You are on page 1of 9

Software Reliability Growth Models, Tools and Data Sets-

A Review
Anurag Kumar
Amity University, Uttar Pradesh
Anurag.ahaar@gmail.com

ABSTRACT
This paper presents effective and comprehensive review on
parametric and non parametric modeling of software reliability.
1.1 Paper Identification
Reliability is one of the most important characteristic of software Our review work has the following characteristics, which help to
quality. More than hundreds of software reliability models were put it out of the crowed-
proposed in last few decades. These models were based on various
phases of software development life cycle. Software Reliability  Time Frames: - Our review covers the time frame of
Growth Models based on software testing were explored a lot over 1980-2015. The paper includes various growth models
the years. Many authors have proposed or discussed parametric published during this time frame, whether these were
models with following characteristics such as realistic and parametric models or non parametric models.
unrealistic assumptions, limitations, applicability, environment  Comprehensiveness: - The papers considered for the
dependability and predictability. On the other hand authors have review were not collected from a single source. We
discussed the shortcomings of parametric models and proposed collected the papers from various digital libraries
non-parametric models as an alternative to address the software including IEEE Explore, ACM Digital Library and
reliability issues. But, both the modeling techniques have their pros Science Direct etc.
and cons. In this paper our aim is to give an insight of parametric  Analysis: - We tried our best to provide the new comers
and non-parametric based software reliability models. The review detailed information about the growth models, data sets
of real life datasets and already proposed tools for software and tools of reliability.
reliability will help the beginner level researcher to continue.
1.2 Assessment Criteria
CCS Concepts
Our way to identifying papers suitable for analysis is spurred by
• Software and its engineering → Software organization and Kitchenham and Charter's idea of a quality check. Our evaluation
properties → Extra-functional properties → Software is engaged particularly on identifying just papers reporting
reliability adequate data to permit analysis crosswise over studies. To permit
this, an essential arrangement of data must be accounted for in
Keywords papers. Without this it is hard to legitimately comprehend what has
Software reliability growth models; tools for software reliability; been done in a study and just as hard to satisfactorily contextualize
parametric technique; non parametric technique. the discoveries.
Our criteria for inclusion of papers is as follows-
1. INTRODUCTION  It must be a new growth model.
Computers are used in a variety of significant and multifaceted  It should have sufficient contextual and model building
areas including Air Traffic Analysis & Control, Nuclear Reactors, data.
Space Research, Hospitals, Schools and Colleges, Military &  It must not be a review or survey paper.
Defense services etc. The increasing dependability of various  It must not be an analysis of already published models.
fields on software creates a need to make the software field more
accurate and reliable [1]. That’s why software reliability is an issue 2. SOFTWARE RELIABILITY
to make all the depending fields work properly [2]. Software Now let’s consider what we mean by software reliability? If we
reliability is emerging as a crucial field of study. talk about the word “reliability” then we know that reliability
means “the quality of being trustworthy or steady”. So for software
The aim of the research work is to present effective and
to be reliable means, the software should be dependable or can be
comprehensive review on parametric and non parametric modeling
trusted by anyone who uses it. The particular definition of software
of software reliability. Various growth models presented by
reliability differs from researcher to researcher. But what comes as
various researchers during last few decades were analyzed by us.
an output from all the definitions proposed till now that software
reliability is the probability that the software system will
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
accomplish its intentional function under some specific limitations
not made or distributed for profit or commercial advantage and that [3]. Software reliability is directly connected to software failure
copies bear this notice and the full citation on the first page. To copy because if software reliability is its correct functioning then
otherwise, or republish, to post on servers or to redistribute to lists, software is said to be unreliable if it’s not functioning as it’s
requires prior specific permission and/or a fee. Request permissions supposed to do or we can say software failure occurred. Let’s take
from Permissions@acm.org. a look at software error, software faults and software failure. An
ISEC '16, February 18-20, 2016, Goa, India error is a mental inaccuracy made by the coder. A fault is the
© 2016 ACM. ISBN 978-1-4503-4018-2/16/02…$15.00
exemplification of that error in the coding of the software while a
DOI: http://dx.doi.org/10.1145/2856636.2856648

80
software failure is defined as the happening of an erroneous output
as an outcome of an input value [4].
4. PARAMETRIC SOFTWARE
Figure 1: Software Faults vs. Failure RELIABILITY GROWTH MODELS
There are two categories of SRGMs; first one is parametric
reliability growth models. The concept of parametric growth
models was the first try to predict the reliability of software
products. These SRGMs are supported by a variety of assumptions
made by the researchers. The methodology used in these models is
based on statistical approaches like regression. During the last
some decades, researchers had proposed so many parametric
growth models. Here we are presenting some PSRGMs selected
from literature:
Bev Littlewood et al. in 1981 present a modification [6] to the
originally and widely used first growth model of reliability given
Various studies in the field of software reliability, has proved that by Jelinski & Moranda [7]. Authors used MLE for parameter
as the size of software system increases the number of faults made estimation to overcome instability problem of the J&M model.
by coder increases, which causes the increased chances of software Yamada et al. in 1983 proposed and analyzed the model of error
failure or we can say less reliability [5]. detection and found the curve of number of errors detected in the
analyzed data is S-shaped, that’s why it’s called s-shaped growth
model [8, 22]. Yoshihiro Tohma et al. in March 1989 published a
3. SOFTWARE RELIABILITY GROWTH structural approach to find out the residual software faults [9].
Application of both the models was shown and they proved that
MODELS the fitness of models to the real data is good enough. George L.
Software reliability is a significant part of software industry. It Gentzler et al. in February 1990, presented data stability to be an
gives a measure to the customer as well as to the developer about application of software reliability models [10]. Sarah Brocklehurst
the faults in the software. So the prediction of the reliability of any et al. in 1990 proposed the concept of recalibration of software
software in this competitive software industry is really essential. reliability growth models [11]. They chose the problem of deciding
Reliability growth models give an estimate to the number of faults that which model will give the better result of reliability priory.
that may occur in near future after the delivery of the product and The advantage of the recalibration was its non dependability on the
thus the models also provide an indication of when to release the model used. Karama Kanoun et al. in 1991 published a research to
software. Software Reliability Growth Models (SRGMs) use the use the already proposed growth models to get better prediction of
past data gathered during the testing process. This data is about the errors or failure and Yoshihiro Tohma et al. gave some new
number of faults occurred over time during testing of the software. parameter estimation techniques [12, 13].
This fault data is used to apply a SRGM to estimate the reliability The growth model given by P.K. Kapur et al. in 1992 [14] was
of the software product. using the process of NHP distribution. The assumption made by
The majority of software reliability growth models have a the authors was that the errors using this process may lead us to
parameter that relates to the total number of defects present in the find out some undetected errors without causing any software
set of code. If this parameter and current number of defects failure. Yamada et al. presented a growth model considering the
discovered is known, then remaining defects in the code can be test effort needed during the period of testing [15]. They used
predicted. Weibull Curve to represent the relationship between test efforts to
the time. The methodology of the model was NHPP [16]. P.
Knowing the number of residual defects helps to make your mind Zeephongsekul, G. Xia, and S. Kumar in 1994 proposed their
up whether or not the code is ready to dispatch and how much reliability growth model [17]. This model was based on the
more testing is necessary if the code is not ready to dispatch. It concept that the flawed resolution of current faults in the system
gives an estimate of the number of failures that the customers will may lead to the new faults and secondary failure of the software
encounter when operating the software. Two types of data is system in near future. In this paper they also presented a cost
required to apply any SRGM- one is the defect data and second model. Jeff Tian in 1995 tried to combine the already proposed
one is the time of the occurrence of the defects in the defect data. approaches, time domain approach and input domain approach, of
reliability growth to make his own new approach [18]. This growth
Figure 2: Residual Defects
model given by Tian had the strengths of both the approaches and
he proved its functionality on software developed by IBM
Software Solutions Toronto Laboratory. A. Pasquini et al. in 1996
published their investigation about the sensitivity of SRGMs to the
faults and also proposed a relationship between this sensitivity and
the testing phase [19]. This investigation was based on a scenario
in which various models of reliability growth were implemented
on the data of testing stage.
In December 1996, Hou et al. used HGDM growth model to give
the solution of two major problems of testing phase, one of them
was to minimize the number of undetected software faults after
testing and other one was to minimize the number of resources

81
needed for the testing and found that optimal resource allocation covered all the pros of conventional NHPP but also provide
method is capable of improving the reliability of the software coverage to other components as well. Huang, Chin-Yu, and Wei-
system under consideration [20]. In 1996 Hou et al. published one Chih Huang in 2008 put the concept in which the assumption of
more paper related to HGDM (Hyper-Geometric Distribution ignoring the debugging time was challenged by the authors [35].
Software Reliability Model) [21]. Authors proposed two new They used the queuing models for predicting the reliability of the
sampling techniques to find out the Bayes estimates of current software systems. In the end authors concluded that considering
reliability as well as current fault number. Tian, Jeff, and Joe the fault detection and correction during the testing process will
Palma in 1998 proposed their new approach which they called tree give the more accuracy to the growth models. Norman
based approach by combining the plus points of software reliability Schneidewind in 2008 published a paper in which he compared
growth models and input growth models [23]. Authors various models and tried to modify some [36]. Yamada model was
implemented their approach on five large software systems taking found most accurate in conclusion. Kapil Sharma et al. chose a
them as examples and got better results for their approach. problem of selecting the appropriate model to test the reliability
[37]. The number of reliability growth models were increasing at a
Chen et al. in 2001 took the problem of over estimation of pace of airplane so in 2010 they gave a distance based model for
reliability growth models because of nature of testing process and selecting the appropriate growth model for the reliability testing.
proposed a model to overcome this limitation of growth models This model takes some input as selection criteria and gives an
using code coverage [24]. In September, 2001 Kuo et al. tried to appropriate model on the basis of the input parameters. In 2011
propose one more growth model using NHPP [25]. Pasquini, Huang et al. said that the debugging process during testing and
Alberto, Giuliano Pistolesi, and Antonio Rizzo in 2001 gave new operation differs in speed and number of faults so they gave the
theory about the failure of software systems that faults are not only model for both the duration of software life: testing and operation
cause of component failure but also as a whole of the system and [38]. Then they proposed the concept of multi points that were the
also because of the human errors [26]. So the authors tried to points on time graph when environment changes. Other parametric
propose a model which extends the already implemented methods models [39, 40, 41, 42] were also new findings in the parametric
to find the growth of other system components such as human growth models’ field.
interface.
Table 1 is showing the summary of all the papers considered by us
Lyu et al. in 2002 considered the CBSE approach of software in parametric category-
development and gave model for the reliability of the software
integrated from various components [27]. This was the model for Table 1: Various PSRGMs
the component based software development techniques to optimize
Reference Year of Proposed Methodology
the reliability constraints and testing schedules. Teng, Xiaolin, and
Publication And Outcomes
Hoang Pham in 2002 proposed first SRGM for the N-Version Number
Programming System which was built by considering error
introduction rate and also considered error removal efficiency [28]. [6] 1981 Concept of MLE introduced
for the conventional reliability
Yashwant K. Malaiya, et al. in December, 2002 first time proposed model.
growth model considering testing coverage [29]. Authors
implemented relationship between test coverage and defect [8] 1983 Proposed s-shaped reliability
coverage. Test-coverage parameters were branches and blocks etc. growth model
Huang et al. in 2003 proposed the paper [30] in which they showed [9] 1989 Structural approach to find out
various NHPP based models using weighted mean, geometric the residual software faults
mean and harmonic mean and also gave model of NHPP by
implementing power law transformation. Huang, Chin-Yu, and [10] 1990 Data stability to be an
Chu-Ti Lin in 2007 published a paper in which they said that the application of software
assumption made in various models proposed till now is not well reliability models. Used Goel-
applicable in practical implementations [31]. The assumption Okumoto growth model.
challenged was that the faults detected are removed immediately [11] 1990 concept of recalibration of
but authors put the concept that it’s not practically possible to software reliability growth
remove all the detected faults immediately, especially dependent models
faults (the faults which are linked to each other), because of
[12] 1991 They gave the concept of
constraints related to cost and debugging team.
analyzing and then dividing
Huang, Chin-Yu, et al. in 2007 again presented a research work in trend of the testing data to
which they compared various SRGMs which were sensible to the apply multiple different
effort needed to resolve the faults [32]. The problem the chose was growth models on the parts of
that some S-Shaped models may not work well if the test-effort data.
was fluctuating, so they introduced a new function. [13] 1991 Analyzed 6 methods and
Zachariah, Babu, and R. N. Rattihalli in year 2007 proposed two proved LSS parameter
new growth models [33]. The attempt made by the authors was estimation is best for hyper
based on the hypothesis that the failure rate of the software during geometric distribution
the testing phase is because of the inappropriate selection of the [14] 1992 New growth model based on
input patterns. Wang et al. proposed a new version of NHPP called non-homogeneous Poisson
MANHPP abbreviation for moving average NHPP to include the distribution
advantages of both the approaches time domain approaches and
structure based one [34]. This MANHPP approach not only

82
[15] 1993 Growth model based on and they used delay function
NHPP, presented relation of in SRGM to overcome this
test effort to the time using problem.
weibull curve
[17] 1994 Improper fault resolution may
result in secondary software [32] 2007 They analyzed the SWGMs
failure in near future. They which were sensitive to the
proved this. test effort and introduced the
logistic function.
[18] 1995 New model by the
combination of time domain [33] 2007 Two growth models using
and input domain approaches. random testing and partition
testing.
[19] 1996 Proved the sensitivity of
growth models to the errors [34] 2007 Moving Average was
based on a considered case implemented with
study. conventional NHPP

[20] 1996 Used HGDM to solve [35] 2008 Proposed a model by


OPT/RA. considering the debugging
time and effort using queuing
[21] 1996 Proposed a release policy for models.
HGDM to minimize cost and
meet reliability deadline. [36] 2008 It was the comparison paper
in which 3 reliability models
[22] 1997 Implemented two Gibb’s were compared.
approaches for sampling.
Used NHPP [37] 2010 Proposed a model selection
method based on distance.
[23] 1998 Combined two approaches to
make their own tree base [38] 2011 Presented concept of multi
growth model points in the software life
cycle to improve reliability
[24] 2001 Solved the problem of over growth prediction in testing
estimation of the growth and operation phase
models.
[39] 2012 Proposed SRGM for multi
[25] 2001 Based on methods for release software products.
parametric decomposition,
proposed one more growth [40] 2013 Modified NHPP technique
model based model considering
imperfect fault debugging.
[26] 2001 Model for growth of other
system components as man- [41] 2014 A model to predict the release
machine interfaces and of the open source software
operators. based on generalized software
reliability model.
[27] 2002 Reliability growth model for
CBSE based software [42] 2015 Growth model based on
systems. Bayesian theory but different
2002 First growth model for NVP from previous models in the
[28] manner that it uses all
which considers error removal
efficiency & error- available data rather than only
introduction rate. failure data.

[29] 2002 first time proposed growth


model considering testing
coverage
5. Non Parametric Software Reliability
Growth Models
[30] 2003 Used harmonic, arithmetic Parametric SRGMs models utilize the statistical approaches for
and geometric means to predicting the reliability growth. Major pitfalls of these models are
implement NHPP models. the assumptions made, because these assumptions are based on the
[31] 2007 Proved that detection and past behavior of the software and may not relate to the real
correction processes during problem. That’s why over the years many researchers proposed
testing can’t run non parametric software reliability growth models. NPSRGMs
simultaneously because generally use the machine learning approaches like neural
debugging process takes time networks, genetic algorithm, support vector machines etc. The

83
advantage of these models over parametric based ones is that, there [47] 2007 GP was used for this model and
is no need of assumptions and the self learning behavior of the boosting is used to improve the
machine learning approaches makes these models more accurate. efficiency of this model.
Ariela Sofer et al. in 1991 presented a new growth model of
[48] 2007 NHPP using nonparametric method to
reliability by modifying their previous model [43]. The new model
overcome the problem of wrong
was modified to estimate not only current but also long-term
assumptions.
predictions. Karunanithi et al. in 1992 proposed one more efficient
model for reliability growth called connectionist model of software [49] 2007 NPSRGM based on order statistic
reliability growth [44]. They used neural network for this model approach and authors solved the
and also presented comparison of this model with other five problem of random sampling of faults
models and the outcome was the better accuracy of the 2009 Authors tried to overcome the pitfalls
[50]
connectionist model. of canonical genetic programming.
In September, 1999 Sitte and Renate published a paper in which These problems were random
they compared already published two great approaches of generation and slow convergence for
reliability estimation, which were neural network based and large data sets.
parameter recalibration based [45]. Both the approaches so far was [51] 2009 Author used meta heuristic approach
proved better to each other but Sitte & Renate find out that, neural to predict the reliability of the
networks are more accurate. Cai, Kai-Yuan et al. in 2001 used complex data sets of software. ACO
neural network once again but this time for the prediction of was used to deal with the constraints
dynamic reliability of the products [46]. Authors presented their of the problem.
new findings on the behavior of the neural network for the
prediction of the software reliability. Costa, Eduardo Oliveira, et [52] 2010 Author used hybrid GA and simulated
al. in September 2007 published this paper, in which they proposed annealing to propose thus model
a growth model based on the genetic algorithm or programming which was based on SVM. This
[47]. Also authors showed the use of boosting technique to perk up model also called as GA-SA-SVM
the effectiveness of the models. Wang, Zhiguo, et al. [48] and model. Use of SA with GA deals with
Wilson et al. [49] in 2007 proposed non parametric NHPP and the convergence problem of GA and
analyzed order statistic model by means of non parametric SVM helps to deal with non linear
technique. Not only neural networks [56, 57] but also genetic data.
algorithm and SVM based models [50, 51, 52, 53, 54, 59] were [53] 2011 Authors integrated ARIMA of Box
proposed by various researchers over the years. Wason et al. [55] and Jenkin and SVM to create new
in 2012 tried new approach for NPSRGMs. They used finite state model because ARIMA deals with
automata first time in the history for reliability growth model. linear data and SVM deals with non
Table 2 is showing some non parametric models proposed over the linear one.
decades. This table only has the models chosen by us and no
[54] 2012 Li et al. published this short paper in
modification and optimization attempts of authors are included in
which they used GA based GEP for
this table. the prediction of growth of reliability.
Table 2: Various NPSRGMs [55] 2012 This was the first of its kind attempt
made by the authors. They used first
Reference Year Methodology Used and Outcome
time finite state automata to represent
Number the software system.

1991 Ariela Sofer et al. previously [60] 2013 In this paper authors used GA and
[43]
proposed parametric model but this simulated annealing with the aim of
time they used non parametric maximizing reliability prediction and
strategy to predict the long term minimizing cost.
faults.
[44] 1992 Authors used neural network to make
their model called connectionist 6. YEAR WISE COMPARISON OF
model and published the results that
this is better than five parametric PSRGMS AND NPSRGMS
models. Here we are providing the chart of various models of both the
categories over the years. This is the plot between number of
[45] 1999 This paper proved that neural models considered (both type) and years. Figure 3 showing this
networks work better than parametric plot. X-axis showing the number of models considered by us from
models. both the categories and y-axis is showing the years in the range of
[46] 2001 Major findings: 10 years.
 Approach is better for
smooth trend
 Training output is better
than prediction.

84
Figure 3: Number of Models vs. Years  Data Modification
 Failure Data Analysis
15  Modeling and Measurement
S-PLUS S-PLUS is one of the tools used for
10 growth modeling based on tree,
Parametric tracking of test cases and SRGMs.
5 Models Some models can easily be
simulated on S-PLUS but not all
0 Non Parametric because of the lots of options
Models required for the models like,
parameter estimation methods etc,
which require lots of effort to use
this tool.
Weibull++ Weibull++ is the product of
ReliaSoft. It is the standard tool for
Weibull analysis for many industries
7. TOOLS USED IN SOFTWARE across the globe. This tool supports
RELIABILITY GROWTH MODELS an all-inclusive array of life span
There are so many tools which are used in software reliability distributions and tenders an
estimation. Which tool would be more efficient, is normal question influential interface especially
which struck in the mind of any researcher, especially for new designed for the need of research
comers in the field. So I have collected the information of some fellows.
research tools to help the novice researchers-
Table 3: Tools used in Reliability Estimation 8. DATA SETS USED IN SOFTWARE
Tools Overview RELIABILITY TESTING
Pham [3] stated that there are two frequently used types of failure
ESTM (Economic Stop It supports multiple models
data:
Testing Model Tool) concurrently and uses MLE
 Data based on Time-domain and
parameter estimation method. But
ESTM only uses time between data  Data based on Interval-domain.
format only. These types of data sets are generally utilized by the experts when
investigating and anticipating reliability. Not all the quality models
ROBUST ROBUST provide the researchers an can deal with both sorts of data.
(Reliability of Basic and easy way of applying earlier growth The time-space methodology includes recording the times when
Ultra-reliable Software models for examination of the fault detected, as showed in Table number 4. First fault occurred at
sysTem) reliability of the software. It 20 minute, 2nd at 32, 3rd at 42 and so on. A few models may have
provides some features to enhance need of the time data between various failures to compensate for
the SRGMs also and can be used for the actual time of failure. In the example considered, 3rd column
statistical methods of early error data 20, 12, 10, 16, 30, 2, 21, 3, 6 and 7 should be used as the time-
detection. domain data set.
SMERFS SMERFS is a tool used for software Table 4
(Statistical Modeling and reliability modeling that enables the Failure Record Failure Time Time Between
Estimation of Software user to execute an analysis of Number (in Min.) Failures (in Min.)
Reliability Functions) reliability. This program has gone
1 20 20
through a number of modifications
over time. The currently used 2 32 12
version of this is 5. It has included 3 42 10
11 of the mostly used models of 4 58 16
literature; 6 using time between error 5 88 30
occurrences as input and 5 using the 6 90 2
number of detected faults per unit of
7 111 21
time.
8 114 3
SoRel [62] Two major modules of SoRel are: 9 120 6
(Software Reliability  Trend Tests 10 127 7
Program)  Reliability Growth Models.
It operates on 2 types of data- Second approach is defined by counting the failures happening
intensity of faults and times of inner during a period of time (e.g. minute, day or a testing session). This
faults. approach is shown in the table number 5. In the table, we recorded
CASRE [61] CASRE is a CASE tool used in six failures in the first 1-hour interval, four failures in the 2 nd
reliability estimation. Main features interval and eight failures in the 3rd and so on.
of this tool are:

85
Table 5 [9] Tohma, Yoshihiro, et al. "Structural approach to the
Time Period (hours) Number of Faults Detected estimation of the number of residual software faults
based on the hyper-geometric distribution." Software
1 6
Engineering, IEEE Transactions on 15.3 (1989): 345-
2 4
355.
3 8 [10] Gentzler Jr, George L., and Nelson M. Andrews. "Data
4 3 stability in an application of a software reliability
model." Selected Areas in Communications, IEEE
The time-domain methodology always offers higher accuracy in Journal on 8.2 (1990): 273-275.
the methods of parameter estimates with current tools but absorbs [11] Brocklehurst, Sarah, et al. "Recalibrating software
more data collection pains than the first one. reliability models." Software Engineering, IEEE
Transactions on 16.4 (1990): 458-470.
9. CONCLUSION AND IMPLICATIONS [12] Kanoun, Karama, Marta Rettelbusch de Bastos Martini,
FOR FUTURE WORK and Jorge Moreira De Souza. "A method for software
Reliability is one of the most imperative characteristics of reliability analysis and prediction application to the
software. More than hundreds of software reliability models were TROPICO-R switching system." Software Engineering,
proposed in last few decades. The aim of the study was to provide IEEE Transactions on 17.4 (1991): 334-344.
the new comers and novice researchers a thorough guide on the [13] Tohma, Yoshihiro, et al. "The estimation of parameters
reliability growth models. We reviewed almost 60 models of of the hyper geometric distribution and its application to
collectively of both type- parametric and non parametric the software reliability growth model." IEEE
techniques. Our research shows the techniques used by researchers Transactions on Software Engineering 5 (1991): 483-
till now to design the growth models and the pros and cons of the 489.
methodologies used by them. Review of parametric and non [14] Kapur, P. K., and R. B. Garg. "A software reliability
parametric models revealed the limitations of both the models. The growth model for an error-removal phenomenon."
limitation of parametric models is that these models use so many Software Engineering Journal 7.4 (1992): 291-294.
assumptions made by researchers and these assumptions vary from [15] Yamada, Shigeru, Jun Hishitani, and Shunji Osaki.
researcher to researcher. If we talk about non parametric models "Software-reliability growth with a Weibull test-effort: a
which use the concept of machine learning approaches take much model and application." Reliability, IEEE Transactions
time in learning and require lots of training data to train the ML on 42.1 (1993): 100-106.
model and also the training data is key factor of accuracy of non [16] As H. Ascher, H. Feingold, Repairable System
parametric models. These limitations of the models can be Reliability: Modeling, Inference, Misconceptions, and
removed in near future. Also we provided the data sets and tools Their Causes, 1984; Marcel Dekker.
which can be used in reliability estimation, this information of [17] Zeephongsekul, P., G. Xia, and S. Kumar. "Software-
tools and data sets are very crucial for the new players in the field. reliability growth model: primary-failures generate
secondary-faults under imperfect debugging."
Reliability, IEEE Transactions on 43.3 (1994): 408-413.
10. REFERENCES [18] Tian, Jeff. "Integrating time domain and input domain
analyses of software reliability using tree-based models."
[1] System Software Reliability by Prof. Hoang Pham
Software Engineering, IEEE Transactions on 21.12
Department of Industrial Engineering Rutgers the State (1995): 945-958.
University of New Jersey, USA. [19] Pasquini, A., A. N. Crespo, and P. Matrella. "Sensitivity
[2] Pai, Ganesh J. "A survey of software reliability models." of reliability-growth models to operational profile errors
preprint: 1304.4539 (2013). vs. testing accuracy [software testing]." Reliability, IEEE
[3] Pham, Hoang. "Springer Series in Reliability Transactions on 45.4 (1996): 531-540.
Engineering." New Jersey, USA (2005). [20] Hou, Rong-Huei, Sy-Yen Kuo, and Yi-Ping Chang.
[4] Oman, Paul, and Jack Hagemeister. "Metrics for "Needed resources for software module test, using the
assessing a software system's maintainability." Software hyper-geometric software reliability growth model."
Maintenance, 1992. Proceerdings, Conference on. IEEE, Reliability, IEEE Transactions on 45.4 (1996): 541-549.
1992. [21] Hou, Rong-Huei, Sy-Yen Kuo, and Yi-Ping Chang.
[5] Pham, Hoang. Software reliability. John Wiley & Sons, "Optimal release policy for hyper-geometric distribution
Inc., 1999. software-reliability growth model." Reliability, IEEE
[6] Littlewood, Bev, and John L. Verrall. "Likelihood Transactions on 45.4 (1996): 646-651.
function of a debugging model for computer software [22] Kuo, Lynn, et al. "Bayes inference for S-shaped
reliability." Reliability, IEEE Transactions on 30.2 software-reliability growth models." Reliability, IEEE
(1981): 145-148 Transactions on 46.1 (1997): 76-80.
[7] Z. Jelinski, P.B. Moranda, "Software reliability [23] Tian, Jeff, and Joe Palma. "Analyzing and improving
research", in Statistical Computer Performance reliability: A Tree-based Approach." IEEE software 2
Evaluation, ed. W. Freiberger, New York and London: (1998): 97-104.
Academic Press, 1972, pp 465-484. [24] Chen, Mei-Hwa, Michael R. Lyu, and W. Eric Wong.
[8] Yamada, Shigeru, Mitsuru Ohba, and Shunji Osaki. "S- "Effect of code coverage on software reliability
shaped reliability growth modeling for software error measurement." Reliability, IEEE Transactions on 50.2
detection." Reliability, IEEE Transactions on 32.5 (2001): 165-170.
(1983): 475-484.

86
[25] Kuo, Sy-Yen, Chin-Yu Huang, and Michael R. Lyu. [41] Washizaki, Hironori, Kiyoshi Honda, and Yoshiaki
"Framework for modeling software reliability, using Fukazawa. "Predicting Release Time for Open Source
various testing-efforts and fault-detection rates." Software based on the Generalized Software Reliability
Reliability, IEEE Transactions on 50.3 (2001): 310-320. Model."
[26] Pasquini, Alberto, Giuliano Pistolesi, and Antonio Rizzo. [42] Wayne, Martin, and Mohammad Modarres. "A Bayesian
"Reliability analysis of systems based on software and Model for Complex System Reliability Growth under
human resources." Reliability, IEEE Transactions on Arbitrary Corrective Actions." Reliability, IEEE
50.4 (2001): 337-345. Transactions on 64.1 (2015): 206-220.
[27] Lyu, Michael R., Sampath Rangarajan, and Aad Van [43] Sofer, Ariela, and Douglas R. Miller. "A nonparametric
Moorsel. "Optimal allocation of test resources for software-reliability growth model." Reliability, IEEE
software reliability growth modeling in software Transactions on 40.3 (1991): 329-337.
development." Reliability, IEEE Transactions on 51.2 [44] Karunanithi, Nachimuthu, Darrell Whitley, and
(2002): 183-192. Yashwant K. Malaiya. "Prediction of software reliability
[28] Teng, Xiaolin, and Hoang Pham. "A software-reliability using connectionist models." Software Engineering,
growth model for N-version programming systems." IEEE Transactions on 18.7 (1992): 563-574.
Reliability, IEEE Transactions on 51.3 (2002): 311-321. [45] Sitte, Renate. "Comparison of software-reliability-
[29] Malaiya, Yashwant K., et al. "Software reliability growth growth predictions: neural networks vs parametric-
with test coverage." Reliability, IEEE Transactions on recalibration." Reliability, IEEE Transactions on 48.3
51.4 (2002): 420-426. (1999): 285-291.
[30] Huang, Chin-Yu, Michael R. Lyu, and Sy-Yen Kuo. "A [46] Cai, Kai-Yuan, et al. "On the neural network approach in
unified scheme of some nonhomogenous poisson process software reliability modeling." Journal of Systems and
models for software reliability estimation." Software Software 58.1 (2001): 47-62.
Engineering, IEEE Transactions on 29.3 (2003): 261- [47] Costa, Eduardo Oliveira, et al. "Exploring genetic
269. programming and boosting techniques to model software
[31] Huang, Chin-Yu, and Chu-Ti Lin. "Software reliability reliability." Reliability, IEEE Transactions on 56.3
analysis by considering fault dependency and debugging (2007): 422-434.
time lag." Reliability, IEEE Transactions on 55.3 (2006): [48] Wang, Zhiguo, Jinde Wang, and Xue Liang. "Non-
436-450. parametric estimation for NHPP software reliability
[32] Huang, Chin-Yu, Sy-Yen Kuo, and Michael R. Lyu. "An models." Journal of Applied Statistics 34.1 (2007): 107-
assessment of testing-effort dependent software 119.
reliability growth models." Reliability, IEEE [49] Wilson, Simon P., and Francisco J. Samaniego.
Transactions on 56.2 (2007): 198-211. "Nonparametric analysis of the order-statistic model in
[33] Zachariah, Babu, and R. N. Rattihalli. "Failure size software reliability." Software Engineering, IEEE
proportional models and an analysis of failure detection Transactions on 33.3 (2007): 198-208.
abilities of software testing strategies." Reliability, IEEE [50] Zhang, Yongqiang, and Huifang Cheng. "Improved
Transactions on 56.2 (2007): 246-253. Genetic Programming Algorithm Applied to Symbolic
[34] Wang, Wen-Li, Thomas L. Hemminger, and Mei-Huei Regression and Software Reliability Modeling." Journal
Tang. "A moving average non-homogeneous Poisson of Software Engineering and Applications 2.05 (2009):
process reliability growth model to account for software 354.
with repair and system structures." Reliability, IEEE [51] Ayyoub, Belal, and Asim El-Sheikh. "A Model for
Transactions on 56.3 (2007): 411-421. System Reliability Optimization Problems Based on Ant
[35] Huang, Chin-Yu, and Wei-Chih Huang. "Software colony Using Index of Criticality Constrain." ICIT 2009
reliability analysis and measurement using finite and Conference-Bioinformatics and Image Volume:
infinite server queueing models." Reliability, IEEE Bioinformatics and Image. 2009.
Transactions on 57.1 (2008): 192-203. [52] Jin, C. "Software reliability prediction based on support
[36] Schneidewind, Norman. "Comparison of reliability and vector regression using a hybrid genetic algorithm and
testing models." Reliability, IEEE Transactions on 57.4 simulated annealing algorithm." IET software 5.4 (2011):
(2008): 607-615. 398-405.
[37] Sharma, Kapil, et al. "Selection of optimal software [53] Lo, Jung-Hua. "A study of applying ARIMA and SVM
reliability growth models using a distance based model to software reliability prediction." Uncertainty
approach." Reliability, IEEE Transactions on 59.2 Reasoning and Knowledge Engineering (URKE), 2011
(2010): 266-276. International Conference on. Vol. 1. IEEE, 2011.
[38] Huang, Chin-Yu, and Michael R. Lyu. "Estimation and [54] LI, JUNG-SHIAN, and MING-WANG GUO. "A Non-
analysis of some generalized multiple change-point Parametric Software Reliability Modeling Approach by
software reliability models." Reliability, IEEE Using Gene Expression Programming " Journal of
Transactions on 60.2 (2011): 498-514. Information Science and Engineering 21 (2005): 1201-
[39] Kapur, P. K., et al. "Two dimensional multi-release 1211.
software reliability modeling and optimal release [55] Wason, Ritika, P. Ahmed, and M. Qasim Rafiq.
planning." Reliability, IEEE Transactions on 61.3 "Automata-Based Software Reliability Model: The Key
(2012): 758-768. to Reliable Software." International Journal of Software
[40] Yan, He. "NHPP software reliability growth model Engineering and Its Applications 7.6 (2013): 111-126.
incorporating fault detection and debugging." Software [56] N. Kiran, V. Ravi, Software reliability prediction using
Engineering and Service Science (ICSESS), 2013 4th wavelet neural networks, in: iccima, IEEE Computer
IEEE International Conference on. IEEE, 2013. Society, 2007, pp. 195–199.

87
[57] S. Zaidi, S. Danial, B. Usmani, Modeling inter-failure objective reliability problem with weight and space
time series using neural networks, in: IEEE International constraints." (2013).
Multitopic Conference, 2008, pp. 409–411. [61] Lyu, Michael R., and Allen Nikora. "CASRE: a
[58] J. Lo, The implementation of artificial neural networks computer-aided software reliability estimation
applying to software reliability modeling, in: tool." Computer-Aided Software Engineering, 1992.
Proceedings of the 21st annual international conference Proceedings., Fifth International Workshop on. IEEE,
on Chinese control and decision, IEEE Press, 2009, pp. 1992.
4385–4390. [62] Kanoun, Karama, et al. "SoRel: a tool for reliability
[59] B. Yang, X. Li, M. Xie, F. Tan, A generic data-driven growth analysis and prediction from statistical failure
software reliability model with model mining technique, data." Fault-Tolerant Computing, 1993. FTCS-23. Digest
Reliability Engineering & System Safety 95 (6) (2010) of Papers., The Twenty-Third International Symposium
671–678. on. IEEE, 1993.
[60] Arjestan, Mina Ebrahimi, and Seyed Hamidreza
Pasandideh. "Applying Meta-heuristic in a multi-

88

You might also like