Journal of Software Engineering and Applications, 2012, 5, 351-363 Published Online May 2012 (
Framework for Extensible Application Testing
Agnieszka Zielińska
Computer Science Institute, AGH University of Science and technology, Kraków, Poland.
Received February 26th, 2012; revised March 27th, 2012; accepted April 29th, 2012
In recent years large corporations as well as smaller commercial enterprises have begun to devote increased attention to
software testing and software quality. This paper introduces a novel tool—the Framework for Extensible Application
Testing (FEAT), implemented by the author and applicable for automatic generation and execution of test cases. The
paper discusses system requirements, design, architecture and modes of operation. It also contains a detailed compari-
son of the FEAT framework with existing test environments, focusing in particular on the STAF/STAX framework. The
final section is devoted to experimental research into the applicability and efficiency of the presented tools in various
projects and configurations, as reflected by quality metrics.
Keywords: Software Testing; Quality Assurance; Testing Automation; Framework for Extensible Application Testing
1. Introduction
In recent years, commercial entities dealing with com-
ponent software development have shown increased
concern with software testing and software quality issues.
Tests need to be frequent, thorough and extensive;
moreover they need to cover a wide range of use cases
and potential applications of software components. Test-
ing processes employ—with ever-increasing frequency—
automated tools and frameworks, enabling developers to
reduce the complexity of software solutions and ensure
their optimal quality. Modern testing environments carry
many benefits, yet they are not free of certain drawbacks
and limitations such as language lock-in, poor support for
inter-scenario dependencies, lack of parallel testing me-
chanisms for distributed architectures and lack of cen-
tralized test resource management tools.
This paper introduces a novel testing environment
called FEAT (Framework for Extensible Application
Testing) developed by the author and capable of auto-
matically testing component software. The environment
is readily adjustable and can be customized to match the
requirements of various testing teams. It boasts many
features which aren’t found in popular testing suites cur-
rently available on the market.
The first section of the paper discusses basic issues
relevant to software testing. This is followed by a discus-
sion of testing principles and a general description of the
testing process itself, where we distinguish specific
modes of testing. Following this discussion, we present
the key aims of the FEAT automatic testing environment
—the author’s main contribution to the field. We also
discuss the general premises on which the system is
based as well as its architecture. An introduction to the
development of test scenarios is presented, along with a
description of test execution in the proposed environment
Section 4 opens the experimental part of the paper,
comparing the features of FEAT with those provided by
popular testing frameworks such as JUnit and TestNG, as
well as advanced environments, including STAF/STAX.
We discuss the advantages and drawbacks of each solu-
tion and present the concepts which have been adopted
and further elaborated by the author in the scope of
Having presented a conceptual outline of the FEAT
framework we follow through with an analysis of the
tool’s applicability to two specific IT projects, differing
with respect to the size of development teams and budg-
ets. The results of this study are presented in Section 5.
The paper ends with a recapitulation of the project’s
goals, conclusions drawn from the above-mentioned use
cases and prospects for further development of the FEAT
2. Basic Concepts Related to Software
The primary goal of software testing is to identify any
defects which may be present in the system being devel-
oped. A defect may occur during any phase of the soft-
ware design, implementation and maintenance process. It
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
may result from coding errors (also called bugs), omis-
sions or misinterpretations of the required functionality
on the part of system developers [2].
Testing is a crucial element of software development
since it guarantees that the application will perform as
expected in all circumstances. Some testing methodolo-
gies actually extend this definition by checking whether
the application doesn’t do more than is expected of it. In
either case, testing works to protect the user from soft-
ware errors which may—depending on the circumstances
—result in loss of time, financial resources or even hu-
man lives [3].
2.1. The Basic Principles of Testing
Testing typically follows the schema outlined in Figure
1. This standard is widely applied throughout the IT
world, and consists of a set of rules defined by respected
authorities in the field. For instance, Davis mentions the
following principles of software testing [4]:
All tests must reflect client requirements.
Testing should be planned far in advance of its actual
Testing should follow the Pareto principle (80/20):
according to this observation 80% of all software er-
rors are caused by 20% of software components. It is
the goal of tests to identify these critical components
and thoroughly verify their correctness.
Testing should begin with general assumptions and
then progress to specific criteria.
Exhaustive testing is impractical.
Tests yield better results when conducted by third
2.2. Types of Tests
Two different approaches may be applied to software
testing. The first is to conduct functional tests, i.e. view
the system from the user’s perspective and treat it as a
“black box” which is expected to perform certain tasks.
In this case the tester is not concerned with the inner
workings of the software being tested.
Not surprisingly, such tests are also referred to as
black box testing [4]. The other approach bases on struc-
tural tests, where the tester has access to source code and
may observe the behavior of individual components of an
application (such as modules and libraries). This mode is
sometimes called white box testing. A typical example of
structural testing involves unit tests where the tester (or
the programmer) develops additional code used solely to
verify the operation of a specific portion of the applica-
tion’s production code [4].
Tests may be conducted manually, by the tester, with
the use of the application’s user interface and a checklist
of actions, or automatically, where there is no need for a
human actor.
While both of the presented divisions focus on the
means of conducting tests, another distinction can be
introduced by considering the scope of testing [5]:
Unit tests;
Integration tests;
System tests.
There is also a division of tests with respect to their
purpose, which we view as the most interesting of all
testing hierarchies. It permits us to select tests suitable
for a given goal. This division is outlined below [6]:
Acceptance tests;
Functional tests;
Regression tests;
Performance and stress tests;
Installation and configuration tests;
Alpha and beta version tests;
Usability tests;
Failure recovery tests.
It should be noted that each development phase of an
IT project can be related to a specific testing phase. This
relationship is outlined in Figure 2.
3. The FEAT Component Application
Testing System
The initial requirements for our testing tool were gath-
Test cases Test data Test results Test report
Create test
Prepare test
Run program on
test data
Compare results
with test cases
Figure 1. Software testing [4].
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing 353
Requirements definition
Component tests
Integration tests
System tests
Acceptance tests
Architecture design
Detailed design
Mappingg requirements
to software
Figure 2. The relationship between project development and
testing phases [4].
ered by studying the habits and opinions of teams devel-
oping component software in a major American corpora-
tion [7]. The presented environment can be applied by
various corporate development teams to conduct auto-
mated software testing.
Following a thorough analysis of requirements we
have formulated some basic assumptions which underpin
the FEAT environment, presented in Figure 3:
1) The environment is to be written in Java;
2) The environment must enable easy, automated test-
ing of distributed Java-based component applications
with focus on simplicity and reusability of tests;
3) The environment must enable easy, automated test-
ing of distributed Java-based component applications
with focus on simplicity and reusability of tests;
4) The environment should focus on testing the APIs
exposed by containers and components deployed in such
containers. Supported containers should include J2EE
and OSGI application servers as well as any container
which provides a Java VM-based platform for pluggable
software modules;
5) FEAT works by wrapping a test engine in a plug-
gable module so that it matches the container technology
in use. Thus, each type of container should correspond to
a specific wrapping;
6) The proposed environment does not approximate a
Test tool
J2EE Application Server
Figure 3. Generalized vision of the placement of the proposed
FEAT testing environment [1].
JUnitEE-type system as unit tests are, by definition, sim-
ple and should not depend on external tools. Rather, the
tests which can be conducted using FEAT can be arbi-
trarily complex and scalable;
7) The author focuses on developing universal com-
ponents from which tests can be constructed and which
can themselves be reused in various projects. Test de-
velopers must be prevented from creating implicit de-
pendencies between tests;
8) The environment should support a flexible hierar-
chy of tests and permit results to be archived in ways ap-
propriate for a given IT project and selected testing meth-
9) When creating a specific test case, the developer
should focus on abstract actions (e.g. install a product;
verify the integrity of a database etc.) rather than on spe-
cific, platform-dependent implementations (taking into
account the operating system, DBMS, application server
etc.) In other words, the environment should conceal tech-
nical details from the user, relegating them to a separate
(hidden) layer;
10) The system should ensure access to the test execu-
tion environment and facilitate interaction of tests with this
environment (contrary to unit tests, which do not require
access to such an environment);
11) The system should provide a selection of useful
auxiliary features such as file downloads, logging, result
storage etc.;
12) The system should feature advanced visualization
of results and generation of testing reports;
13) The system should introduce a mechanism for
automatic removal of test engines and/or other resources
(e.g. databases), restoring the initial state of the tested
14) The system should support pluggable installation
and execution layers (such as SSH and STAF), response-
ble for deploying test engines on remote machines. The
transport layers used for remote execution of tests should
also be pluggable (e.g. RMI, servlets, custom protocols
based on serializable Java objects etc.) [1].
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
3.1. System Overview
A general overview of the architecture of the FEAT test-
ing environment is presented in Figure 4. Below we de-
scribe the basic principles of its operation [1]:
The Controller receives test Scenarios;
The Scenerio Manager scenarios (which may be
written in an arbitrary programming language) and
converts them to Java classes;
Classes are forwarded to the Test Execution Core
which executes the scenarios by invoking test frag-
ments on appropriate test engines, deployed on re-
mote hosts;
The Resource Manager should gather information on
the resources available for testing, including hosts,
databases, application servers etc. The Test Execu-
tion Core should reserve selected resources (on de-
mand) and assign them to a given scenario (by sup-
plying its description). Resource access is query-
based: for example, the user may request a machine
with DB2 v9.1 and WebSphere v6.1 installed. Thus,
test developers only indicate generic requirements in-
stead of selecting specific machines for testing;
The Resource Manager may be extended with
plugins and/or interface with an external management
system (for the purposes of reserving access to hosts);
Test Engines are installed on remote machines. Once
tests are complete, these engines are automatically
uninstalled to restore the local environment to its ini-
tial state;
Test Engines execute test fragments and return re-
sults to the Test Execution Core from which they are
dispatched to the Results Manager;
The Results Manager may also be extended with
plugins, for example to write results to files, custom
databases etc.;
The Visualizer be attached to the Results Manager in
order to display graphical representations of test re-
The testing environment is configured in two phases.
Phase one involves configuring the controller by deter-
mining which Resource Manager and which Result
Manager it should interact with; which mode it should
operate in and which services it should expose. This con-
figuration does not directly influence tests. Phase two
focuses on verifying the correctness of the test scenarios,
checking their dependencies, determining the order in
which tests should be executed etc.
3.2. System Architecture
One of the key aspects of FEAT is its layered architect-
ture. Each layer is dedicated to a specific goal and may
interact with other layers in pursuit of that goal. The lay-
Database Plugin
Tests Execution CoreTest
Result database Result File
Test Results
Files Plugin
Results Manager Test Machine 2
Test Machine 1
java object for
java object
represe nting
J2EE Application
Figure 4. General architecture of the FEAT testing environment [1].
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing 355
ers depicted in Figure 5 are briefly discussed below:
1) Transport Layer: executing Java code on remote
2) Deployment Layer: installing test engines on re-
mote hosts;
3) Services: services such as file download, file trans-
fer etc.;
4) Resources: various resources, including test hosts,
databases, application servers etc.;
5) Test Engine API: an API facilitating access to
lower layers (Transport Layer, Deployment Layer, Ser-
vices, Resources). It manages these layers and exposes
their functionality to upper layers (Test fragments and
installers and Test Scenarios);
6) Test fragments and installers: this layer executes
test fragments (each test fragment is an atomic unit, in-
dependent of other fragments);
7) Test Scenarios: this layer executes entire test sce-
3.3. Development of Test Cases
At this point it seems appropriate to explain how test
scenarios are created. Each test scenario corresponds to a
specific test case and consists of the following elements
Scenario identifier (ID);
Scenario name;
Textual scenario description;
Dependent scenarios (which need to be executed prior
to the given scenario);
Programming language in which the scenario has
been implemented;
Name of class implementing the scenario;
Resources required for the test scenario to execute
(resource name, resource class and—optionally—
other parameters such as version number).
A sample XML file describing two simple scenarios is
presented below:
<?xml version=“1.0” encoding=“UTF-8”?>
<scenario id=“scen_20”
<description>Test if database is cor-
rectly installed</description>
<resource name=“DB”
class=“feat.resources.Database” re-
<scenario id=“scen_21” name=“Scenario2”
depends=“scen_20” >
<description>This scenario tests
creation of database
<resource name=“DB”
class=“feat.resources.Database” re-
<property name=“type”>DB2</property>
<property name=“version”>9.1</property>
The <testsuite> tag indicates a test suite and
contains a separate <scenario> tag for each scenario
which belongs to the given suite. The <description> tag
contains a textual description of the scenario. Each sce-
nario must also specify a list of required resources by
supplying their abstract names For instance,
<resource name=“DB”
means a database with the name DB2, managed by the
feat, resources, database class. This database is required
for the test scenario to operate (required = “true”).
In addition, each scenario follows a specific course of
action which may be described using various languages
Transport Layer
Test Scenarios
Tset fragments and installers
Peployment Layer Services Resources
Test Engine API
Figure 5. Layered architecture of the FEAT testing environment [1].
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
(for example, by a Java class or a Jython script). The
required language and the name of the scenario file are
given as:
Test scenarios may be bundled in test suites and may
also specify internal dependencies (for instance, “if a
given scenario fails, do not execute these dependent sce-
narios”). Such dependencies are expressed via the de-
pends attribute in the scenario description. For example,
the following description
<scenario id=“scen_21”
name=“Scenario2” depends=“scen_20” >
3.4. Test Execution Process
Given the proposed architecture, automatic execution of
tests in the FEAT environment proceeds according to the
following plan [1]:
1. Controller initialization:
1.1. Loading resources.
1.2. Installation of test engines on remote hosts.
2. Scenario enactment:
2.1. Loading scenario.
2.2. Converting scenario to Java object.
2.3. Enactment of the next waiting scenario:
2.3.1. Allocating resources.
2.3.2. Retrieving/creating test fragments.
2.3.3. Executing test fragments on test engines.
2.3.4. Retrieving results.
2.3.5. Uninstalling test engines installed by the
2.4. Collating scenario results.
3. Closing and uninstalling engines.
3.5. Testing Progress Report
While test scenarios and test fragments are executing on
remote hosts, it is possible to monitor their progress and
results via a Web-based interface. To this end, the Jetty
Figure 6. Testing progress report [1].
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing 357
web server is used to serve a website at http://localhost:
8088. Key configuration info is available under the Gen-
eral tab:
Main: information about the system on which tests
are being executed: version of the FEAT testing en-
vironment, build version of the tested application,
elapsed testing time, ID and full name of the operat-
ing system, Java environment version, JAVA_HOME
value, total RAM, free RAM and number of available
processor cores;
Scenarios: information on the executing test scenar-
ios: scenario ID, programming language (Java, Py-
thon etc.), scenario name;
Test Fragments: information on individual test frag-
ments which make up the scenarios: fragment name,
class, platform, resources utilized;
Execution: information on the current state of the
executing scenarios (Scenarios) and test fragments
(Test Fragments). This menu also enables users to
view execution logs (Log browsing) (Figure 6).
4. Comparison of JUnit, TestNG,
Let’s assume that we have been tasked with testing
component software developed by a large corporation.
The development team consists of experts from around-
the world, operating in various time zones. Let us fur-
thermore assume that a test build of the software is cre-
ated daily around 1:00 a.m. Tests are to be divided into
the following phases [1]:
BVT (Build Verification Tests)—performed each
time a new build is deployed;
CVT (Component Verification Tests)—if BVT pass
successfully. CVT consist of regression tests con-
ducted on a range of hosts with different configure-
Which tool—JUnit, TestNG, STAF/STAX or FEAT—
would ensure the most efficient, most reliable and fastest
testing under the proposed conditions? In order to answer
this question, let’s analyze the information presented in
Figure 7, which lists individual testing stages and the as-
sociated benefits of each testing environment.
As can be seen, the highest efficiency—both for BVT
and CVT—is provided by the FEAT environment. Tests
can be performed automatically, with no human interact-
tion, and may occur at any time (e.g. immediately upon
the deployment of a new build, even late at night). When
the testers come to work on the following day, they can
peruse an automatically generated report detailing the
conducted tests. It should be noted that the FEAT envi-
ronment combines the benefits of the solutions presented
in the previous sections. It derives its support for small
test fragments from JUnit and inherits support for test
suites from TestNG. In addition, it enables automatic
execution of individual steps of the testing process; thus
it shares many features with the STAF/STAX environ-
ments frequently used in large corporations [8].
It is the goal of FEAT to fully automatize testing, in-
cluding preparation of the testing environment. This is
why the author has extended the proposed tool with addi-
tional features such as automatic installation and execu-
tion, file transfers, automatic allocation of testing re-
sources and centralized logging.
We can therefore conclude that the FEAT environment
exhibits all the attributes of a self-contained automatic
software testing solution and that—in terms of novel
features—it compares favorably to the environments and
tools currently available on the market. It can be ex-
pected that the concepts upon which FEAT bases will
one day become standard practice in commercial and open-
Test monitoring
Test execution
Small, reusable test fragments
Development of comliacted scenarios
Development of tests in multiple languages
Dependent tests
Complex test execution
Complex test scenarios
Assignment of test machines
File transfer between machines
Installation of new application version for tests
Allocation of test resoures
Logs and resuit analysis
Support for exchangeable modules
JUnit TestN
manual manual
manual manual
Figure 7. Comparison of JUnit, TestNG, STAF/STAX and FEAT [8,9].
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
source testing solutions, including those employed by
large IT enterprises.
Comparison of Test Implementations in
STAF/STAX and FEAT Environments
Compared to STAF, FEAT is better suited for Java-based
environments since it does not require additional virtual
machines to operate. As FEAT may be executed in any
Java virtual machine, no additional code is required to
perform mediation between the internal interface of the
tested application and the testing environment itself.
STAF only supports communication via basic strings
and lists. In contrast, FEAT enables transfer of configu-
ration parameters, data and test results as serialized Java
objects with arbitrary internal structure. During devel-
opment of tests, syntactic verification occurs at the com-
pilation stage, and not—as in STAF—at runtime [10].
When preparing tests in an IDE (such as Eclipse) the
programmer may take advantage of the code autocom-
pletion feature to find out which arguments are required
for a given test fragment. This reduces the need to con-
sult documentation and increases code reusability.
STAF is a highly generic environment, built upon the
notion of agents and services deployed therein. Efficient
development of tests requires users to deploy an addi-
tional layer of testing logic suited for a particular project.
In contrast, FEAT aims to support creation of tests “out
of the box”. As the test suite grows and matures, it may
be extended by adding (or exchanging) plugins, installers
or fragments, usually with no need to rewrite code [10].
FEAT defines a selection of interfaces and contracts
which need to be implemented by the test code. This
feature results in reusable modules which can be repur-
posed not only within the context of a specific project,
but in the scope of an entire enterprise (or branch
Software testing often calls for examining the same
functionality under various hardware and software con-
figurations (different operating systems, application con-
tainers, databases etc.) FEAT supports this goal by sepa-
rating the generic test scenario from implementation- and
configuration-specific details. Test fragments may be
differentiated into versions, depending on the engine in
which they’re executed. Moreover, each scenario may be
executed on various sets of resources, which enriches the
set of test configurations without calling upon the user to
develop additional scenarios.
Finally, FEAT supports central management of re-
sources used in tests. The Resource Manager module may
be swapped while the central database remains unchanged.
This feature mitigates the problem of having several
testers share a single machine and interfere with one an-
other’s work.
5. Experimental Assessment of the Manual
Software Testing Process with
STAF/STAX as Compared to FEAT
In the following subsection of our work, we will discuss
the software testing process on the basis of two distinct
use cases [1]:
a small enterprise employing up to 5 developers and
testers with overlapping areas of responsibility, ap-
plying the MSF methodology.
a large corporation with a well-defined organizational
structure and clearly delineated developer/tester re-
sponsibilities, applying the RUP software develop-
ment methodology. This use case is based on infor-
mation acquired from a large real-world international
For each of the presented cases we will briefly de-
scribe the testing methodology, development team com-
position and software being produced. Following a pre-
liminary analysis of the project team, we will assess the
time required to manually perform tests using STAF/
STAX and FEAT. This will enable us to compare the cost-
effectiveness of test preparation and execution phases.
Our analysis will yield conclusions regarding the ap-
plicability and effectiveness of selected tools for different
teams and software projects.
5.1. Quality Metrics
Published works [11,12] propose a great variety of met-
rics and quality assessment methods which may be ap-
plied by gauging the cost and expected duration of a pro-
ject—e.g. COCOMO (COnstructive COst MOdel) or the
function points method. For the purposes of the presented
work, we will only discuss some basic quality metrics
and means to determine the percentage of undiscovered
errors. Particular attention should be devoted to the qual-
ity of the testing process itself—its effectiveness and the
reliability of the results it yields – as these values enable
us to determine the quality of the software being tested.
The following metrics may be used when assessing the
quality of testing [13]:
Requirements Coverage metric: numerical assess-
ment of the portion of requirements which have suc-
cessfully passed tests. The following formula applies:
RC = no. of requirements (P, I, E, S)/no. of require-
ments (A, T).
This metric may be derived at many stages in the test-
ing process, to determine the coverage of code by planned
(P), implemented (I) or executed (E) tests. Moreover, it
can be used to assess how many requirements have
passed tests successfully (S). This value may be com-
pared to the total number of requirements (A) or to the
number of requirements which were subjected to testing
(T). In the context of testing quality, the requirements
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing 359
coverage metric can serve as the coefficient of tested
requirements, namely:
RC = no. of tested requirements (E)/total no. of re-
quirements (A).
Code Coverage metric: this metric enables us to de-
termine the fraction of source code subjected to test-
ing. It is given by the following formula:
CC= no. of executed code units/total no. of code units.
The number of code units is defined as the number of
lines of code or, alternatively, as the number of potential
control paths that can be traversed. Source code should
be understood as an expression of the developer’s inten-
tions; however it must be noted that even complete code
coverage and absence of errors do not warrant a good
match between the resulting software and the end user’s
Defect Removal Effectiveness metric, usually given as:
DRE = no. of errors identified during testing/total no.
of identified errors.
The total number of identified errors is the sum of the
number of errors identified during testing and the number
of errors identified following testing (i.e. during deploy-
ment or during production runs involving the end user).
This metric indicates how efficient the testing process
is in identifying errors; however in order to accurately
derive it we must remain aware of how many errors are
identified by end users. Thus, our assessment must be
based on historical data and take into account the ex-
periences of similarly competent testing teams working
on similar projects with the use of similar testing meth-
The Defect Removal Effectiveness metric is applied
by the CMM (Capability Maturity Model) to assess the
maturity of an IT enterprise.
5.2. Case Study 1: Small Enterprise
5.2.1. Team Description
The small enterprise development team consists of 8
persons, one of whom plays the role of product manager,
project manager and release manager. That person is re-
sponsible for planning, estimating the duration of project
phases, assigning tasks, overall project management and
release deployment. Four persons are employed as de-
velopers while the three remaining team members are
testers who nevertheless actively participate in develop-
ment (due to manpower limitations). Their assignments
may change depending on the circumstances. Each team
member handles clearly defined tasks. Team roles are
mutually dependent and shared. The team applies the
MSD (Microsoft Solution Framework) methodology.
5.2.2. Project Outline
The project is driven by client demands and has a spe-
cific timeframe for each implementation phase as well as
a final deadline for software delivery. The client is par-
ticularly concerned with meeting this deadline while en-
suring high quality of the delivered software.
Unfortunately, due to the lack of sufficient manpower,
the team finds it difficult to meet all of the client’s re-
5.2.3. Methodology
The team applies the MSF (Microsoft Solution Framework)
methodology, developed by Microsoft Consulting Ser-
vices and known throughout the world as an established
standard for the development of IT projects by Microsoft
and its partners.
In general, the project can be considered a success if it
is realized within budget and in agreement with the pre-
determined deadlines.
5.2.4. Tool Effectiveness Analysis
How should this type of project be handled, given the
specified methodology, requirements and team composi-
tion? The most significant problem seems to be the rela-
tively small size of the team and, in particular, the need
to intermittently engage testers in development of addi-
tional features. The question arises: how to optimally
exploit the available resources? The following analysis
begins with a phased implementation plan.
First, let us assume that the team does not perform any
automatic testing. All tests need to be conducted manu-
ally. Whenever new features are introduced, regression
tests must take place, given the strong dependencies be-
tween any new features and existing system components.
While such tests are being conducted, testers cannot par-
ticipate in further development, due to their relatively
high workload. Two different scenarios may be applied
at this point:
Testers focus on manual tests run in parallel with im-
plementation. This ensures better product quality but
at the cost of missing deadlines (testers cannot engage
in development and the team is reduced to four active
developers—cf. column 1 in Figure 8).
A single tester focuses on testing while the remaining
two collaborate with developers on implementation of
new features. The project proceeds as scheduled,
however it may fail to meet client requirements due to
the relatively low testing coverage (column 2 in Fig-
ure 8).
Thus, given a manual testing regimen, it seems un-
likely that all the goals of the project can be met. Ei-
ther the project will be delayed or the quality of the
delivered software will be compromised.
Let us now consider the potential advantages of ap-
plying automatic testing tools in the scope of this project:
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
Time for tests (%) Time for development (%)
1 2
Time [%]
Figure 8. Comparison of project implementation time given
manual (1) and automatic testing (2) [1].
During the initial implementation phase all three test-
ers design automatic tests of expected features, basing
on the conceptual system design. Implementation
proceeds slowly.
Once the tests are implemented, testers may devote
their attention to development of additional features
while devoting only a small percentage of their work-
load to analysis of test results. This greatly speeds up
further development.
Automatic tests can proceed in parallel with imple-
mentation throughout the duration of the project (column
2 in Figure 9). As testers are not required to personally
execute and monitor tests, they can devote more of their
time to implementation work.
Let us now analyze the project from the perspective of
IT industry standards:
Number of requirements: 150;
Number of implemented classes: 80;
Lines of code: 80 classes * ca. 1000 lines = 80000
Lines of test code: 60 classes * ca. 1000 lines = 60000
Under these conditions the implemented tests would
cover approximately 90% of code, exposing appro-
ximately 94% of all errors and covering approxima-
tely 130 requirements;
The quality metrics, as given by formulas presented
in Section 5.1, would therefore equal:
RC = 130/150 = 0.86
CC = 72,000/80,000 = 0.9
DRE = 0.94
From the above analysis and the derived quality metrics
Time for automatic tests results analysis (%) (1) and time for
automatic tests execution (5) (2)
Time for development (%)
1 2
Time [%]
Time for automatic tests implementation (%)
Figure 9. Manual tests: 1—deadline missed, acceptable
software quality; 2—deadline met, insufficient software
quality [1].
it follows that automatic testing would enable this project
to be delivered on time, with no undue impact on soft-
ware quality. Thus, automatic testing proves efficient and
highly desirable from the business standpoint. No other
testing environment seems to fulfill the stated require-
ments of this project, given the circumstances under
which it is being implemented.
5.3. Case Study 2: Large Enterprise
5.3.1. Team Description
This use case bases upon data gathered while working
with automatic testing tools in a large real-world corpo-
ration. The development team consists of developers,
testers and a project manager. The role of each team
member is explicitly defined. The team is geographically
distributed; some members operate in Poland while oth-
ers are based in the USA. The impact of the FEAT on
business goals will be assessed based on data from two
real-life IT development projects.
5.3.2. Project Outline
The goal of the project is to deliver a custom component
to external clients. The team applies the RUP (Rational
Unified Process) iterative methodology. Extensive test-
ing of the product, covering as much of its functionality
as possible, is critically important. Tests are to be divided
into the following phases:
BVT (Build Verification Tests)—performed each
time a new build is deployed;
CVT (Component Verification Tests)—performed if
BVT pass successfully;
IVT (Integration Verification Tests)—integration
SVT (System Verification Tests)—system-wide tests;
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing 361
RT (Regression Tests).
Proper testing coverage is a prerequisite of suitable
QA (Quality Assurance).
5.3.3. Methodology
As mentioned above, the development team applies the
RUP (Rational Unified Process) methodology, whose
principles, advantages and drawbacks are outlined in Sec-
tion 2 of this paper.
5.3.4. Tool Effectiveness Analysis
The analysis of the effectiveness of testing tools is based
on real-world data acquired from a corporation:
Test scenario: Selected scenario from the BVT;
Condition: full automation of test scenario, including
environment configuration, post-test cleanup and col-
lation of test results for analysis;
As depicted in Figure 10;
Duration of a single test scenario:
when using SSH, JunitEE and ant: 28 h—a sig-
nificant portion of that time has to be devoted to
proper implementation of the test, preparation of
the testing environment and post-test cleanup;
when using the FEAT environment: 8.7 h;
Time saved: 28 h - 8, 7 h = 19.3 h for a single test
50 requirements per functionality;
2 to 5 (average: 3.5) scenarios per requirement = 175
test scenarios;
Number of implemented classes: 120;
Lines of code: 120 classes * ca. 1000 lines = 120000
Lines of test code: 200 classes * ca. 1000 lines =
200,000 lines;
Time for results analysis
Automation with SSH Automation with FEAT
Time [hours]
Time for test execution
Time for test scenario im
Figure 10. Manual tests with SSH and JUnit and with
Under these conditions the implemented tests would
cover approximately 95% of code, exposing ap-
proximately 97% of all errors and covering approxi-
mately 48 requirements;
The quality metrics, as given by formulas presented
in Section 5.1, would therefore equal:
RC = 50/48 = 0.86
CC = 114,000/120,000 = 0.95
DRE = 0.97
Business impact:
3377.5 h saved (175 * 19.3 h);
422 PD (Project Days) or 84 PW (Project Weeks) or
21 PM (Project Months) or nearly 1.8 PY (Project
Years) saved;
costs reduced by 211K USD (assuming 120KUSD/
5.3.5. Conclusion
The above analysis suggests that FEAT-type automated
testing environments carry advantages for large-scale IT
projects as they ensure repeatability and increased quality
of tests while saving time and reducing costs. This trans-
lates into significant improvements both in terms of qual-
ity metrics and business competitiveness.
6. Summary
The main goal of the presented work was to develop a
custom environment called FEAT (Framework for Ex-
tensible Application Testing), which could be used to
automatically test component software while ensuring
high adaptability to the requirements of various testing
teams The presented environment can be applied to eas-
ily and automatically test distributed component applica-
tions written in Java, with focus on simplicity and reus-
ability of tests. It focuses on tests of APIs provided by
containers and components deployed therein. Supported
technologies include J2EE and OSGI application servers
and all other containers which provide a pluggable com-
ponent-based system operating within the Java VM. The
presented environment does not attempt to emulate
JUnitEE as unit tests are, by definition, simple and
should not call upon external dependencies. Tests con-
ducted with the use of FEAT may be arbitrarily complex;
however, we wish to stress the tester’s ability to develop
universal components which can be assembled into tests
of various software systems. In addition, FEAT provides
support for a flexible testing hierarchy where results can
be logged and presented in accordance with the require-
ments of specific projects and specific testing method-
ologies. In developing a test case, the tester focuses on
abstract actions (such as: “install a product” or “verify
the integrity of a database”) while omitting low-level
implementation details, relating to given operating sys-
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
tems, database management systems or application serv-
ers. In other words, the proposed environment relegates
all technical aspects to a separate layer which is con-
cealed from the end user. FEAT provides a testing envi-
ronment and ensures suitable interaction between this
environment and the tests themselves (contrary to unit
tests which do not require any such environment). More-
over, FEAT comes with a set of useful support services,
including on-demand file transfer, logging, collating re-
sults etc. It also contains an advanced visualization sub-
system which may be used to generate testing reports [1].
The FEAT environment combines the advantages of
several other testing platforms mentioned in this paper.
For instance, it derives the notion of small test fragments
from JUnit and the ability to compose test suites from
TestNG. In addition, it enables automatic execution of
individual steps of the testing process thus it shares many
concepts with the STAF/STAX environments frequently
used by large corporations.
It is the goal of FEAT to fully automatize testing, in-
cluding preparation of the testing environment. This is
why the author has extended the proposed tool with addi-
tional features such as automatic installation and execu-
tion, file transfers, automatic allocation of testing re-
sources and centralized logging.
The use case analysis presented in Section 6 indicates
that automatic testing environments, including FEAT,
can be beneficial both for medium-scale and large IT en-
terprises. Such environments facilitate the testing process,
increasing its reproducibility and efficiency, while con-
serving the financial resources assigned to the project
and thus improving the enterprise’s competitiveness.
The FEAT environment exhibits all the attributes of a
self-contained automatic software testing solution. In
terms of novel features it compares favorably to the en-
vironments and tools currently available on the market.
We can conclude that the concepts underpinning this
environment will one day become standard practice in
commercial and open-source testing solutions, including
those applied in large IT enterprises.
The FEAT environment, in its current form (as pre-
sented in this paper), is a complex, feature-rich system.
Even so, its modular architecture supports further exten-
sions which can be implemented by developing addi-
tional modules. One example of such an extension would
be to swap the application container for a Web browser.
Under such circumstances, the testing plugin would be
installed directly in a specific browser (Internet Explorer,
Mozilla Firefox etc.) and verify that a given Web page
has been correctly rendered by the browser. In order to
perform this test it would take a screenshot of the
browser window and then analyze its contents with the
use of image recognition algorithms.
Another possible extension might involve a groupwise
resource manager, which would monitor (via a dedicated
database) all the resources available to a group of testers,
including test hosts, application servers, databases and
external services. It could also support a booking system,
enabling each tester to reserve access to certain resources
for a specified amount of time. The resource manager
would then ensure that no single resource is simultane-
ously assigned to two different testers. It might manage
resources on a project-wide or company-wide basis while
delivering reports and recommendations (for instance, if
a given test host with a specific operating system proves
to be a highly contested resource, it might issue a request
for another such machine to be provided to the testing
The FEAT testing environment could also provide
virtualization capabilities, particularly important for large
corporations. To this end it could be extended with a
dedicated subsystem responsible for managing virtual
system images and instantiating them on test hosts, as
required by testers.
In the context of potential wide-scale adoption of
FEAT, it might prove beneficial to create a database of
tests and toolsets, facilitating rapid development of new
tests and easy configuration of working environments.
This feature could be supported by a custom module or
by an existing software solution such as TPM (Tivoli
Provisioning Manager).
Yet another possible extension would be to develop
custom testing plugins for popular IDEs such as Eclipse,
NetBeans or Visual Studio. Configuring tests would then
be supported by a dedicated user-friendly GUI where the
tester might select items from a menu and arrange them
using a drag&drop mechanism. Tests could also be exe-
cuted from within the IDE itself, in a way similar to
JUnit integration in the Eclipse environment [1].
Clearly, the FEAT environment, as developed by the
author and presented in this paper, could be extended in
numerous ways, resulting in a highly complex and adap-
table framework.
[1] L. Pobereżnik and A. Zielińska, “Automatyzacja Procesu
Testowania Oprogramowania Komponentowego W Het-
erogenicznym Środowisku Produkcyjnym,” AGH, Kraków,
[2] G. J. Myers, C. Sandler, T. Badgett and T. M. Thomas, “Sz-
tuka Testowania Oprogramowania,” Helion, Gliwice, 2005.
[3] W. Hetzel, “The Complete Guide to Software Testing,”
John Wiley & Sons, New York, 1988.
[4] I. Sommerville, “Inżynieria Oprogramowania,” WNT, War-
szawa, 2003.
[5] R. Patton, “Software Testing,” 2nd Edition, Sams, Indian-
apolis, 2005.
Copyright © 2012 SciRes. JSEA
Framework for Extensible Application Testing
Copyright © 2012 SciRes. JSEA
[6] W. E. Perry, “Effective Methods for Software Testing,” 3rd
Edition, John Wiley & Sons, Indianapolis, 2006.
[7] J. D. McGregor and D. A. Sykes, “A Practical Guide to
Testing Object-Oriented Software,” Addison-Wesley Pro-
fessional, Upper Sadle River, 2001.
[8] C. Beust and H. Suleiman “Next Generation Java Testing:
Test NG and Advanced Concepts,” Addison Wesley, San
Francisco, 2007.
[9] A. Hunt and D. Thomas, “JUnit. Pragmatyczne Testy Jed-
nostkowe W Javie,” Helion, Gliwice, 2006.
[10] STAF/STAX documantation, 2009.
[11] C. Y. Laporte, “An Overview of Software Quality Con-
cepts and Management Issues,” Hershey, 2005.
[12] L. Westfall, “Defect Removal Effectiveness,” Austin, 1996.
[13] S. H. Kan, “Metrics and Models in Software Quality Engi-
neering,” 2nd Edition, Addison-Wesley Professional, Bos-
ton, 2003.