An Implementation of a QoE Evaluation Technique Including Business

Model Parameters

Diego Rivera

1

, Ana R. Cavalli

1

, Natalia Kushik

1

and Wissam Mallouli

2

1

SAMOVAR, T

´

el

´

ecom SudParis, CNRS, Universit

´

e Paris-Saclay, 9 rue Charles Fourier 91011, EVRY, France

2

Montimage, Paris, France

Keywords:

Quality of Experience (QoE), Business Model, QoE Evaluation, Extended Finite State Machine (EFSM).

Abstract:

The expansion of Internet-based services has increased the need to ensure a good quality on them. In this con-

text, a preliminary work we developed exposes a Quality of Experience (QoE) evaluation framework based

on the mathematical formalism of EFSMs, which includes business-related variables into the prediction anal-

ysis. In this paper, we present an implementation of this QoE evaluation framework using the Montimage

Monitoring Tool (MMT). The implementation presented in this paper is based on three main algorithms: (1)

generation of the traces of a given length of the EFSM-based OTT model, (2) computation of the QoE for

each trace using a suitable QoE model, and (3) computation of the number of conﬁgurations reachable from

the initial state of the EFSM. We use this implementation to calculate the amount of conﬁgurations captured

by the model of a real OTT service, analyzing how this value varies with respect to the depth (trace length) of

the analysis and which is the distribution of the QoE values of the computed conﬁgurations. This information

will enable the service provider to characterize the QoE of all possible scenarios and to introduce changes if

required, in order to maximize the revenues provided by the chosen business model and the QoE of end-users.

1 INTRODUCTION

With the expansion of Internet as an effective media

to transport data, new types of services have found a

business opportunity. In this sense, the Internet-based

services arose as a competitive alternative to the tradi-

tional telecommunication services commonly offered

(and distributed) by the major actors of the telecom-

munication market. A particular example of them are

the multimedia services. New companies in the mar-

ket, such as Skype or Netﬂix, have proposed a new

offer with more competitive prices in comparison to

the ones offered by traditional telcos. This business

strategy has led them to gain an important portion of

the market in a short time.

This last fact can be explained mainly by the use

of Internet as the main distribution method for their

services, excluding the network operators from any

revenues of these services. This is the main feature

that deﬁnes what an Over-The-Top Service (OTT) is,

i.e. services offered using Internet as the distribution

platform, but without involving the network operators

in the business. It is important to remark that the use

of Internet to distribute content allows an easy and

fast deployment of the service. In addition, it presents

the advantage of avoiding the costs of building and

managing the distribution network, but it introduces

other difﬁculties to the distribution process.

The Internet was conceived as a best effort net-

work, meaning that the delivery of the data is not en-

sured. Some network protocols such as TCP try to

ﬁx this, but they do not ensure any quality parame-

ters, typically required in multimedia services as, for

example, audio or video streamings.

In this paper

1

, we analyze the quality concept

starting from three points of view: the technical posi-

tion, related with the quality of the network itself; the

view of the user, which is more related to the qual-

ity level experienced and the fulﬁllment of his/her ex-

pectations; and ﬁnally, the view of the business in-

vestor, which is interested in the maximization of the

revenues. For each one of these actors, we use a par-

ticular quality concept that is integrated into the eval-

uation framework of a real OTT case study.

This work expands the previous research by in-

troducing an implementation of the Quality of Ex-

perience Evaluation Framework published in (Rivera

1

This work has been developed in the frame of the Celtic+

Project NOTTS http://projects.celticplus.eu/notts/

138

Rivera, D., Cavalli, A., Kushik, N. and Mallouli, W.

An Implementation of a QoE Evaluation Technique Including Business Model Parameters.

DOI: 10.5220/0006005001380145

In Proceedings of the 11th International Joint Conference on Software Technologies (ICSOFT 2016) - Volume 2: ICSOFT-PT, pages 138-145

ISBN: 978-989-758-194-6

Copyright

c

2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

et al., 2015), which is based on the Extended Fi-

nite State Machines (EFSMs) mathematical formal-

ism. The implementation is based on the Montimage

Monitoring Tool (MMT) that allows modeling and

simulating the execution of EFSMs. This tool was

modiﬁed by introducing three algorithms: (1) the es-

timation of the l-equivalent of an EFSM and the traces

(paths) deﬁned in this l-equivalent; (2) the calculation

of the QoE of each path by applying an appropriate

QoE evaluation model; and (3) the computation of the

number of conﬁgurations of the machine – namely

“user scenarios” – in order to analyze the effective-

ness of representing an OTT service as an EFSM.

The rest of the paper is structured as follows. Sec-

tion 2 contains the preliminaries, including a brief de-

scription of the QoE evaluation framework this work

is based on. Section 3 introduces the algorithms that

implement the QoE framework presented in the pre-

liminaries, being all of them implemented as an ex-

tension of the Montimage Monitoring Tool. Section 4

presents the analysis of a real case of study (the beIN

Sports Connect Service) including the results and the

discussion of the test. Finally, Section 5 presents the

conclusions of this work.

2 PRELIMINARIES

2.1 OTT Services

A formal deﬁnition of an Over-The Top Service

(OTT) was provided by Green et al., who deﬁned it as

a “service that delivers value to the customers without

involving any carrier in the planning, selling, provi-

sioning or servicing of the offer and, of course, with-

out involving any traditional telco in the revenues of

these services” (Green et al., 2007). From this def-

inition, it is important to remark that three main ac-

tors involved in the whole scenario are: (1) the ser-

vice providers, who provide the content and make it

available to the customers via Internet; (2) the users

or customers, who are the target people that will con-

sume the content provided by the OTT provider; and

(3) the network operators (Internet Service Providers

– ISPs), who are the “third-party” companies that are

in charge of the technical administration of the net-

work and its commercialization. The latter are the

ones that provide the access to Internet to both the

OTT providers and the ﬁnal customers.

Despite the ISPs are the only actors that are in

charge of conducting the content from the service

provider to an end-user, they do not take any part

of the revenues of the service. In addition, the OTT

trafﬁc usually represents a huge load for the network

components, which raises the complexity of manag-

ing the network and, therefore, the maintenance costs.

This fact has led the ISPs to push the OTT providers to

share their revenues, even by imposing restrictions to

the OTT trafﬁc, such as limiting the bandwidth avail-

able for these types of services.

2.2 Quality in Different Dimensions

Starting from its conception, Internet was designed as

a best-effort network, meaning that the delivery of the

data is not guaranteed for any service that uses this

network as the distribution mechanism. However, the

arrival of more complex services arose the issue of

ensuring a good level of quality, concept which deﬁ-

nition depends on the point of view from which it is

being analyzed (Gozdecki et al., 2003).

The quality concept can be deﬁned, as seen from

an engineer, as a set of measurable, technical param-

eters (Gozdecki et al., 2003). In this case, the concept

of Quality of Service (QoS) commonly refers to these

types of parameters that can be obtained by direct

measuring the network. This deﬁnition is aligned with

the main goals of this work, given that the framework

proposes a methodology to map these values into the

user experience of a given service. Typical examples

of these variables are the delay or packet loss, all of

them are usually easy to measure or monitor at any

point of the network.

Even when the given deﬁnition tries to widely

cover all the aspects of quality, it does not cover the

totality of possible dimensions of the quality concept.

As stated by the International Organization for Stan-

dardization (ISO), quality is deﬁned as “(the) degree

to which an inherent characteristics (a distinguishing

feature) fulﬁlls requirements (a need or expectation

that is stated, generally implied or obligatory)” (ISO,

2015). In this deﬁnition, the expectation is also a part

of the concept, thus it is important to consider the

quality deﬁnition from the point of view of the user

in order to have a broad, comprehensive conception.

This idea was captured by Le Callet et al. with

the deﬁnition of Quality of Experience (QoE) as “the

degree of delight or annoyance of the user of an appli-

cation or service. It results from the fulﬁllment of his

or her expectations with respect to the utility and/or

enjoyment of the application or service in the light

of the user’s personality and current state” (Le Cal-

let et al., 2013). This has become the most accepted

description of QoE since it captures the fact that the

experience of a service is a subjective opinion aligned

with user’s expectations. Based on this deﬁnition, we

will understand the QoE as the quality level subjec-

tively perceived by a user, which is related with the

An Implementation of a QoE Evaluation Technique Including Business Model Parameters

139

fulﬁllment of his/her expectations.

At this point, we have analyzed the quality deﬁni-

tion given in (ISO, 2015) as seen from the engineer’s

and user’s point of view. However, it is also possible

to understand it from the point of view of the business.

This approach was formalized in (Van Moorsel, 2001)

with the deﬁnition of the Quality of Business (QoBiz)

as “all of the parameters that can be expressed in mon-

etary units”. In this study, Van Moorsel identiﬁes a

direct relation between the QoE and QoBiz, based on

pricing schema of the service and the willingness to

pay of the customer (Van Moorsel, 2001). These rela-

tionships have been reafﬁrmed by Liao et al., stating

that customers make comparisons between the price

and their expectations with their previous experiences

with similar services (Liao et al., 2015). Having this

in mind, we will use this fact to consider the pric-

ing schema as the main QoBiz parameter of the QoE

analysis, in order to study how these types of busi-

ness decisions can inﬂuence the user’s expectations

and, therefore, the QoE.

2.3 Extended Finite State Machines

Multiple deﬁnitions of this concept have been given in

the literature, being most of them more related with

mathematical formalisms that go beyond the scope

of this work. In the context of this work, we will

use the following deﬁnition given in (Petrenko et al.,

2004). Given X (the set of inputs), Y (the set of out-

puts), R (the set of parameters) and V (the set of con-

text variables), we denote R

x

⊆ R the set of input pa-

rameters and D

R

x

the set of valuations (as vectors) of

these parameters for an input x ∈ X. Similarly, for

an output y ∈ Y we deﬁne the set of output parame-

ters and their valuations R

y

and D

R

y

. Finally, D

V

de-

notes a set of vectors of context variables valuations

v. Being this said, an Extended Finite State Machine

(EFSM) M over X, Y, R, V is a pair (S, V ) of a ﬁnite

set of states S and a ﬁnite set of transitions T be-

tween states in S, such that each transition t ∈ T is

a tuple (s, x, P, op, y, up, s

0

) where: s, s

0

∈ S are the ini-

tial and ﬁnal state of the transition respectively; x ∈ X

is the input of the transition; y ∈ Y is the output of the

transition; P, op and up are functions deﬁned over

input parameters and context variables in V , where

P : D

R

x

× D

V

→ {Tr ue, False} is the predicate of the

transition, op : D

R

x

×D

V

→ D

R

y

is the output parame-

ter function of the transition, and up : D

R

x

×D

V

→ D

V

is the context update function of the transition.

An example of an EFSM is presented in

Figure 1 where we can identify, for example,

the set of inputs X = {connect, option, lo-

gin

credentials, validate, user card data, “live”,

stop stream, “home button”}, the set of states

S = {Idle, Choosing subscription or login, wait-

ing for personal data, waiting for card data, ser-

vice ready, stream delivery, stop stream}, and,

for example, a transition t = (stream delivery,

NULL, ‘if(stream ﬂag ==1)’,

/

0, stream data, f

1

,

stream delivery).

2.4 Business Model Aware QoE

Framework

As stated in the previous paragraphs, this paper aims

to integrate business-related parameters into the eval-

uation of the QoE. This new approach required the de-

velopment of a new QoE evaluation framework ﬂexi-

ble enough to include such kind of new variables that

can inﬂuence the perceived quality of an OTT service.

The framework is used to analyze and calculate

the QoE of an OTT service. An EFSM is used to rep-

resent: the stages of the user-service interaction, the

inputs given to the service and the outputs to an end-

user. This constitutes the ﬁrst step of the framework,

producing a preliminary model of the service that rep-

resents the functional and some non-functional re-

quirements of the service.

This preliminary model is then augmented in or-

der to include the quality parameters that will be an-

alyzed. This augmentation is done through introduc-

ing context variables in the machine representing the

quality parameters. To accomplish this step, it is re-

quired to provide: (1) the speciﬁcation about how

these variables are measured and, (2) how and where

in the model their values are updated. This second

step ﬁnalizes with an augmented model, establishing

how the quality indicators are updated at each step.

Finally, the model can be used to analyze the QoE

of the service. To achieve this, one can calculate the

l-equivalent form of the model, which shows all the

possible end-user scenarios reachable from the initial

state to a ﬁxed length l as branches of a tree-shaped

model. At each one of these branches it is possible to

apply a proper QoE model that correlates the values

of the context variables in order to obtain the value of

the QoE of the branch.

In this paper, we use this approach to model and

augment a real OTT service, namely beIN Sports

Connect. The augmented EFSM is shown in Figure 1.

Further details about how to obtain this model can be

found in (Rivera et al., 2015).

In the model used in this work, we consider three

groups of quality parameters: objective, subjective

and business-related parameters. With these variables

we aim to model the stream state of the service (the

video is being streamed or not), the conﬁdence of the

ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends

140

Figure 1: Augmented EFSM for beIN Sports Connect Service.

Table 1: Summary of the quality parameters/variables for

the beIN Sports Connect Service.

Parameter type Variable Name Values Weight

Objective stream ﬂag {0, 1} 0.5

Subjective service conﬁdence {0, 1, 2} 1.0

Business-related price premium {0, 1, 2} 1.0

user (the level of trust of the user with the service),

and the willingness to pay of the user (namely the

“loyalty” of the user with the service). For each one

of them, we consider a single discrete context variable

taking at most three values. In Table 1 we present a

summary of the quality parameters considered for the

beIN Sports Connect Service.

3 QoE FRAMEWORK

IMPLEMENTATION

The framework presented in the Section 2.4 was de-

signed to provide a mechanism to evaluate the QoE

using EFSMs while considering how the business as-

pects of the service impact the QoE value. The im-

plementation is formed by three main algorithms pre-

sented in the following sections.

3.1 QoE Evaluation Algorithms

3.1.1 Generation of the l-equivalent

As stated before, once the service is modeled as an

EFSM and augmented with quality parameters, one

can generate its l-equivalent form before starting any

further analysis. This process is performed by the ap-

plication of Algorithm 1.

This procedure analyzes recursively the provided

EFSM, traversing the graph using a Depth-First

Search (DFS) strategy. At each step, the algorithm

analyzes each transition of the actual state, determines

if the conditions for the transition stand, and uses the

updating functions to change the variable values for

the next step. The rest of the paths are calculated re-

cursively starting from the current state, updating the

length of the analysis and the context variable val-

ues. Finally, the obtained paths are augmented with

the data calculated for the current step, and the re-

sult is returned. In this algorithm, the base case is

reached when the procedure is called with a length of

0, returning the vector variable and the current state

reached.

An Implementation of a QoE Evaluation Technique Including Business Model Parameters

141

Algorithm 1: l-equivalent generation.

function BUILDPATHS(state, len, vars)

if len ≤ 0 then

return {(state, vars)}

end if

paths ← {}

for all t in transitions[state] do

step ← (state, vars)

stand ← true

for all cond in t.conditions do

stand ← stand && cond(vars)

end for

if !conditionsStand then

continue

end if

for all f in t.updateFunctions do

for all var in vars do

if var.name == f .name then

var.val ← f . f unc(var.val)

end if

end for

end for

lowPaths ← buildPaths(t.to, len − 1, vars)

for p in lowPaths do

paths ← paths ∪ {[p, step]}

end for

end for

return paths

end function

3.1.2 Computation of the QoE

Once the l-equivalent of the EFSM is derived, the next

step is to calculate the QoE for each trace. This is

completed by the procedure shown in Algorithm 2.

Algorithm 2: QoE Evaluation of the traces.

function EVALQOE(traces, qoeModel)

results ← {}

for all trace in traces do

tr

f

← []

for all s in trace do

qoeVal ← qoeModel(s.vars)

tr

f

← tr

f

∪ (s.state, s.vars, qoeVal)

end for

results ← results ∪ {tr

f

}

end for

return results

end function

The proposed method iterates over every path pre-

viously calculated, computing the QoE at each step of

the path using the given QoE model as a mathemati-

cal function. In our particular case, we use a linear

combination as such mathematical function. Finally,

this algorithm returns the set of all paths (the same in-

formation that was used as input) augmented with the

new information of the computed QoE.

Algorithms 1 and 2 can also be used in combi-

nation with active monitoring and DPI techniques.

In (Rivera and Cavalli, 2016), we present the design

of an MMT extension that identiﬁes the actual stage

of the user-service interaction in the model and the ac-

tual set of values of the quality parameters. With this

information, it is possible to use both Algorithms 1

and 2 in order to predict the future scenarios for an

end-user and their corresponding QoE values.

3.2 Concrete Implementation

In order to analyze a real case study, it is required

to implement the algorithms presented below using a

tool that allows: (1) representing an OTT service as

an EFSM model and, (2) interact with the OTT ﬂow

in order to extract the information about the values of

the quality parameters.

3.2.1 Montimage Monitoring Tool (MMT)

The MMT tool is an online monitoring solution that

provides real-time visibility of network trafﬁc, ap-

plication communication, ﬂows and usage levels. It

facilitates network security, performance monitoring

and operation troubleshooting. MMT rules engine

can correlate network and application events in order

to detect operational, security and performance inci-

dents or to generate new events.

MMT is composed of three complementary, but

independent, modules. First, MMT-Probe is the core

packet capture and extraction module. It analy-

ses network trafﬁc using Deep Packet/Flow Inspec-

tion (DPI/DFI) techniques and also allows analyzing

any structured information generated by applications

(e.g., traces, logged messages, simulated events).

Second, MMT-Correlation is an analysis engine based

on formal properties that analyzes and correlates net-

work and application events to detect operational and

security incidents or derive new events. Third, MMT-

Reporting is a visualization application that allows

collecting and aggregating analysis reports to present

them via a graphical user interface.

3.2.2 Implementation Details

The architecture of the MMT tool described below

allowed us to take advantage of its DPI/DFI tech-

niques in order to perform an online analysis of the

OTT ﬂow. The events generated by the MMT-Probe

module – which contain information about the OTT

ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends

142

stream – are analyzed by the MMT-Correlation mod-

ule and the results are ﬁnally presented using the

MMT-Reporting module. Following this schema, our

implementation will take advantage of two main char-

acteristics of this architecture.

Firstly, the DPI capabilities of MMT-Probe will

allow recognizing the values of the quality parameters

and the actual stage of the user-service interaction.

This information will be useful to automatically pre-

dict future conﬁgurations of the machine and, there-

fore, future values of the QoE.

Secondly, the MMT-Correlation module has been

implemented using the Node.js technology (based on

the Javascript language), allowing a clear representa-

tion of the EFSM model using the Javascript format.

4 ANALYSIS OF AN OTT

SERVICE

As stated below, the augmentation of the service

model with discrete and ﬁnite quality parameters in-

troduces a maximum number of EFSM conﬁgura-

tions. However, the predicates on the transitions and

the initial values of the context variables set a limit on

the valid conﬁgurations for the OTT service. In order

to present the advantage of this feature, we empiri-

cally compute the amount of different conﬁgurations

contained in the model. Using this information it will

be possible to characterize each reachable conﬁgura-

tion and its QoE value, showing a distribution of the

QoE values in all the possible conﬁgurations.

4.1 Computation of the Total Number

of Conﬁgurations of the Model

As mentioned before, we will use the conﬁgurations

captured in the EFSM in order to analyze the OTT

service. For this goal, each conﬁguration is based

on two main features of the EFSM: the states of the

OTT model and the context variables of the machine

representing the quality parameters. This completely

deﬁnes a conﬁguration of the machine as the tuple

C = (s, v) where s ∈ S is a state of the machine, and

v ∈ D

V

is the vector of the values of context vari-

ables. Each tuple is related to a speciﬁc QoE value

at a speciﬁc stage of the user-service interaction, thus

by computing the whole set of conﬁgurations of the

model it will be possible to predict all the possible

QoE values for the service. This analysis allows iden-

tifying the conﬁgurations where the QoE value is low,

which is useful information for the service provider in

order to determine how to optimize the resource dis-

tribution, maximizing both the revenues of the busi-

ness and the quality of an end-user.

Algorithm 3: Computation of the total number of conﬁgu-

rations.

function COMPUTECONFS(len, state, vars, con f s)

if len ≤ 0 then

return con f s = con f s ∪ {(state, vars)}

end if

for all t in transitions[state] do

stand ← true

for all cond in t.conditions do

stand ← stand && cond(vars)

end for

if !conditionsStand then

continue

end if

for all f in t.updateFunctions do

for all var in vars do

if var.name == f .name then

var.val ← f . f unc(var.val)

end if

end for

end for

buildPaths(t.to, len − 1, vars)

end for

end function

For a general EFSM representing any OTT model,

the cardinality of the sets previously mentioned, can-

not be estimated in advance, thus the number of possi-

ble conﬁgurations represented with the machine can-

not be determined in a general way. However, it

is possible to reﬁne this analysis using a concrete

instance of an OTT model, represented by a ﬁxed

amount of states and context variables. In this sense,

modeling a real OTT service by using the proposed

methodology can aid us with the analysis of the pos-

sible conﬁgurations of the service and further predic-

tions about reachable QoE values.

When the quality parameters are discrete and the

maximum number of their values is known in ad-

vance, one can estimate the maximal number of con-

ﬁgurations of the machine as |C| = |S| ·

∏

|V |

i=1

|v

i

|,

where |S | is the number of states of the machine, |V | is

the number of context variables in analysis, and |v

i

| is

the number of possible values of the context variable

v

i

. This integer |C| represents the theoretical max-

imum number of conﬁgurations since a single user

might experience all the range of values at every sin-

gle state of the machine. However, the usage of the

EFSM formalism and the introduction of predicates

in the model allows to reduce the amount of possible

conﬁgurations reachable from the initial state.

An Implementation of a QoE Evaluation Technique Including Business Model Parameters

143

In order to show this effect, we introduce a third

algorithm (Algorithm 3) that allows us to compute the

possible conﬁgurations contained within the model.

It is based on the Algorithm 1, recursively comput-

ing the result. The process determines at each step

which are the transitions that will be executed given

the values of the parameters, and recomputes the val-

ues of the variables before calling recursively the al-

gorithm on the next state of the machine. Once the

required length was reached, the algorithm adds the

current conﬁguration to the ﬁnal set.

Using the implementation of Algorithm 3, we per-

formed experiments to empirically calculate: (1) the

maximum number of possible conﬁgurations repre-

sented with the machine, (2) how this value changes

with respect to the depth of the l-equivalent, and (3)

which is the distribution of the QoE values for the

conﬁgurations represented in the model. The follow-

ing sections present the details about the correspond-

ing numerical analysis executed.

4.2 Experimental Conﬁguration

The methodology and algorithms presented above

were used to analyze the beIN Sports Connect Ser-

vice. The three algorithms were implemented as an

extension of the MMT, and the model of the beIN

Sports Connect service was represented as an input

of the tool. In order to show the effectiveness of the

proposed approach, we used the Algorithm 3 to ana-

lyze the increase of the total number of conﬁgurations

with respect to the length of l-equivalent, ranging the

values of length of the tree from 2 to 18.

For this simulation, the initial values of the pa-

rameters were set to represent that the stream is not

playing (stream ﬂag = 0), and a end-user with a neu-

tral opinion of the service (service conﬁdence = 1)

who does not have a preference for a brand, called or

brand switcher (price premium = 1). However, the

conditions here exposed do not emulate a service fail-

ure, i.e. the transition labeled with 7 in Figure 1 is

never triggered since its execution is constrained to

an external change on the stream ﬂag variable.

In order to ﬁx this, we replaced the predicate of

the transition with two different ones simulating the

following scenarios: (1) the transition will trigger au-

tomatically if the simulation stays on the ‘delivery’

state more than twice consecutively (“ﬁxed failure”),

and (2) the transition will be triggered randomly with

probability 0.5 (“random failure”).

Finally, in order to compute the QoE of each con-

ﬁguration of the machine, we use the QoE model pro-

posed in (Sandoval et al., 2013), which is based on

a linear combination of the parameters. For the pur-

Table 2: Conﬁgurations and QoE distribution for scenario

1.

Length

Avg. number Average QoE distribution

of conf. ]0,1[ [1,2[ [2,3[ [3,4[ [4,5]

2 3 0 0 2 1 0

3 7 0 0 4 3 0

4 12 0 0 6 6 0

5 17 0 0 7 10 0

6 24 0 0 8 14 2

7 41 4 9 9 15 4

8 48 5 12 10 15 6

9 51 6 12 12 15 6

10 53 6 12 14 15 6

11 54 6 12 15 15 6

12 54 6 12 15 15 6

Table 3: Conﬁgurations and QoE distribution for scenario

2.

Length

Avg. number Average QoE distribution

of conf. ]0,1[ [1,2[ [2,3[ [3,4[ [4,5]

2 3 0 0 2 1 0

3 7 0 0 4 3 0

4 12.3 0.3 0 6 6 0

5 18.3 0.9 0.4 7 10 0

6 28.1 1.7 2.4 8 14 2

7 36.9 2.7 6.2 9 15 4

8 42.9 3.9 8.1 9.9 15 6

9 48.1 5.4 10.9 10.8 15 6

10 52.5 5.8 12 13.7 15 6

11 53.7 6 12 14.7 15 6

12 54 6 12 15 15 6

poses of this work, we use ﬁxed weights for each vari-

able shown in Table 1, leaving the experimental eval-

uation for ﬁnding their best values as future work.

4.3 Results and Discussion

The results of the experiments are presented in Ta-

bles 2 and 3. The results for lengths between 13 are

18 are not shown, showing the same values as length

12 in both cases.

In both scenarios, we observe a steady increase on

the number of conﬁgurations as long as the depth of

the analysis increases. However, this increase stabi-

lizes at the length of 11 and 12 for the ﬁxed and ran-

dom failure emulation respectively, where the maxi-

mal number of conﬁgurations reaches 54. After these

values, the distribution of the QoE values remains the

same in both experiments.

When observing the calculated QoE values, we

notice that at low lengths no conﬁgurations have a

QoE lower than 2, which can be explained by the

fact that the prediction has not considered the effect

of service failures yet. At the same time, it is possi-

ble to observe how the subjective and business-related

variables affect the QoE: with low depths of analysis

we can observe that the QoE of the computed conﬁg-

ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends

144

urations are spread in the 2 to 4 range (from “bad”

to “good” in the MOS scale). By introducing the

price premium and service conﬁdence variables (and

their respective updating functions) the model now

considers the effects of loyalty and past experiences

on his/her expectations.

As expected, when performing a deeper analy-

sis, we observe conﬁgurations with low QoE values

showing the effects of emulating failures of the ser-

vice. Despite the growth of the length, the conﬁgura-

tions with low QoE do not grow considerably in both

scenarios, showing a normal distribution with average

of 3, once the number of conﬁgurations has reached

its maximum.

This last fact represents a potential of this type of

analysis: it is possible to observe that there are 54

reachable conﬁgurations of the user, where 21 have

a QoE value equal or higher than 3. The rest of the

conﬁgurations have a “bad” or lower QoE value. This

information is useful to the service provider in order

to take countermeasures with these unsatisﬁed users

or invest more to improve the service offered.

Finally, it is important to notice that despite the

nature of the simulation of the failure events (random

or ﬁxed), it is possible to reach the same number of

maximal conﬁgurations: 54 in both cases, number

affected by the number of the context variables and

predicates inserted in the model. In addition, this ef-

fect of the analysis shows the advantage of the ap-

proach: the EFSM retains and limits the maximal

amount of conﬁgurations, that can be reached at a

ﬁxed length. In this sense, this fact allows us to

limit the length of the l-equivalent up to this value,

on which all the possible scenarios of a ﬁnal user are

reached. We conjecture that this is the optimal length

of the l-equivalent machine, where a deeper analysis

no longer adds different scenarios to consider.

5 CONCLUSION

In this paper, we presented the implementation of a

business-aware QoE evaluation framework. The im-

plementation is based on the Montimage Monitoring

Tool and it is composed of three basic algorithms.

We used this implementation to analyze the beIN

Sports Connect Service in order to show the advan-

tages of representing an OTT service using the EFSM

formalism. In this direction, we computed the amount

of different scenarios of an end-user. The implemen-

tation allows to simulate how this number varies with

the depth – length of the l-equivalent – of the analy-

sis, and which is the distribution of the QoE values at

different depths of the study.

With this analysis, we found that the number of

conﬁgurations will reach its maximum after a ﬁxed

value for the length. This result allowed us to limit the

depth of the analysis to this value of length. It is im-

portant when using the ﬁrst two algorithms to predict

future scenarios in an online basis, since it permits to

limit the amount of computation needed to calculate

all the possible future scenarios. In addition, the anal-

ysis of the distribution of the QoE values allowed us

to characterize in advance the amount of users that

might be classiﬁed as “unsatisﬁed”. This information

can be crucial for the service provider in order to im-

prove the service offered or compensate the users who

experiment low QoE of the service.

REFERENCES

Gozdecki, J., Jajszczyk, A., and Stankiewicz, R. (2003).

Quality of service terminology in ip networks. Com-

munications Magazine, IEEE, 41(3):153–159.

Green, W., Lancaster, B., and Sladek, J. (2007). Over-the-

top services. Pipeline, 4(7).

ISO (2015). Iso 9000:2015 quality management sys-

tems — fundamentals and vocabulary. Standard ISO

9000:2015.

Le Callet, P., M

¨

oller, S., and Perkis, A. (2013). Qualinet

whitepaper on deﬁnitions of quality of experience.

Seminar report, European Network on Quality of Ex-

perience in Multimedia Systems and Services.

Liao, C.-H., Tsou, C.-W., and Yi-Chung, S. (2015). The

roles of perceived enjoyment and price perception

in determining acceptance of multimedia-on-demand.

International Journal of Business and Information,

3(1).

Petrenko, A., Boroday, S., and Groz, R. (2004). Conﬁrming

conﬁgurations in efsm testing. IEEE Transactions on

Software Engineering, 30(1):29–42.

Rivera, D. and Cavalli, A. (2016). Qoe-driven service op-

timization aware of the business model. In 2016 30th

International Conference on Advanced Information

Networking and Applications Workshops (WAINA),

pages 725–730.

Rivera, D., Kushik, N., Fuenzalida, C., Cavalli, A., and Yev-

tushenko, N. (2015). Qoe evaluation based on qos

and qobiz parameters applied to an ott service. In

Web Services (ICWS), 2015 IEEE International Con-

ference on, pages 607–614.

Sandoval, J., Ehijo, A., Casals, A., and Estevez, C. (2013).

New model and open tools for real testing of qoe in

mobile broadband services and the transport protocol

impact: The operator’s approach. In IEEE LATIN-

COM, pages 1–5.

Van Moorsel, A. (2001). Metrics for the internet age: Qual-

ity of experience and quality of business. In Fifth In-

ternational Workshop on Performability Modeling of

Computer and Communication Systems, pages 26–31.

Universit

¨

at Erlangen-N

¨

urnberg, Germany.

An Implementation of a QoE Evaluation Technique Including Business Model Parameters

145