WEBTESTE: A STRESS TEST TOOL

Hugo Saba, Eduardo Manuel de Freitas Jorge, Victor Franco Costa

Nova Tecnologia Inform

atica

Rua Macaubas 520, Ap. 503, Rio Vermelho, Salvador/BA, Brazil

Hernane Borges de Barros Pereira

Fundac¸

ao Visconde de Cairu - CEPPEV

Rua da Brisa 42, Ap. 1701, Pituba - Salvador/BA - Brasil

Keywords:

Stress Test, Web Server, WebTeste.

Abstract:

The usage of web applications has became a very common activity in the organizations’ scope. From the

software engineering perspective, the incessant search for production of more robust softwares, with better

quality, is a continuous requirement. The main purpose of this paper is to present the WebTeste, which is a test

tool used to verify the robustness of a web application. After comparison of several simulation’s results, the

use of distributed and ordered computers suggests more reliable tests. In addition, the analysis of the obtained

results can suggest a new (re)design of a web system. The WebTeste could be used to perform stress test in

order to verify the robustness of a web application more precisely.

1 INTRODUCTION

The success of a test used to identify defects is ob-

tained when improper operations of the analyzed sys-

tem that is being veriﬁed are identiﬁed, showing that

anomalies there exist (Sommerville, 2000).

In the stress test case, (Bazzana, 2001) argues that

location of architectural bottlenecks, impact of hw/sw

changes and scalability of the solution are some im-

portant aspects with respect to start this kind of test.

Because of applications based on web are in con-

tinuous evolution, they have peculiar characteristics

different from others software categories (Peters and

Pedrycz, 2000).

Web applications must attend to a variable num-

ber of simultaneous accesses, which will be foreseen

by the designer. The stress test is very well applied

to the context of some web applications, once this

kind of application has requirements such as unin-

terrupted working (i.e. the application must work 24

hours/day) and attending several numbers of simulta-

neous requests (Trepper, 2000).

The main goal of this research is to specify and

design a test tool, called WebTeste, which performs

stress test to solve problems related to web application

failure. In order to analyze the effectiveness of the

WebTeste, some simulations based on different sce-

narios have been carried out.

2 THE WebTeste TOOL

The main proposal of the WebTeste is to perform

stress tests in web applications, generating informa-

tion that allows software designers to make the best

decision for the deﬁnition of architectural project. In

this way, the obtained information gives support to

designers during the ofﬁcial homologation process of

the web application architecture.

The WebTeste has been designed with the capac-

ity of emulating an unlimited number of users access-

ing to a web application (i.e. the capacity of realiz-

ing simultaneous requests) that will be tested. The

WebTeste has a distributed and extensible architecture

and uses a test plan deﬁned from the system require-

ments collected during the speciﬁcation of the web

application.

The WebTeste also uses as technology to simulate

the requests to the web applications, the HTTPUNIT

framework, a JUNIT extension (Httpunit, 2004). The

WebTeste is also composed of an algorithm to simu-

late the web access in a distributed and ordered way.

In this way, WebTeste guarantees that the stress only

occurs in the computer where the web application is

placed, obtaining a more rigorous test. The WebTeste

allows the designer to be sure about the dimension of

the web application architecture.

Considering that the deﬁnition of the test plan is

one of the major difﬁculties of test tools, the WebTeste

246

Saba H., Manuel de Freitas Jorge E., Franco Costa V. and Borges de Barros Pereira H. (2006).

WEBTESTE: A STRESS TEST TOOL.

In Proceedings of WEBIST 2006 - Second International Conference on Web Information Systems and Technologies - Internet Technology / Web

Interface and Applications, pages 246-249

DOI: 10.5220/0001246802460249

 SciTePress

is also composed of a module, called WebMonitor,

that will help in this task. This module, in a interactive

and simple way, makes quickly and reliably possible

the editing of a test case. Succinctly, this work con-

templates (1) the possibility of realizing distributed

and synchronized tests, (2) the automatic creation of

the test plan by the WebMonitor module, (3) the uni-

ﬁcation of the logs generated by each solicitant com-

puter in the case of distributed tests and (4) a method

of the monitoring and analysis of results.

2.1 The WebTeste Architecture

During the WebTeste development some problems

have been identiﬁed and some changes have been nec-

essaries.

The WebTeste consists of a distributed and ordered

architecture composed of four components (Figura 1):

Supervisors (A)

HTTP/Requests

(D)

Web Environment (B)

Connections

Computer with Web

Application

Log (C)

Figure 1: WebTeste Architecture.

• Set of Supervisors (A): Component responsible

for the management process of all test. It sends

the test plan to the other participants of the test and

puts together the logs created by them, generating a

centralized log, that is stored in XML format. This

allows to guarantee the synchrony among all tests

by the supervisors. This component establishes the

ﬂow of tests to the web environment;

• Web environment (B): It can be composed of sev-

eral elements (e.g. ﬁrewalls, web servers in clus-

ter, application servers etc.). The web application,

used to the validation of the WebTeste architectural

model, is placed in this component;

• Log (C): It represents the centralized log that is

generated by the set of supervisors during the sim-

ulation of the requests to the web environment;

• HTTP/Requests (D): This component corresponds

to the requests to the web environment.

2.2 WebTeste Requirements

WebTeste is composed of functional and no-functional

requirements. Functional requirements consist of a

set of information (i.e. Creation of Test Plan, Con-

ﬁguration of distributed test and Uniﬁcation of logs)

based on three XML ﬁles (i.e. testplan.xml,

config.xml, logs.xml).

On the other hand, no-functional requirements are

composed of the following ﬁve elements: Creation of

Test Plan, Distributed Object Remote Method Invoca-

tion (RMI), Web Request Framework - HTTPUNIT

framework, Algorithm of Synchrony based on New-

top algorithm (Ezhilchelvan et al., 1995) and Test En-

vironment.

3 SIMULATIONS

Several simulations have been carried out. In this pa-

per, the conﬁgurations of some of them are presented.

These conﬁgurations have taken into account three

test plans (i.e. three speciﬁc tasks run in a web ap-

plication), two scenarios (i.e. homogeneous and het-

erogeneous laboratories, in which the distributed tests

have been performed), and one centralized test that

does not depend on the type of the laboratory, but

the computer conﬁgurations and the network connec-

tions.

• Test Plans have been created by the WebMonitor

and used in the scenarios #01 and #02 (Figures 2,

and 3, respectively). The test plan used in the sim-

ulations represents a purchase in a virtual shop and

consists of two, ten and twenty requested links to-

wards the purchase conclusion or while the user

navigate in web application;

• The main goal of the proposed types of tests is

to demonstrate the saving of response time when

the test is performed in a distributed way with re-

spect to the centralized test, avoiding the stress in

the computers that generate the requests:

– The centralized test is performed from a com-

puter that simulates all deﬁned requests to web

server;

– The distributed test is performed from two or

more computers to web server; each one sim-

ulating the half or less of all deﬁned requests.

Distributed tests are organized in two categories

with respect to the conﬁguration of the comput-

ers used in the test: homogeneous and heteroge-

neous laboratories.

For this, both centralized and distributed tests have

been performed in the two simulation scenarios

(Figures 2 and 3) taking into account different

quantities of threads in blocks from 100 (minimum

quantity) to 2000 with an increment of 100, and

from 2500 to 10000 (maximum quantity) with an

increment of 500.

WEBTESTE: A STRESS TEST TOOL

247

3.1 Scenarios

The web architecture for the WebTeste simulations has

consisted of the Tomcat web server (Tomcat, 2004),

version 5.5, working in a computer Intel Pentium Mo-

bile 1.60 gHz with 512mb of RAM and Microsoft

Windows XP Professional. In addition, eight com-

puters with different conﬁgurations distributed in two

scenarios (i.e. homogeneous and heterogeneous labo-

ratories, Figures 2, and 3 respectively) have been used

to simulate the supervisors deﬁned in the current ver-

sion of the WebTeste architecture and to perform the

stress test in a web application.

Web Server

PC #1

Ethernet (10/100 Mbps)

Processor: AMD DURON 750

RAM: 128 mb

HD: 20 gb

OS: Windows 98

Processor: INTEL

Pentium Mobile 1.60 gHz

RAM: 512 mb

HD: 60 gb

OS: Windows XP Professional

Application: Apache Tomcat 5.5

PC #2

PC #3 PC #4

Lab with 04 PCs

Lab with 02 PCs

Figure 2: Simulation scenario #01 for the WebTeste per-

forming: Homogeneous Laboratory.

Web Server

Ethernet (10/100 Mbps)

Processor: INTEL Celeron

Mobile 1.4 gHz

RAM: 256 mb

HD: 40 gb

OS: Windows XP Home

Processor: INTEL

Pentium Mobile 1.60 gHz

RAM: 512 mb

HD: 60 gb

OS: Windows XP Professional

Application: Apache Tomcat 5.5

PC #3

PC #2 PC #4

Lab with 04 PCs

Lab with 02 PCs

Processor: INTEL Pentium

Mobile 1.7 gHz

RAM: 512 mb

HD: 60 gb

OS: Windows XP Home

Processor: INTEL Pentium 4

1.5 gHz

RAM: 512 mb

HD: 10 gb

OS: Windows 2000 Professional

Processor: INTEL Pentium

Mobile 1.6 gHz

RAM: 512 mb

HD: 80 gb

OS: Windows XP Professional

Lab with 01 PC

PC #1

Figure 3: Simulation scenario #02 for the WebTeste per-

forming: Heterogeneous Laboratory.

The web application used to perform the tests con-

sists of a computer system that virtually simulates a

computer store. In this system, none kind of database

persistence is used, only in memory.

3.2 Results

After the simulations, the results of the centralized

and distributed tests have been obtained. It is impor-

tant to (re)comment, that in the centralized test only

one computer has been used to perform the supervi-

sor’s function. On the other hand, in the distributed

test, two and four computers have been used as super-

visors.

Figure 4 depicts a comparative chart of the averages

response times (ms) obtained from the WebTeste sim-

ulations taking into account the scenarios #01 and #02

for the test plan composed of two links. The arrow in

the Figure 4 indicates that has occurred some prob-

lem during the performance of the last set of threads

(i.e. 10000) in the homogeneous laboratory with 02

computers. This problem can be related to the traf-

ﬁc ﬂow of the used LAN and/or to the low computer

conﬁguration (i.e. AMD Duron 750 processor, RAM

128mb).

0 2000 4000 6000 8000 10000

0,0

5,0x10

1,0x10

1,5x10

2,0x10

2,5x10

Average of Response Time (ms)

Number of Threads

01 PC, 02 Links

02 PC, 02 Links - Homogeneous Lab

02 PC, 02 Links - Heterogeneous Lab

04 PC, 02 Links - Homogeneous Lab

04 PC, 02 Links - Heterogeneous Lab

Figure 4: Comparative Chart with respect to Averages of

Response Times for the WebTeste simulation scenario #01

and #02 for the test plan composed of two links.

As can be observed, the distributed test has been

more efﬁcient than the centralized test, due to while

the threads number started in the centralized test for

each client (i.e. supervisor computer) increased, the

response time also considerably increased. When the

test has been performed in a distribute way, it has been

noticed that the same request numbers had a lower

response time, because of the fact of each supervisor

computer starts a lower thread number.

Moreover, another interesting analysis can be car-

ried out ﬁxing the number of computers. For instance,

Figures 5 and 6 respectively show the comparative

charts of the average of response time (ms) obtained

from the WebTeste simulations considering the cen-

tralized tests and the distributed tests composed of

four computers. Two problems are indicated by an

arrow in the Figures 5 and 6:

• The ﬁrst one refers to the arrow in the Figure 5

WEBIST 2006 - INTERNET TECHNOLOGY

248

that indicates the occurrence of some problem dur-

ing the performance of the last set of threads (i.e.

10000) in the centralized test for 10 links. This

problem is only related to the trafﬁc ﬂow of the

used LAN, once the computer conﬁguration used

(i.e. Intel Pentium Mobile 1.7gHz, RAM 512mb)

is upper than the computer conﬁgurations used in

the homogeneous laboratory (i.e. AMD Duron 750

processor, RAM 128mb);

• The second one refers to the arrow in the Figure 6

that indicates the occurrence of some problem dur-

ing the performance of the last three set of threads

(i.e. 9000, 9500 and 10000) in the distributed test

for 04 computers. This problem is not only related

to the trafﬁc ﬂow of the used LAN, but also to the

low conﬁguration of a computer used in the het-

erogeneous laboratory (i.e. Intel Celeron Mobile

1.4gHz, RAM 256mb). This situation indicates that

a low computer conﬁguration in a heterogeneous

scenario can reduce the performance of the stress

test.

0 2000 4000 6000 8000 10000

-5,00E+007

0,00E+000

5,00E+007

1,00E+008

1,50E+008

2,00E+008

2,50E+008

3,00E+008

3,50E+008

4,00E+008

4,50E+008

5,00E+008

5,50E+008

6,00E+008

Average of Response Time (ms)

Number of Threads

01 PC, 02 Links

01 PC, 10 Links

01 PC, 20 Links

Figure 5: Comparative Chart with respect to Averages of

Response Times for the WebTeste simulation based on the

test plan composed of one computer and two, ten and twenty

links.

From the obtained results, it has been evident that

the thread numbers started in the tester computer can

have an inﬂuence on the response time of the web en-

vironment.

4 CONCLUDING REMARKS

The current version of WebTeste performs ordered

and distributed tests and takes into account the fault-

tolerance processing, the facility of the test plan de-

sign, and the monitoring and analysis of results.

The obtained results from the WebTeste tool ap-

plication have been interesting. WebTeste has pre-

0 2000 4000 6000 8000 10000

-2,00E+007

0,00E+000

2,00E+007

4,00E+007

6,00E+007

8,00E+007

1,00E+008

1,20E+008

1,40E+008

1,60E+008

1,80E+008

2,00E+008

2,20E+008

2,40E+008

2,60E+008

2,80E+008

Average of Response Time (ms)

Number of Threads

DTest 04 PCs 02 Links

DTest 04 PCs 10 Links

DTest 04 PCs 20 Links

Figure 6: Comparative Chart with respect to Averages of

Response Times for the WebTeste simulation scenario #02

for the test plan composed of four computers and two, ten

and twenty links.

sented efﬁciency during the distributed and synchro-

nized tests, due to the signiﬁcant reduction of the av-

erage of response times (Figures 4, 5 and 6).

The stress test using the WebTeste throughout the

design and implementation of web applications will

can contribute with the robustness and reliability of

these applications, improving their quality.

As future research activities, it is proposed to per-

form the stress test based on WebTeste tool taking into

account access to the database in order to verify some

aspects such as the relation between the stress test tool

and the database persistence, and/or the variation of

response times.

REFERENCES

Bazzana, G. (2001). Ensuring the quality of web sites and

e-commerce applications. In Wieczorek, Martin and

Meyerhoff, Dirk, editor, Software Quality: State of the

Art in Management, Testing, and Tools, chapter 13,

pages 178–191. Springer-Verlag, Berlin.

Ezhilchelvan, P. D., Mac

edo, R. A., and Shrivastava, S. K.

(1995). Newtop: A fault-tolerant group communica-

tion protocol. In International Conference on Dis-

tributed Computing Systems, pages 296–306.

Httpunit (2004). Http unit. http://httpunit.sourceforge.net//.

Peters, J. F. and Pedrycz, W. (2000). Software Engineering:

An Engineering Approach. John Wiley & Sons, New

York.

Sommerville, I. (2000). Software Engineering. Addison-

Wesley, London.

Tomcat (2004). Apache jakarta tomcat.

http://jakarta.apache.org/tomcat/.

Trepper, C. H. (2000). E-Commerce Strategies. Microsoft

Press, Redmond.

WEBTESTE: A STRESS TEST TOOL

249