Defining Controlled Experiments Inside the Access Control Environment
Said Daoudagh
1,2 a
and Eda Marchetti
1 b
1
ISTI-CNR, Pisa, Italy
2
Department of Computer Science, University of Pisa, Pisa, Italy
Keywords:
Access Control, Controlled Experiment, Goal-Question-Metric, Testing, XACML.
Abstract:
In ICT systems and modern applications access control systems are important mechanisms for managing
resources and data access. Their criticality requires high security levels and consequently, the application of
effective and efficient testing approaches. In this paper we propose standardized guidelines for correctly and
systematically performing the testing process in order to avoid errors and improve the effectiveness of the
validation. We focus in particular on Controlled Experiments, and we provide here a characterization of the
first three steps of the experiment process (i.e., Scoping, Planning and Operation) by the adoption of the Goal-
Question-Metric template. The specialization of the three phases is provided through a concrete example.
1 INTRODUCTION
Nowadays, quality of Information and Communica-
tion Technology (ICT) systems and modern appli-
cations is strictly tied with the security and privacy.
Among security mechanisms, a critical role is played
by Access Control (AC) systems, which aim to ensure
that only the intended subjects can access the pro-
tected data and get the permission levels required to
accomplish their tasks and no much more.
Due to the complexity of AC systems, for ensur-
ing the required security level, a key factor becomes
the application of effective and efficient testing ap-
proaches: knowing in advance the criticality of the
systems lets to put in practice efficacious corrective
actions so as to improve the overall security of the sys-
tem. However, testing phase is a time consuming, er-
ror prone and critical step of the development process,
which involves different activities: from test strategy
selection, to the test case derivation, from execution to
the final test results evaluation. Bad choices in each
stage of the testing phase may compromise the entire
process, with the risk of releasing inadequate secu-
rity solutions that allow unauthorized access from the
security perspective or unlawful processing from the
legal perspective.
In the last years different proposal are tar-
geting the efficacious management of the testing
a
https://orcid.org/0000-0002-3073-6217
b
https://orcid.org/0000-0003-4223-8036
phase by proposing techniques that combines Model
Based Testing (MBT) (Utting et al., 2012) and Test
driven Development (TDD) (Nanthaamornphong and
Carver, 2017) techniques so that model-based tests
can guide the development. Among them the Model-
Based Test Driven Development (MBTDD) (Sadeghi
and Mirian-Hosseinabadi, 2012) one of the first ten-
tative for extending the TDD cycle is extended with
MBT steps. However, the main issues of the MBTDD
is that it does not deal with the reuse of test cases
along the iterations. For this different improved so-
lutions have been conceived in order to better sup-
port the test phase development, management and
evaluation (Harumi et al., 2016). In line with these
proposals, this paper provides a revised approach of
MBTDD in the context of testing access control sys-
tems. In particular, the paper focuses on the use of
controlled experiments for ensuring the integrity and
replicability of the testing results.
In literature, different solutions are currently
available for testing AC systems and their behav-
ior (Bertolino et al., 2013; Bertolino et al., 2014a;
Hu et al., 2017), but there are not standardized guide-
lines for correctly and systematically performing the
testing process in order to avoid errors and improve
the effectiveness of the validation. In particular, the
lack of a formalized specification of the testing ac-
tivity can have the following consequences: impossi-
bility of replicating and controlling the process espe-
cially in case of regression testing (Yoo and Harman,
2012), difficulties in the generalization of the testing
Daoudagh, S. and Marchetti, E.
Defining Controlled Experiments Inside the Access Control Environment.
DOI: 10.5220/0009358201670176
In Proceedings of the 8th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2020), pages 167-176
ISBN: 978-989-758-400-8; ISSN: 2184-4348
Copyright
c
2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
167
results and consequent derivation of statistical signif-
icance values; and problems in defining and sharing
a common testing knowledge so as to avoid recurring
failures and speeding up the corrective process.
A reply to these issues comes from the software
engineering context, where Controlled Experiments
(CEs) (Juzgado and Moreno, 2001; Wohlin et al.,
2012; Basili and Rombach, 1988) are commonly used
to investigate the cause-effect relationships of intro-
ducing new methods, techniques or tools and to build
a body of knowledge supported by observation and
empirical evidence. Therefore, the controlled experi-
ments let to validate the different activities of the test-
ing process by means of the identification of impor-
tant variables, the definition of specific testing mod-
els and objectives, and the derivation of empirical ev-
idence. In the controlled experiment different treat-
ments can be applied to, or by, different subjects,
while other variables are kept constant and the effects
on response variables are measured.
Authors in (Juzgado and Moreno, 2001; Wohlin
et al., 2012) categorize experiments as either
technology-oriented or human-oriented, depending
on whether artifacts or human subjects have given
various treatments. In this paper, we revise and cus-
tomize the technology-oriented experiments in order
to provide general guidelines for correctly end ef-
fectively performing the testing of the AC systems.
Therefore, we provide the characterization of the first
three (over the five) steps of the Experiment Process
that namely are Scoping, Planning, and Operation.
We refer to (Daoudagh et al., 2020) for a concrete ap-
plication example as well as a detailed checklist of the
required implementation steps.
Outline. Section 2 introduces the main concepts
used along the rest of the paper related to Controlled
Experiment, the Goal-Question-Metric, and Access
Control, as well as the related work; Section 3 illus-
trates our proposal of a family of Controlled Experi-
ments in the context of Access Control. In particular,
in Sections 4, 5 and 6 we detail the first three phases
of the Controlled Experiment; finally, Section 7 con-
cludes the paper and depicts the future work.
2 BACKGROUND AND RELATED
WORK
In this section we firstly describe the main concepts
related with (1) Controlled Experiment; (2) Access
Control & Testing; and (3) Goal-Question-Metric
(GQM) used along the rest of the paper and their re-
lated works.
Controlled Experiment. Experiments (or CEs) are
used in software engineering to investigate the cause-
effect relationships. They consist of a well-defined
Experiment Process including five specific phases: (i)
Scoping, (ii) Planning, (iii) Operation, (iv) Analysis
and Interpretation, and (v) Presentation and Package.
However, in the Experiment Process it is not manda-
tory to finish an activity before starting the next one.
As a consequence, it is possible to go back and refine
a previous activities before continuing with next one.
In this sense it is partially iterative.
The purpose of a CE is therefore to systematically
define the elements necessary for ensure the integrity
and replicability of the obtained results. Very briefly,
the main element are: (1) objects on which the experi-
ment is run are the experimental units and can involve
the all the systems or part of it; (2) subjects that repre-
sent artifacts on which the methods or techniques are
applied; (3) the outcome of an experiment is referred
to a quantitative response variable (also called De-
pendent Variable); (4) each considered characteristic
target of the experiment to be studied that can affect
the response variable is called a factor (also called In-
dependent Variables); (5) the possible values of the
factors are called levels; and (6) parameter, i.e., any
other invariable (qualitative or quantitative) character-
istic of the software project that does not influence the
result of the experiment.
Consequently, in each experiment a combination
of alternatives of factors are applied by a subject on an
unit. A defined and precise specification of the exper-
iment guarantees both: the External replication (Judd
et al., 1991), i.e., reproducing the experiment in dif-
ferent contexts and environments so as to increase the
confidence in experiment results; the Internal replica-
tion, i.e., the repetition of the experiment more time
in the same environment or condition to increases the
reliability of the experiment results.
In Software Engineering field, CEs are gaining
a lot of attention (Sjøberg et al., 2005; Ko et al.,
2015) and different proposals are trying to give guid-
ance on how to conduct CEs (Juzgado and Moreno,
2001; Wohlin et al., 2012). Following this tendency,
our proposal want to come up with a Goal Definition
Framework that enables one to conduct technology-
oriented experiments in the Access Control (AC) con-
text. More precisely, the novelty of our proposal is
to provide general guidelines for correctly end effec-
tively performing the testing of AC systems.
Access Control & Testing. AC systems are means
to help organizations to improve their security from
the point of view Confidentiality, Integrity and Avail-
ability (i.e., the CIA Triad). Often AC systems are
MODELSWARD 2020 - 8th International Conference on Model-Driven Engineering and Software Development
168
regulated by Access Control Policies (ACPs) that
define which subject is allowed to access a pro-
tected resources. ACPs are usually written by us-
ing the eXtensible Access Control Markup Language
(XACML) (OASIS, 2013) standard. This standard
defines both a reference architecture and a language
based on XML to express ACPs and AC request/re-
sponse.
One of the main the main components of XACML
standard is Policy Decision Point (PDP), which eval-
uates the ACP against the request and returns the
response, including the authorization decision. For
more details about the XACML standard we remaind
the reader to its specification (OASIS, 2013).
In literature, several works are focused on AC sys-
tems testing, and they can be mainly divided into
the following research fields: i) test strategies defi-
nition (Bertolino et al., 2013; Bertolino et al., 2018);
ii) test strategy assessment (Bertolino et al., 2014b;
Lonetti and Marchetti, 2018; Daoudagh et al., 2019a);
iii) test cases generation and execution (Bertolino
et al., 2010; Hu et al., 2017); iv) test execution and
oracle derivation which are focused on approaches for
evaluating theAC replies to specific inputs (Daoudagh
et al., 2015; Calabr
`
o et al., 2017; Bertolino et al.,
2018; Daoudagh et al., 2019b).
The lack of formality of the conducted studies in
the above work do not enable external replication of
the result. Differently our work want to contribute to
formally and thoroughly conduct CEs in the context
of AC.
Goal-Question-Metric. Originally presented
in (Basili and Rombach, 1988), the Goal-Question-
Metric (GQM) paradigm proposes a top-down
approach to define measurement: goals lead to
questions, which are then answered by metrics. A
GQM model is a hierarchical structure as presented
in Figure 1 starting with a goal by specifying purpose
of measurement, object to be measured, issue to be
measured, and viewpoint from which the measure
is taken (Conceptual level). The goal is refined into
several questions that usually break down the issue
into its major components (Operational level). Each
question is then refined into metrics, some of them
objective and others subjective (Quantitative level).
The same metric can be used to answer different
questions under the same goal as well as different
goals (Basili et al., 1994).
In security domain there are a few proposals using
the GQM and they are used to mainly identify secu-
rity requirements and metrics. For example, authors
in (Islam and Falcarin, 2011) used GQM approach
to define clear and comprehensible measures for a
Figure 1: The Goal Question Metric (GQM) model
(adopted from (Basili et al., 1994)).
set of established security requirements. The GQM
approach based on Standard security metrics and on
Service Oriented Architecture maturity is presented
in (Kassou and Kjiri, 2013), where scholars aimed at
supporting organizations to assess SOA Security as
well as to ensure the safety of their SOA based col-
laborations. To assessing the security of data stored in
cloud storage, authors in (Yahya et al., 2015) attempt
to provide practical guidance and example of mea-
surements using GQM. A more recent work is pre-
sented in (Weldehawaryat and Katt, 2018) where the
authors presented a quantitative evaluation approach
for defining security assurance metrics using two per-
spectives, vulnerabilities and security requirements.
Differently from the above works, our proposal
aims at enabling the derivation of metrics for answer-
ing questions related to investigation goals in the con-
text of AC. In particular, the intention is to enable CEs
in the context of AC by covering all the phases of the
process. In this paper however we focus on the first
three phases of the process and we refer to (Daoudagh
et al., 2020) for more details about the remaining
phases.
3 A GQM PROPOSAL FOR
ACCESS CONTROL TESTING
The general idea behind our proposal is to provide a
set of CE families useful for formally and thoroughly
describing scientific investigations in the context of
Access Control (AC) systems. Indeed, our intuition
is to use the standard and consolidated GQM tem-
plate (Basili and Rombach, 1988), as guidance to se-
lect, and consequently classify, concepts of interest in
the domain of AC. Then, exploiting the knowledge
and the techniques typical of the software testing sci-
entific environment, a concrete AC-based goal defini-
tion framework can be derived. This set will be well-
defined, specific and achievable AC testing goals to
be exploited for different experimentations.
The proposal of the paper, although grounded in a
domain-related AC testing, represents an example of
realization of CE families, that can be easily applied
in all the domains where a scientific investigation in
Defining Controlled Experiments Inside the Access Control Environment
169
which a formal and rigorous fashion should be per-
formed.
As in Figure 2, the proposal is composed of five
conceptual components: the Goal Question Metric
1 , the Access Control Context 2 and Software
Testing 3 , which represent the conceptual models of
the target experiment.
These models are integrated in the Goal Defini-
tion Framework component 4 so as to define a spe-
cialized GQM, which is the common basic knowledge
for the AC families. Then, the GQM exploited in the
Main Research Goal component 5 for defining sci-
entific testing goals in AC testing process and there-
fore for selecting specific and achievable AC testing
goals 6 to be evaluated in real context.
Figure 2: GQM Access Control Model.
In the following sections we illustrate how the use
of a specialized GQM can be a important innovation
for the development of Controlled Experiments in AC
context. In particular, by referring to the structure
of a CE presented in Section 2, we detail the execu-
tion of the first three steps of the process (i.e., Scop-
ing, Planning and Operation), which are those that
need to be specialized for the AC domain. We refer
to (Daoudagh et al., 2020) for a complete example in-
cluding also the last two phases.
4 EXPERIMENT SCOPING
The purpose of the scoping phase is to determine
the foundations of the experiment by defining goals
according to a specific framework. As described
in the previous section, The idea here is to use
the Goal-Question-Metric (GQM) method, integrated
with concepts of AC and Software Testing for deriv-
ing a specialized template for the definition of CEs
goals in the AC testing context. By referring to Fig-
ure 2, the scoping phase exploits the domain specific
concepts of components 1 , 2 and 3 so as to define
a reference framework, i.e., Goal Definition Frame-
work (component 4 ). In the remainder of the section
the use of the three components is better detailed.
According to (Basili and Rombach, 1988), the
GQM template consists of five elements: (1) object
of study is target entity of the experiment. It can be a
product, process, resource, model, metric or theory.
(2) purpose defines the intention of the experiment
is. It may be to evaluate the impact of two different
techniques or to characterize the learning curve of an
organization. (3) quality focus is the primary effect
under study in the experiment. It can y be effective-
ness, cost, reliability etc. (4) perspective describes the
viewpoint from which the experiment results are in-
terpreted. Examples are developer, project manager,
customer and researcher. (5) contextis the environ-
ment in which the experiment is run. It defines which
personnel is involved in the experiment (subjects) and
which software artifacts, called objects
1
are used in
the experiment.
Consequently, the intention of the GQM template
is to Analyze <Object(s) of study> for the purpose
of <Purpose> with respect to their <Quality focus>
from the point of view of the <Perspective> in the
context of <Context>.
Table 1: AC concepts.
GQM elements AC concepts
Object of study XACML-based PDPs
XACML-based ACPs
Purpose -
Quality focus -
Perspective ACP Architect
AC System Developer
AC System Administrator
Context Subjects (XACML Policies)
Objects (XACML-based PDPs)
AC Model. There are different access control
model in literature, among them, in this study we re-
fer to the Attribute-Based Access Control (ABAC)
model and in particular to its implementation, i.e.,
the XACML standard. More precisely, we refer to
both the ACP model and the XACML reference ar-
chitecture.The objective here is to characterize the
CE in the context of AC by gathering the main con-
cepts, terms and components that can be used to for-
mulate an interesting goals from the scientific point
of view. The selected elements are then used in the
GQM template for the object of the study, the pur-
pose, the perspective and the context. The classi-
1
Note that the objects here are generally different from
the objects of study
MODELSWARD 2020 - 8th International Conference on Model-Driven Engineering and Software Development
170
fication we propose in this paper is summarized in
table Table 1. In particular, the first column (GQM
elements) lists the GQM element while the second
one (column AC concepts), reports concepts useful
for defining meaningful research investigation in the
context of AC.
Software Testing. In literature different proposals
exist that leverage well-known software techniques
to test ACPs and AC mechanisms. By analyzing
current literature, we summarize in Table 2 in the
column Software Testing concepts some of the
main software testing concepts useful in generic con-
trolled experiment. We also classify them accord-
ing to the GQM template elements (column GQM
elements).
Table 2: Software Testing concepts.
GQM elements Software Testing concepts
Object of study Test case generation strategy
Test case prioritization technique
Mutation Generators
Test case reduction technique
Oracle Derivation
Purpose Characterize
Evaluate
Quality focus Effectiveness
Cost
Size
APFD
Performance
Perspective Researcher
Tester
Project manage
User
Context -
However, without the pretend to be exhaustive and
in the aim of simplicity, the table reports a simplifica-
tion of a possible classification. In particular, in this
paper we limit ourself to the definition and assessment
of a test case generation strategies, because they are
recognized as ones of the most crucial activities of
the testing process. In the assessment of the effective-
ness of a test strategy, concepts as coverage criteria
and mutation analyses or test oracle are often used,
and therefore included in Table 2. We also add the
prioritization and reduction concepts because they are
commonly adopted techniques for reducing the num-
ber of test case to be executed and consequently the
effort and time due to overall testing phase.
Goal Definition Framework. On the bases of the
concepts of Table 2 the specialized Goal Definition
Framework is derived. This is a comprehensive
framework based on the GQM for the definition of
research investigation goals for testing tools, method-
ologies and strategies in the AC (both ACPs and AC
mechanisms) context. To the best of the authors’
knowledge, this proposal is the first attempt to provide
a formally and thoroughly solution for the definition
of a Controlled Experiment in AC domain. Table 3
reports the conceived framework, which represent the
output of component 4 of our proposal depicted in
Figure 2.
Specifically Table 3 has a column for each of
the five GQM where the identified AC and Software
Testing concepts are reported: namely Object of
study, Purpose, Quality focus, Perspective,
and Context.
Research Goals in AC Context. Combining the el-
ements of the different columns of Table 3 a defined
and focused scientific investigation goals that enable
the specification of CE in the context of AC can be
identified. Thus, the Goal Definition Framework lets
the definition of families of goals for the access con-
trol systems testing. In Table 4 a not exhaustive list
of the mostly adopted research goals are reported.
In particular, the fist column (Research Goal) re-
ports a label associated to each defined goal, whereas,
the second column (Goal Definition) contains the
definition of the goal using the GQM template cus-
tomized with a specific combination of the elements
of Table 3. Note that not all the possible combina-
tions of those elements enable the definition of an in-
teresting a well-defined goal. It is up to the user of
the framework to choose the correct combination de-
pending on the concrete objective.
5 EXPERIMENT PLANNING
The Planning activity consists of different steps where
foundation of the experiment is defined. More pre-
cisely, the context of the experiment is determined
and the hypothesis is stated formally, including a null
hypothesis and an alternative hypothesis. Then, we
need to determine variables, both independent vari-
ables (inputs) and dependent variables (outputs), and
to identify the subjects of the study. After the de-
sign step, which includes choosing a suitable exper-
iment design, the instrumentation of the experiment
is defined by identifying and preparing suitable ob-
jects and measurement procedures. As a part of the
planning, it is important to consider the question of
Defining Controlled Experiments Inside the Access Control Environment
171
Table 3: Goal definition framework in the context of XACML Testing.
Object of study Purpose Quality focus Perspective Context
Test case generation strategy Characterize Effectiveness Researcher Subjects (XACML Policies)
Test case prioritization technique Evaluate Cost Tester Objects (XACML-based PDPs)
Mutation Generators Assess Size Project manager
Test case reduction technique APFD User
XACML-based PDPs Performance ACP Architect
XACML Policies AC System Developer
XACML-based Oracle Derivation AC System Administrator
Table 4: Main Research Goals in the context of XACML Systems Testing.
Research Goal Goal Definition
Goal 1: Policy Testing Analyze test case generation strategies for the purpose of evaluation with respect to their effectiveness and size of test suite
produced from the point of view of the researcher in the context of XACML policy testing.
Goal 2: PDP Testing Analyze test case generation strategies for the purpose of evaluation with respect to their effectiveness and size of test suite
produced from the point of view of the researcher in the context of XACML policy decision point testing.
Goal 3: Mutation PDP Analyze mutation generators for the purpose of evaluation with respect to their applicability from the point of view of the
researcher in the context of XACML policy decision point testing.
Goal 4: Mutation Policy Analyze mutation generators for the purpose of evaluation with respect to their effectiveness and size of test suite produced
from the point of view of the researcher in the context of XACML policy testing.
Goal 5: Prioritization Analyze test case prioritization techniques for the purpose of evaluation with respect to their effectiveness (rate of fault
detection, using APFD (Average Percentage Faults Detected) metric) from the point of view of the researcher in the context of
XACML policy testing.
Goal 6: Reduction Analyze test case reduction techniques for the purpose of evaluation with respect to their effectiveness (rate of fault detection,
using APFD (Average Percentage Faults Detected) metric) from the point of view of the researcher in the context of XACML
policy and PDP testing.
validity of the results we can expect. Validity can
be divided into four major classes: internal, external,
construct and conclusion validity.
In the remainder of the section, in order to clar-
ify the steps of the Planning phase we refer to a real
example: the testing of a PDP engine. This can be
translated into the selection of the best test strategy
for testing the PDP in order to improve its quality and
reduce the testing effort. As reported in Table 5, in
this case three sub-goals, each focuses on a specific
research question, are identified:
RQ1 Effectiveness: How much does the qual-
ity of a test suite produced by Strategy
1
(T GS
1
)
differ from the quality of test suite produced by
Strategy
2
(T GS
2
) in terms of Effectiveness, i.e.,
the mutation score?
RQ2 Size: How much does the cost of a test suite
produced by Strategy
1
differ from the cost of test
suite produced by Strategy
2
in terms of Size, i.e.,
the number of test cases?
RQ3 APFD: How much does the Average Per-
centage Faults Detected (APFD) of a test suite
produced by Strategy
1
differ from the APFD of
test suite produced by Strategy
2
?
Context Selection. The first activity of the planning
phase is the Context Selection. According to (Wohlin
et al., 2012) the experiment contexts can be classified
as in Table 6. Considering the PDP testing example,
because two test strategies should be compared, the
context is a Multi-test within object study.
Considering the implementation of the considered
experiment, the Policy Decision Point is the Sun-
PDP (Sun Microsystems, 2006) is the target PDP,
(One Object). We decided for Sun’s PDP engine
because it is currently one of the most mature and
widespread used engine for XACML policy imple-
mentation, which provides complete support for all
the mandatory features of XACML 2.0 as well as a
number of optional features. The strategies to be com-
pared are the Multiple test strategy (Bertolino et al.,
2013) and XACMET test strategy (Daoudagh et al.,
2019b); a set of real world XACML policies are used
for test case derivation (Multiple Subjects), and mu-
tation techniques adopted to assess the test strate-
gies considered; the comparison is done by evaluat-
ing the effectiveness, the size and the APFD of the
test suite generated for each XACML policy, apply-
ing both strategies.
Hypothesis Formulation. We consider the follow-
ing null hypotheses:
MODELSWARD 2020 - 8th International Conference on Model-Driven Engineering and Software Development
172
Table 5: Sun PDP Testing Goal.
Policy Decision Point Testing Goal (Goal 2)
Analyze Multiple and XACMET Strategies for the purpose of evaluation with respect to their effectiveness and size of test suite produced from the point of
view of the researcher in the context of Sun PDP testing.
Research Questions
RQ 1: Effectiveness RQ 2: Size RQ 3: APFD
Research Subgoals
Analyze Multiple and XACMET Strategies for
the purpose of evaluation with respect to their
test suite effectiveness from the point of view of
the researcher in the context of Sun PDP testing
without constraints.
Analyze Multiple and XACMET Strategies for
the purpose of evaluation with respect to their
cost in terms of number of test cases generated
from the point of view of the researcher in the
context of budget programming.
Analyze Multiple and XACMET Strategies for
the purpose of evaluation with respect to their
effectiveness in terms of APFD from the point of
view of the researcher and quality manager in the
context of interruption of Sun PDP testing activ-
ity.
Metrics
m1: Effectiveness m1: Size of the test suite m1: APFD
Table 6: Experiment context classification.
# Objects
One More than one
# Subjects per
object
One Single object
study
Multi-object
variation study
More than one Multi-test
within object
study
Blocked
subject-object
study
H
0E f f
: µ
E f f St1
= µ
E f f St2
the Strategy1 finds on
average the same number of faults, i.e., the effec-
tiveness, as the Strategy2, where µ denotes the av-
erage percentage of the killed mutants using the
complete test suites generated by the two strate-
gies;
H
0Size
: µ
NSizeSt1
= µ
NSizeSt2
the size of test suite is
equal for strategy1 and strategy2;
H
0APFD
: µ
APFDSt1
= µ
APFDSt2
the average APFD
is equal for strategy1 and strategy2.
A null hypothesis states that there are no real un-
derlying trends or patterns in the experiment setting;
the only reasons for differences in the observations
are coincidental. This is the hypothesis that we wants
to reject with a high significance as possible.
When the null hypothesis can be rejected with rel-
atively high confidence, it is possible to formulate an
alternative hypothesis, as following:
H
1E f f
: µ
E f f St1
6= µ
E f f St2
the Strategy1 and Strat-
egy2 find on average a different number of faults,
i.e. their effectiveness are Not equal;
H
1Size
: µ
SizeSt1
6= µ
SizeSt2
the size of test suite is
Not equal for strategy1 and strategy2;
H
1APFD
: µ
APFDSt1
6= µ
APFDSt2
the average APFD
is Not equal for strategy1 and strategy2.
Variables Selection. By referring to the PDP test-
ing example, the unique independent variable is the
test case generation strategy with two levels or al-
ternatives (treatments) for the main factor: {Multiple
and XACMET}. The dependent variables are the Ef-
fectiveness, the Size of the test suites and the APFD
metrics.
The object of the experiment a complex object
composed by SunPDP (Gold PDP) and some mutated
versions of it. Because mutation techniques are con-
sidered in the experiment, the mutation generator can
be identified as a possible Parameter.
Selection of Subjects. The selection of subjects is
important when conducting an experiment, because
closely connected to the generalization of the results
from the experiment. In order to generalize the results
to the desired population, the selection must be repre-
sentative for that population, thus, it is also called a
sample from a population. In the example considered
the XACML Policies are the Subjects.
Experiment Design. The design we use is the
paired comparison design , a particular kind of one
factor with two treatments (Wohlin et al., 2012). The
same design is called “randomized paired compar-
ison design: two alternatives on one experimental
unit” in (Juzgado and Moreno, 2001). In this design,
each subject uses both treatments on the same object,
i.e., both Strategies are applied to each XACML pol-
icy and the obtained test suites are evaluated using
the SunPDP and its mutants. This design allows to
Defining Controlled Experiments Inside the Access Control Environment
173
compare the two treatments (Multiple and XACMET
strategies) against each other; the most common oper-
ation is to compare the means of the dependent vari-
able (Effectiveness, Size and APFD) for each treat-
ment. In particular, both Strategies are applied to each
XACML policy.
Instrumentation. The overall goal of the instru-
mentation is to provide means for performing the ex-
periment and to monitor it, without affecting the con-
trol of the experiment. If the instrumentation affects
the outcome of the experiment, the results are invalid.
In the planning of an experiment, the instruments
are chosen. Before the execution, the instruments are
developed for the specific experiment. The instru-
ments for an experiment are of three types, namely
objects, guidelines and measurement instruments.
In the example considered the Object is the Sun
PDP.
Usually in a testing experiment the number and
the nature of faults in the testing objects is known
in advance. In the experiment considered the mu-
tation technique can be applied to the Object for
deriving a controlled number of faulty versions of
the Sun PDP. In this case for instantiating the mu-
tation generator parameter, different levels can be
considered, such as µJava) (seung Ma et al., 2005),
Javalanche (Schuler and Zeller, 2009), Major (Just,
2014) or Judy (Madeyski and Radyk, 2010), that may
influence the result of the experiment.
Guidelines and Measurement. Guidelines are
procedural steps for executing the Controlled Exper-
iment. They include process descriptions, checklists,
tools and facilities useful for performing measure-
ment and enabling the result analysis and interpreta-
tion. Among the available proposals for automating
the overall testing process of AC, in this paper we
use the solution provided in (Daoudagh et al., 2019a),
so as to be compliant with the selected research goal
(Goal 2 of Table 5). The selected framework enables
the collection of the target measures, i.e., Effective-
ness, Size and APFD as reported in Table 5.
6 EXPERIMENT OPERATION
The third activity of the experimental process is Oper-
ation, which consists of three steps: preparation, exe-
cution and data validation. Figure 3 reports the activ-
ity diagram of the experiment operation phase consid-
ering the Sun PDP Testing of the Goal 2 of Table 5.
As in the Figure 3, during the preparation step,
subjects, the object and parameters are instantiated on
Figure 3: Experiment Operation Activities.
the selected Testing Framework. In particular during
the activities A , B and C the following steps are
performed: the Subject represented by XACML poli-
cies, are selected; the treatments, i.e., the test case
generation strategies (Multiple and XACMET in the
example)and the Object of the experiment, i.e., the
Sun PDP, are defined. Then, the test cases (i.e., the
XACML requests) and the required mutants (i.e., the
mutated version of SunPDP) are derived during the
activities
D , E and F . Afterwards, the execu-
tion step, which consists of XACML requests evalua-
tion, data collection, and measures computation, is in
charge of activities G and H . Finally, data valida-
tion consists of data selection, filtering and measures
computing, i.e., the calculation of Size, Effectiveness
and APFD metrics that are in charge of activities I
of Figure 3.
Once the experimental data will be collected the
Null Hypothesis test could be performed. In the
MODELSWARD 2020 - 8th International Conference on Model-Driven Engineering and Software Development
174
considered experiment this can be translated into the
H
1E f f
, H
1Size
and H
1APFD
defined in Section 5. Ac-
cording to (Juzgado and Moreno, 2001; Wohlin et al.,
2012), we can apply the Paired T-Test to formally ver-
ify the Null Hypothesis with the confidence level of
95%. This choice was a natural consequence of the
type of design adopted, i.e., the paired comparison.
Therefore, following the standard best practices, we
can accept a probability of 5% of committing a Type-
1-Error (Juzgado and Moreno, 2001; Wohlin et al.,
2012), i.e., the Null Hypothesis is rejected if the com-
puted p-value is less or equal to 0,05 (alpha = 0.05).
7 CONCLUSIONS
In this paper we presented a family of controlled ex-
periments in the context of AC testing. The idea was
to define a set of standardized guidelines for correctly
and systematically performing the testing process in
order to avoid errors and improve the effectiveness of
the validation. The proposal relies on a characteriza-
tion of the first three steps of the experiment process
(i.e., Scoping, Planning and Operation) by leverag-
ing the Goal-Question-Metric template. Thus, we de-
tailed the activities necessary for performing the first
three steps of the experiment process (i.e., Scoping,
Planning and Operation). The example of the testing
of the Sun PDP engine is taken as a reference for bet-
ter explaining the three phases.
It was out of the scope of the paper providing the
complete list of testing goals or the realization of all
the possible testing frameworks. The example pro-
vided in the paper wanted to highlight the peculiar-
ity of the Controlled Experiments and the potentiality
they represent for the testing activity.
As a future work we intent to provide other im-
plementations of the Controlled Experiments for dif-
ferent testing purposes, so as to demonstrate its flex-
ibility and adaptability. We want also to apply the
proposed Controlled Experiment in real environments
so as to collect testing results and perform statistical
analysis.
ACKNOWLEDGEMENTS
This work is partially supported by CyberSec4Europe
Grant agreement ID: 830929.
REFERENCES
Basili, V. R., Caldiera, G., and Rombach, H. D. (1994). The
goal question metric approach. In Encyclopedia of
Software Engineering. Wiley.
Basili, V. R. and Rombach, H. D. (1988). The tame
project: towards improvement-oriented software envi-
ronments. IEEE Transactions on Software Engineer-
ing, 14(6):758–773.
Bertolino, A., Daoudagh, S., Lonetti, F., and Marchetti, E.
(2018). An automated model-based test oracle for ac-
cess control systems. In Proceedings of the 13th In-
ternational Workshop on Automation of Software Test,
AST ’18, pages 2–8, New York, NY, USA. ACM.
Bertolino, A., Daoudagh, S., Lonetti, F., Marchetti, E., Mar-
tinelli, F., and Mori, P. (2014a). Testing of polpa-
based usage control systems. Software Quality Jour-
nal, 22(2):241–271.
Bertolino, A., Daoudagh, S., Lonetti, F., Marchetti, E., and
Schilders, L. (2013). Automated testing of extensible
access control markup language-based access control
systems. IET Software, 7(4):203–212.
Bertolino, A., Le Traon, Y., Lonetti, F., Marchetti, E., and
Mouelhi, T. (2014b). Coverage-based test cases se-
lection for xacml policies. In Proceedings of ICST
Workshops, pages 12–21.
Bertolino, A., Lonetti, F., and Marchetti, E. (2010). Sys-
tematic XACML Request Generation for Testing Pur-
poses. In Proc. of 36th EUROMICRO Conference
on Software Engineering and Advanced Applications
(SEAA), pages 3 –11.
Calabr
`
o, A., Lonetti, F., and Marchetti, E. (2017). Access
control policy coverage assessment through monitor-
ing. In Proc. of TELERISE, pages 373–383.
Daoudagh, S., El Kateb, D., Lonetti, F., Marchetti, E., and
Mouelhi, T. (2015). A toolchain for model-based de-
sign and testing of access control systems. In Proc.of
MODELSWARD, pages 411–418. IEEE.
Daoudagh, S., Lonetti, F., and Marchetti, E. (2019a). A
framework for the validation of access control sys-
tems. In Saracino, A. and Mori, P., editors, Proceed-
ings of the 2nd International Workshop on Emerging
Technologies for Authorization and Authentication.
Daoudagh, S., Lonetti, F., and Marchetti, E. (2019b).
XACMET: XACML modeling & testing an auto-
mated model-based testing solution for access control
systems. Software Quality Journal.
Daoudagh, S., Lonetti, F., and Marchetti, E. (2020). As-
sessing testing strategies for access control systems:
A controlled experiment. In Proceedings of ICISSP
2020, Valletta, Malta, February 25-27, 2020.
Harumi, T. et al. (2016). D-mbtdd: An approach for reusing
test artefacts in evolving systems. In 2016 46th Annual
IEEE/IFIP International Conference on Dependable
Systems and Networks Workshops (DSN-W). IEEE.
Hu, V. C., Kuhn, R., and Yaga, D. (2017). Verification and
test methods for access control policies/models. NIST
Special Publication, 800:192.
Islam, S. and Falcarin, P. (2011). Measuring security
requirements for software security. In 2011 IEEE
Defining Controlled Experiments Inside the Access Control Environment
175
10th International Conference on Cybernetic Intelli-
gent Systems (CIS), pages 70–75.
Judd, C. M., Smith, E. R., and Kidder, L. H. (1991). Re-
search methods in social relations, fort worth: Holt,
rinehart and winston.
Just, R. (2014). The major mutation framework: Efficient
and scalable mutation analysis for java. In Proceed-
ings of the 2014 international symposium on software
testing and analysis, pages 433–436. ACM.
Juzgado, N. J. and Moreno, A. M. (2001). Basics of soft-
ware engineering experimentation. Kluwer.
Kassou, M. and Kjiri, L. (2013). A goal question metric
approach for evaluating security in a service oriented
architecture context. CoRR, abs/1304.0589.
Ko, A. J., Latoza, T. D., and Burnett, M. M. (2015). A
practical guide to controlled experiments of software
engineering tools with human participants. Empirical
Softw. Engg., 20(1):110–141.
Lonetti, F. and Marchetti, E. (2018). On-line tracing of
xacml-based policy coverage criteria. IET Software,
12(6):480–488.
Madeyski, L. and Radyk, N. (2010). Judy - a mutation test-
ing tool for java. IET Software, 4(1):32–42.
Nanthaamornphong, A. and Carver, J. C. (2017). Test-
driven development in scientific software: a survey.
Software Quality Journal, 25(2):343–372.
OASIS (2013). eXtensible Access Control
Markup Language (XACML) Version 3.0.
http://docs.oasis-open.org/xacml/3.0/
xacml-3.0-core-spec-os-en.html.
Sadeghi, A. and Mirian-Hosseinabadi, S.-H. (2012).
Mbtdd: Model based test driven development. Inter-
national Journal of Software Engineering and Knowl-
edge Engineering, 22(08):1085–1102.
Schuler, D. and Zeller, A. (2009). Javalanche: Efficient mu-
tation testing for java. In Proceedings of ESEC/FSE,
pages 297–298, New York, NY, USA. ACM.
seung Ma, Y., Offutt, J., and Kwon, Y. R. (2005). Mujava
: An automated class mutation system. Journal of
Software Testing, Verification and Reliability, 15:97–
133.
Sjøberg, D. I., Hannay, J. E., Hansen, O., Kampenes, V. B.,
Karahasanovic, A., Liborg, N.-K., and Rekdal, A. C.
(2005). A survey of controlled experiments in soft-
ware engineering. IEEE transactions on software en-
gineering, 31(9):733–753.
Sun Microsystems (2006). Sun’s XACML Implementation.
http://sunxacml.sourceforge.net/.
Utting, M., Pretschner, A., and Legeard, B. (2012). A tax-
onomy of model-based testing approaches. Software
Testing, Verification and Reliability, 22(5):297–312.
Weldehawaryat, G. K. and Katt, B. (2018). Towards a
quantitative approach for security assurance metrics.
In The Twelfth International Conference on Emerging
Security Information, Systems and Technologies; SE-
CURWARE 2018 September 16, 2018 to September
20, 2018-Venice, Italy. International Academy, Re-
search and Industry Association (IARIA).
Wohlin, C., Runeson, P., H
¨
ost, M., Ohlsson, M. C., and
Regnell, B. (2012). Experimentation in Software En-
gineering. Springer.
Yahya, F., Walters, R. J., and Wills, G. B. (2015). Using
goal-question-metric (gqm) approach to assess secu-
rity in cloud storage. In International Workshop on
Enterprise Security, pages 223–240. Springer.
Yoo, S. and Harman, M. (2012). Regression testing mini-
mization, selection and prioritization: A survey. Softw.
Test. Verif. Reliab., 22(2):67–120.
MODELSWARD 2020 - 8th International Conference on Model-Driven Engineering and Software Development
176