Note that if we have more situations (e.g., previ-

ous services orchestration executions), we can include

them in the deﬁnition context. This provides more

accurate models of QoS attributes normalization (see

Section 3.2.2).

Table 4: Preferences and Preference Strengths for Response

Time.

Good S

perceived

Neutral

Good No h

perceived

No h

Neutral No

0=null, 1=very weak, 2=weak, 3=moderate, 4=strong,

5=very strong, 6=extreme

3.2.2 The QoS Attributes Normalization Step

In this step, the goal is to normalize QoS attributes

values. We do not use linear transformations to nor-

malize them, but we preferably exploit informations

provided by the client. For that purpose, the client

(decision maker) uses his expertise to judge given sit-

uations and fulﬁll the matrix of judgements like the

one given in Table 4. Firstly, he is asked for each QoS

attribute about his preferences between pairs of situ-

ations (including the two reference situations). If the

client prefers situation S

to S

for a QoS attribute k,

this is noted as follows:

≻S

and means that for the normalized QoS attributes val-

ues q

> q

. This is mapped in Table 4 into the clas-

siﬁcation of the situations by their order of preference

depending on the values of the QoS attribute k.

Secondly, the client expresses his strengths of

preference about the difference of attractiveness be-

tween the same situations. The strengths of pref-

erence are characterized with seven levels in the

MACBETH method: 0=null, 1=very weak, 2=weak,

3=moderate, 4=strong, 5=very strong, 6=extreme

(see Table 4). If the client is not able to give his

strengths of preference but only his preferences, this

is noted as positive or more shortly P. For a QoS

attribute k, the client prefers the situation S

to S

with a difference of attractiveness characterized by a

strength h

∈ {0, ..., 6} i.e.,

≻

This is equivalent to :

− q

= h

.α (2)

where α is a coefﬁcient necessary to meet the con-

dition q

and q

∈ [0, 1].

When all the strengths of preference between sit-

uations are provided and the matrix of judgements is

fulﬁlled (e.g., see Table 4), a system of equations can

be extracted. Each strength of preference expressed

on pair of situations gives an equation under the form

of equation 2. By solving this system of equations,

ICE-B 2011 - International Conference on e-Business

106

Figure 4: The Main Steps of MACBETH Method.

the normalized QoS attributes values are quantiﬁed in

the interval [0,1].

Example: for the strengths of preferences ex-

pressed in Table 4 for the response time, the system

of equations is the following:

good

= 1); q

good

− q

= 1 − q

= h

.α

neutral

= 0); q

− q

neutral

= q

− 0 = h

.α

In the above system, the unknown variables are q

and α, as h

, h

∈ {0, ..., 6} and are given by the

client. So, the system of equations can be solved and

the normalized response time of the perceived QoS

level can be computed. Note that the same procedure

is established for each QoS attribute (i.e., reliability

and availability).

3.2.3 Model Parameters’ Determination Step

In this step, we have to determine the parameters

of the 2-additive Choquet Integral, which are ν

and

(See formula 1). In the case of three QoS at-

tributes (q

, q

rel

, q

) or more simply (q

, q

), the

2-additive Choquet Integral takes the form:

orch

= ν

+ ν

−

− q

−

− q

| −

− q

(3)

Once the strengths of preferences are given (e.g.,

see Table 5), a system of equations can be extracted.

Then, the parameters of the aggregation operator can

be computed by solving the system of equations. We

restrict the client to only fulﬁll the ﬁrst diagonal of

the matrix as we consider providing all the strengths

of preferences over all the combinations of pairs of

situations is more complex and a hard task.

For example, the system of equations extracted

from Table 5 is:

(0,1,1)

orch

− q

(1,1,0)

orch

= h

.α = −ν

+ ν

− 0.5I

+ 0.5I

(1,1,0)

orch

− q

(0,1,0)

orch

= h

.α = ν

+ 0.5I

− 0.5I

(0,1,0)

orch

− q

(1,0,1)

orch

= h

.α = −ν

+ ν

− ν

(1,0,1)

orch

− q

(0,0,1)

orch

= h

.α = ν

− 0.5I

+ 0.5I

(0,0,1)

orch

− q

(1,0,0)

orch

= h

.α = −ν

+ ν

+ 0.5I

− 0.5I

(1,0,0)

orch

− q

(0,0,0)

orch

= h

.α = ν

− 0.5I

+ ν

= 1

The system can be put in a matrix form. There-

fore, it is resolvable only and only if the matrix is

non-singular.

The client can verify if the computed weights (ν

)

and the interaction parameters (I

) corresponds to his

preferences. This can be done by computingthe satis-

faction degrees of the binary situations and verifying

if they are conform with his preferences. Otherwise,

he can modify his strengths of preferences for best

translation of his preferences.

3.2.4 Aggregation Step

The QoS attributes values being normalized in Sec-

tion 3.2.2 and the Choquet Integral parameters being

computed in Section 3.2.3, we can now aggregate the

perceived QoS level of the services orchestration by

applying formula (3). The resulting value from the

aggregation represents the satisfaction degree of the

client from QoS point of view.

In the next section, we present a use case of the

satisfaction degree following this approach.

QUALITY OF SERVICE AGGREGATION IN E-BUSINESS APPLICATIONS

107

Table 5: Client’s Judgements for Choquet Integral’s Parameters Determination.

(0, 1, 1) (1, 1, 0) (0, 1, 0) (1, 0, 1) (0, 0, 1) (1, 0, 0) (0, 0, 0)

(0, 1, 1) No h

P P P P P

(1, 1, 0) No h

P P P P

(0, 1, 0) No h

P P P

(1, 0, 1) No h

P P

(0, 0, 1) No h

(1, 0, 0) No h

4 USE CASE

E-business applications is a growing area and we can

found several e-business applications in the market re-

lating to the same business goal. We have presentedin

the previous section a B2C services orchestration al-

lowing the client to request for an insurance-car quo-

tation. Let us consider another services orchestration

satisfying the same business goal, but differs slightly.

The second services orchestration follows the model

of B2B2C applications as it involves many enterprises

(see Figure 5). These enterprises collaborate together

in order to provide to the client the best offer. Thisser-

vices orchestration is roughly the same as the already

presented above (see Figure 2) but differs when the

quotation request is accepted. In this latter case, the

quotation request is sent to two car-insurance subcon-

tractors and partners (enterprise B and C in Figure 5),

each of them establishing a quotation. Note that the

enterprises B and C do not have the same process to

establish quotations. When all quotation proposals

are received by enterprise A, they are submitted for

analyse and comparison. Then, a commercial offer

is packaged comprising the best insurance quotation

and some commercial and promotional offers as al-

ready discussed in Section 3. Finally, the commercial

offer is sent to the client.

We will now compute the satisfaction degree of

each services orchestration using the provided infor-

mations from the same client. This allows the client

to discriminate the two e-business applications and

choose the best one from his QoS point of view. Let

us consider that the client’s QoS expectations of these

e-business applications are:

good

= desired QoS level = (60, 1, 1)

neutral

= adequate QoS level = (120, 0.7, 0.6)

We will not detail the workﬂow pattern-based aggre-

gation phase but we give directly the perceived QoS

levels of the two services orchestration. We denote

Figure 5: B2B2C services orchestration.

the perceived QoS level of the ﬁrst e-business appli-

cation (B2C) as S

, while the perceived QoS level of

the second e-business application (B2B2C) is denoted

as S

= (q

, q

rel

, q

) = (78, 0.8, 0.95)

= (q

, q

rel

, q

) = (95, 0.9, 0.9)

Given these four situations S

good

, S

neutral

, S

the client has to compare them and express his

strengths of preferences over the difference of attrac-

tiveness between them for each QoS attribute (we

omitted details due to lack of space).

This allows us to normalize QoS attributes values

of each situation. The normalized QoS levels of situ-

ations S

and S

are as follows:

ICE-B 2011 - International Conference on e-Business

108

60 80 100 120

0.2

0.4

0.6

0.8

Perceived Response Time

Normalized Response Time

Response Time Normalization

0.6 0.7 0.8 0.9 1

0.2

0.4

0.6

0.8

Perceived Availability

Normalized Availability

Availability Normalization

0.7 0.8 0.9 1

0.2

0.4

0.6

0.8

Perceived Reliability

Normalized Reliability

Reliability Normalization

Consumer Informations

Linear Equations

Figure 6: QoS Attributes Normalization.

Normalized S

= (

)

Normalized S

= (

)

As we discussed above, we do not use linear equa-

tions to normalize QoS attributes but we are based

only on client’s informations (i.e., preferences and

strengths of preferences on QoS attributes). The use

of linear equations takes only into account the satis-

factory value (i.e., the best QoS attribute value) and

the unsatisfactory value (i.e., the lower QoS attribute

value) and assume that the client’s satisfaction is lin-

ear between them. This assumption is not very ac-

curate. Figure 6 shows that the client’s satisfaction

based on his provided informations can be non-linear

(curve in continuous line). We consider that this

method models the best the client’s satisfaction.

QoS attributes values being normalized, we have

to compute the parameters ν

and I

to build the QoS

aggregation model based on the 2-additive Choquet

Integral. To this end, the client has to compare ﬁctive

situations and expresses his strengths of preferences

on the difference of the attractiveness between them.

This leads to the following computed parameters:

= 0.125, ν

= 0.55, ν

= 0.325

= 0.1, I

= 0.05, I

= 0.2

By applying the formula 3 at the normalized sit-

uations S

and S

, we get the following satisfactions

degree:

orch

) = 0.2535, q

orch

) = 0.4476

According to these measured satisfaction degrees, it is

clear that the client is more satisﬁed with the second e-

business application (B2B2C) despite the fact that its

services orchestration model contains more services.

This is explained by the fact that this client has put

much more importance at the reliability than the re-

sponse time (ν

> ν

5 CONCLUSIONS

This paper presents an approachthat measuresthe sat-

isfaction degree of services orchestrations related to

the client’s QoS expectations. The approach newly

combines workﬂow patterns aggregations rules and

the 2-additive Choquet Integral. We have shown how

does the measured satisfaction degree allow the client

to discriminate several e-business applications having

the same business goal. This permits to the client

to choose the best satisfactory e-business application

from his QoS point of view. Moreover, for one e-

business application, this approach allows us to detect

the deviation of its perceived QoS level by comparing

the measured satisfaction degree with that resulting

from the previous e-business application’s executions.

The presented approach exploits the services’ per-

ceived QoS levels that are all obtained along the exe-

cuted/runtimed services orchestration. Thus, the sat-

isfaction degree is obtained when the execution termi-

nates. Our future work aims to evaluate the satisfac-

tion degree of the services orchestration throughout

its execution and predicts the deviation of perceived

QoS level inside QoS expectations range. Whenever

we detect potential deviation, we will try to recover

it by dynamically adapting/modifyingthe services or-

chestration.

ACKNOWLEDGEMENTS

This work is partially funded by the FEDER MES

project granted by the French Rhˆone-Alpes Region.

REFERENCES

Cardoso, J., Miller, J., Sheth, A., and Arnold, J. (2002).

Modeling quality of service for workﬂows and web

QUALITY OF SERVICE AGGREGATION IN E-BUSINESS APPLICATIONS

109

service processes. Journal of Web Semantics, 1:281–

308.

Cliville, V., Berrah, L., and Mauris, G. (2007). Quan-

titative expression and aggregation of performance

measurements based on the macbeth multi-criteria

method. International Journal of Production Eco-

nomics, 105(1):171–189.

Coppolino, L., Romano, L., Mazzocca, N., and Salvi, S.

(2007). Web services workﬂow reliability estimation

through reliability patterns. Security and Privacy in

Communications Networks and the Workshops,.

Cortellessa, V. and Grassi, V. (2007). Reliability model-

ing and analysis of service-oriented architectures. In

Test and Analysis of Web Services, pages 339–362.

Springer-Verlag.

Costa, C. B., Corte, J., and Vansnick, J. (2005). On the

mathematical foundation of MACBETH. In Multiple

Criteria Decision Analysis: State of the Art Surveys,

volume 78 of International Series in Operations Re-

search & Management Science, pages 409–437.

Springer New York.

Fakhfakh, N., Verjus, H., Pourraz, F., and Moreaux, P.

(2011). Measuring the satisfaction degree of quality

attributes requirements for services orchestrations. In

in The Fourth International Conference on Commu-

nication Theory, Reliability, and Quality of Service

(CTRQ 2011), pages 89–94.

Forman, E. H. and Selly, M. A. (2001). Decision by objec-

tives: how to convince others that you are right. World

Scientiﬁc Pub Co Inc.

Gallotti, S., Ghezzi, C., Mirandola, R., and Tamburrelli,

G. (2008). Quality prediction of service composi-

tions through probabilistic model checking. In QoSA

’08: Proceedings of the 4th International Conference

on Quality of Software-Architectures, pages 119–134,

Berlin, Heidelberg. Springer-Verlag.

Grabisch, M. and Labreuche, C. (2005). Fuzzy measures

and integrals in mcda. In Multiple Criteria Decision

Analysis: State of the Art Surveys, volume 78 of Inter-

national Series in Operations Research & Man-

agement Science, pages 563–604. Springer New York.

Grabisch, M. and Labreuche, C. (2008). A decade of ap-

plication of the choquet and sugeno integrals in multi-

criteria decision aid. 4OR: A Quarterly Journal of Op-

erations Research, 6:1–44.

Herssens, C., Jureta, I. J., and Faulkner, S. (2008). Captur-

ing and using qos relationships to improve service se-

lection. In Proceedings of the 20th international con-

ference on Advanced Information Systems Engineer-

ing, CAiSE ’08, pages 312–327, Berlin, Heidelberg.

Springer-Verlag.

Jaeger, M. C., Rojec Goldmann, G., and Muhl, G. (2005).

Qos aggregation in web service compositions. In

EEE ’05: Proceedings of the 2005 IEEE International

Conference on e-Technology, e-Commerce and e-

Service (EEE’05) on e-Technology, e-Commerce and

e-Service, pages 181–185, Washington, DC, USA.

IEEE Computer Society.

Mayag, B., Grabisch, M., and Labreuche, C. (2010). A

characterization of the 2-additive choquet integral

through cardinal information. Fuzzy Sets and Systems,

In Press, Corrected Proof:–.

Menasc´e, D. A. (2003). Automatic qos control. IEEE In-

ternet Computing, 7:92–95.

Parasuraman, A., Zeithaml, V. A., and Berry, L. L. (1994).

Alternative scales for measuring service quality: A

comparative assessment based on psychometric and

diagnostic criteria. Journal of Retailing, 70(3):201 –

230.

Qi, M. and Huang, X. (2005). The design and analysis

of three-dimensional e-business model. In Proceed-

ings of the 7th international conference on Electronic

commerce, ICEC ’05, pages 136–138, New York, NY,

USA. ACM.

Rosenberg, F. (2009). QoS-Aware Composition of Adap-

tive Service-Oriented Systems. PhD thesis, Technical

University Vienna, Austria.

Sato, N. and Trivedi, K. S. (2007). Stochastic modeling

of composite web services for closed-form analysis of

their performance and reliability bottlenecks. In IC-

SOC ’07: Proceedings of the 5th international confer-

ence on Service-Oriented Computing, pages 107–118,

Berlin, Heidelberg. Springer-Verlag.

Szydlo, T. and Zielinski, K. (2008). Method of adaptive

quality control in service oriented architectures. In

Heidelberg, S. B. ., editor, Computational Science

ICCS 2008, volume 5101, pages 307–316. Springer

Berlin / Heidelberg.

van der Aalst, W., ter Hofstede, A., Kiepuszewski, B., and

Barros, A. (2003). Workﬂow patterns. In Distributed

and Parallel Databases, volume 14 (1), pages 5–51.

Zhong, D. and Qi, Z. (2006). A Petri net based approach for

reliability prediction of web services. In OTM Work-

shops (1), pages 116–125.

ICE-B 2011 - International Conference on e-Business

110