SRGM with Imperfect Debugging by Genetic Algorithms

SRGM with Imperfect Debugging by Genetic Algorithms
Dr.R.SatyaPrasad O.NagaRaju Prof.R.R.LKantam
Associate.Professor Asst.Professor. Professor
Dept.of Computer science &Engineering Dept.of Computer science &Engineering Dept.of Statistics
Acharya Nagarjuna University charya Nagarjuna University Acharya Nagarjuna University
Nagarjuna Nagar-522524 Nagarjuna Nagar-522524 Nagarjuna Nagar-522524
profrsp@gmail.com onrajunrt@gmail.com
Abstract
Computer software has progressively turned out to be an
essential component in modern technologies. Penalty costs resulting from
software failures are often more considerable than software developing costs.
Debugging decreases the error content but expands the software development
costs. To improve the software quality, software reliability engineering plays
an important role in many aspects throughout the software life cycle. In this
paper, we incorporate both imperfect debugging and change-point problem into
the software reliability growth model(SRGM) based on the well-known
exponential distribution the parameter estimation is studied and the proposed
model is compared with the some existing models in the literature and is find to
be better.
Key words: - Software Reliability, NHPP, Mean value function, Genetic
Algorithms
1. Introduction:
We are witnessing our increasing dependence on software systems, as
they are becoming more and more complex, thus harder to develop and
maintain. Software systems are present in many safety – critical applications
such as power plants, health care systems, air-traffic, etc. they all require high
quality, reliabilities and safety.

67
Software reliability is the probability that software will not cause the
failure of a product for a specified period of time. This probability is a function
of the inputs, as well as a function of the existence of faults in the software.
Various NHPP SRGMs have been studied with various assumptions.
Many of the SRGMs assume that each time a failure occurs, the fault that
caused it can be immediately removed and no new faults are introduced, which
is usually called perfect debugging, Imperfect debugging models have been
proposed, with relaxation of the above assumption (hoba, 1984, pham 1993).
The other assumption of many NHPP SRGMs is that each failure occurs
independently and randomly in time according to the same distribution during
the fault detection process (Musa at. Al 1987). However in more realistic
situations, the failure distribution can be affected by many factors, such as the
running environment, testing strategy and resource allocation. Once these
factors are changed during the software testing phase, this could result in a
software failure intensity function that increase or decrease non-monotonically.
It is identified as a change point problem (Zhao, 1993). In software reliability
estimation the change point effect should be considered simultaneously, if there
is a change point exists, otherwise the estimators of the model cannot express
the factual software reliability behavior.
2. A General NHPP Model:
Let { }
( ), 0
N t t ≥ be a counting process representing the cumulative
number of software failures by time t. The N (t) process is shown to be a NHPP
with a mean value function m(t). Mean value function represent the s-expected
number of software failures by time t. Goel and Okumoto (1979) assume that
the number of software failures during non-overlapped time intervals is s-
independent and the software failure intensity ( )
t
λ is proportional to the
residual fault content. Thus m(t) can be obtained by solving the following
differential equation.

68
( )
( )
( ) ( )
dm t
t b a m t
dt
λ = = − …………………… (2.1)
where ‘a’ denotes the initial number of faults contained in a program
and b represents the fault detection rate. The solution of
( )
( ) 1 bt
m t a e−
= − …………………… (2.2)
In software reliability, the initial number of faults and the fault detection
rate are always unknown. The maximum likelihood technique can be used to
evaluate the unknown parameters.
The conditional software reliability, R(x/t), is defined as the probability
that there is no failure observed in the time period (t, t+x), given that the last
failure occurred at a time point ( )
0, 0
t t x
≥ > . Given the mean value function
m(t), the conditional software reliability can be shown as
( ) ( ) ( ) ( )
( )
1
/ exp 1
m t x m t b x
bt
R x t e a e e
− + − − +
−
= = − − ……………. (2.3)
For a more general NHPP SRGM, we can extend and modify Eq. (2.1)
as following:
( )
( )
( ) ( ) ( ) ,
dm t
t b t a t m t
dt
λ = = − ……………………. (2.4)
where a(t) is the time-dependent fault content function which includes
the initial and introduced faults in the program and b(t) is the time-dependent
fault detection rate. One can define a(t), b(t) to yield more complex or less
complex analytic solution for m(t). Various a(t), b(t) express different
assumptions of the fault detection processes (Pham et al., 1999).
3. Imperfect-software-debugging models.
Following the general NHPP model, a constant a(t) implies the perfect
debugging assumption, i.e., no new faults are introduced during the debugging

69
process. Pham (1993) introduced an NHPP SRGM that is subject to imperfect
debugging. He assumed if detected faults are removed, then there is a
possibility to introduce new faults with a constant rate β . Let a(t) be the
number of faults to be eventually detected (denoted by “a”) plus the number of
new faults introduced to the program by time t, the mean value function m(t)
can be given as the solution of the following system of differential equations.
( ) ( )
( ) ( )
( )
,
a t m t
m t
b a t m t
t t t
β
∂ ∂
∂
= − =
∂ ∂ ∂
………………… (3.1)
( )
(0) , 0 0
a a m
= =
where a is the number of faults to be eventually detected. Solving the
above equations, we can obtain the mean value function and conditional
software reliability, respectively, as follows:
( ) ( )
1
1 ,
1
bt
a
m t e β
β
− −
= −
−
……………………..(3.2)
( ) ( ) ( )
( )
1
/ exp bt
R x t m x e β
− −
= −
4. An NHPP model with change-point.
Many SRGMs suppose the fault detection rate is a constant, or a
monotonically increasing function. The failure intensity is expected to be a
continuous function of time. For examples, Goel and Okumoto (1979)
presented a G-O model subjected to a constant fault detection rate and Yamada
et al. (1983) modified the G-O model and created an increasing fault detection
rate function, which represents the debugging process with the learning
phenomenon. Both of the models were proposed with continuous failure
intensity function. As the earlier mention, the fault detection rate can be
affected by many factors such as the testing strategy and resources allocation.
During a software testing process, there is a possibility that the underlying fault
detection rate function is changed at some time moment τ called ‘change-
point’. Considering the change- point problem in software reliability models is
intended to be more close to the reality.

70
Zhao (1993) modified the Jelinski and Moranda (1972) model to estmate
the allocation of change-point and the failure intensity function. He assumed
that the observed inter-failure times follow the same distribution F at the
beginning. After τ failure are observed. The remaining items have the
distribution G. Distribution F and G are from the same parametric family.
Chang (1997) considered the change-point problems in the NHPP
SRGMs. The parameters of the NHPP with change-point models are estimated
by the weighted least square method. Let the parameter τ be the change point
that is considered unknown and is to be estimated from the data. The fault
detection rate function is defined as
( ) 1
2
, when 0 t t,
, when t. t,
b
b t
b
≤ ≤
=
>
By the assumptions, the mean value function, m(t) and the intensity
function, ( )
t
λ , can be derived as
( )
( )
( )
( )
1
1 2
1 , when 0 t ,
1 , when t. ,
b
b b t
a e
m t
a e
τ
τ τ
τ
τ
−
− − −
− ≤ ≤
=
− >
……………………… (4.1)
( )
( )
( )
1
1 2
1
2
, when 0 t ,
, when t. ,
b
b b t
ab e
dm t
t
dt ab e
τ
τ τ
τ
λ
τ
−
− − −
≤ ≤
=
>
Introducing the above mean value function, the conditional software
reliability function for any times x given t can be shown as
( ) ( ) ( )
( )
{ }
/ exp
R x t m t x m t
= − + −
( )
( )
{ }
( )
( )
{ }
( ) ( )
( )
{ }
1
1
1 2
1
1 2 1
) 1
exp ,when t t+x ,
exp ,when t +t ,
exp ,when
b t x
b t
b b t x
b t
b b t b b x
a e e
a e e x
a e e t
τ τ
τ τ τ τ
τ
τ
τ
− +
−
− − + −
−
− − − − − + −
− − ≤ ≤
= − − ≤ +
− − <
…………. (4.2)
5. Imperfect-software-debugging model with change- point.

71
To consider the NHPP SRGM that integrates imperfect debugging with
change-point problem, the following assumptions are made:
(a) When detected faults are removed at time t, it is possible to introduce
new faults with introduction rate ( )
t
β
( ) ( )
1
2
,when 0 t ,
is the change-point
, when t.>
t
β τ
β τ
β τ
≤ ≤
=
(b) The fault detection rate represented as following is a step function.
( ) 1
2
, when 0 t ,
b , when t> ,
b
b t
τ
τ
≤ ≤
=
(c) A NHPP models with the fault detection phenomenon in the software
system.
In earlier studies, the parameter, τ is considered as unknown and is to
be estimated from the collected failure data (Zhao, 1993; Hinkley, 1970;
Chang 1997). Because the testing strategy and resource allocation can be
tracked all the time during the fault detection process, it may be more
reasonable to reconsider that the change point τ is given. Therefore, we
can assume but not necessary the parameter τ as allocated in a certain
time point and is known in advance. According to these assumptions,
one can derive the new set of differential equations to obtain the new
mean value function:
( )
( ) ( ) ( )
( )
( )
( )
( )
,
m t a t m t
b t a t m t t
t t t
β
∂ ∂ ∂
= − =
∂ ∂ ∂
( ) ( )
0 , 0 0
a a m
= =
Solving the differential equations under the assumptions
(a) And (b) Yields
( ) ( ) ( )
( )
1 1
1
1
1
b b t
a
m t e
β τ τ
β
− − + −
= −
−
……………………………..(5.1)
( ) ( ) ( )
( )
{ }
/ exp
R x t m t x m t
= − + −

72
( ) ( ) ( )
( )
( ) ( ) ( ) ( )
( )
( )
1 1 1 1
2 1 2 2 2 2
1 1
1
1 1 1
2
exp , when 0<t
1
exp 1 , when t
1
b t b t x
b b t b x
a
e e
a
e e
β β
β τ β τ β
τ
β
τ
β
− − − − +
− − − − − − −
− − ≤
−
=
− − >
−
6. The Present model under study:
To consider the new model of NHPP SRGM that imperfect debugging
with change point problem, the following assumption are.
a) When defected faults are removed at time t, it is possible to introduce
new faults with introduction rate β .
b) The faults detection rate represented as a step function is given below
( ) 1,when 0
b t b t
= ≤
The unknown parameters are to be estimated from the collected failure data
(Zhao, 1993, Hinkley 1970, Chang 1997) According to these assumptions;
one can derive the new set of differential equation to obtain the new mean
value function.
( )
( ) ( )
m t
b t a m t
t
∂
= −
∂
( )
( )
( )
( ) ( )
; 0 , 0 0
m t m t
t a a m
t t
β
∂ ∂
= = =
∂ ∂
( ) ( ) 1
b t m t b
β
= + at ( ) 1
0
t b t b
= =
where b(t) is the time dependent faults detection rate running the general NHPP
model, a constant b(t) implies the perfect debugging assumption i.e. no new
faults are introduced during debugging process.
………………..(6.1)

73
Taking this mean value function we propose to suggest a new
SRGM with the help of a NHPP. It is reliability Parameter, predictive validates.
Its applicability as a SRGM can also be assessed through fitness of models.
The developed function will become more complex than the other
models. Let the faults introduction rate is a constant ( )
β during the fault
detection process, the mean value function time 0 t
≤
Eq. [m(t)] assumes that the intensity function ( )
t
λ is not a continuous
function of time except when 1
b Following the same definition of Goel and
Okumto (1979), the conditional reliability function of this developed model can
be obtained
( ) ( )
( )
{ }
/ exp ( )
R x t m t x m t
= − + −
( )
( ) ( )
( )
( )( )
( )
1 1
1
1 1
1
1
1 1
1
/ exp
1 1
a b t a b x
a b t x
a b b
a
e e a
b
R x t
a a
e e a b t
b b
β β
β
β
β β
β β
β
+ +
+ +
+
− + −
=
+ + +
…… (6.2)
The R(x/t) model can be used to the construct the problem with single type of
fault. However based on the severity that assesses the impact of the fault on the
user, software faults can be classified into varies types.
The further modified model can be applied to conduct the software
reliability estimation problem not only for the imperfect debugging and change
point case but also the multiple fault type’s problem. The only difficulty is that
more parameters need to be estimated at the same time.
7. Numerical examples and model evaluation:
To verify the proposed model that incorporates both imperfect debugging and
change-point problems, four data sets are introduced. Two of them are
collected from real software development project and the others are obtained
from simulation.

74
The first set of software failure data to be analyzed in this section is
taken from Misra (1983. The purpose of the first example is to illustrate the
process of model creation. In the data set software faults are classified into
three different types. Critical (type1), major (type2) and minor (type 3). The
total testing time and number of software failure for each week are recorded.
Pham (1993) used the same data to illustrate the results of his imperfect
debugging model based on the following setting of parameter. 0.05
β = .
According to the assumption of NHPP the remaining parameters 1
,
a b
can be estimated by solving the following likelihood function, for its
Maximum.
( )
( ) ( )
( ) ( ) ( ) ( )
( ) ( )
( )
1
1
2
1
1 1 1
1
,
!
m t m t
i j i j
i j i j
y t y t e
n
i j i j
i j i j i j
m t m t
L a b
y t y t
− − −
−
−
= = −
− −
=
−
∏∏
where ( )
i j
y t is the cumulated number of types I faults before time ( )
i j
y t and n
is the interval domain size It should be point out that a similar analysis was
conducted using the method of maximum likelihood discussed in (Pham,
1993). The estimates for parameters .200
a = 1 0.000355
b =
The results show that the fault detection rates increase for all three types
of fault after time τ table 1 present the evaluation results of the developed
model. Where mean square errors (MSE) measure weighted average of the
square of the distance between actual data and the model estimates and is
defined as
( ) ( )
2
3
1 1
MSE= / .
n
i j i j
i j
y t m t d
= =
−
d: the degree of freedom.
For the purpose, of comparison MSE uses the degree of freedom by
assigning a large penalty to a model, with more parameters. In this proposed
model larger penalty is assigned. The smaller the MSE value, the better the

75
model fits. It is exhibited that the new model yields a little more conservative
results than the other NHPP SRGM but not significant, since the change-point
is created and may not exist.
The proposed model using the data set collected from the system T1 in
Musa (1979) are also examined.
Table-1
Comparison of goodness-of-fit of imperfect debugging NHPP models (data
from Misra)
Imperfect NHPP SRGMs
Our
Model
With
change-point
No change-point
Maxim Mize log-
likelihood
-177.53 -153.998 -154.584
MSE (critical fault) 0.0734 0.082 0.086
MSE (major fault) 1.9210 1.998 2.028
MSE (minor fault) 4.1315 4.240 4.286
Table 2 Summarizes the MSE values for investigated models. It shows that
the MSE-fit and MSE-predict values for the new model are smaller than
the other one which do not consider change-point problem (24.50 vs 25.11
vs. 62.91 and 1.01vs 1.52vs 1.175 respectively)
Table-2
Comparison of descriptive and predictive power of imperfect debugging
NHPP models (data from Mua)
Our
Model
With change-point No change-point
Maxim Mize log-
likelihood
-199 -215 -223
MSE (fit 24.50 25.11 62.91
MSE (predict) 1.00 1.01 1.75
The results of using simulation data to verify the new models are shown in
Table.3. In the data set 2, the descriptive error for the model without

76
considering change-point is poor and unacceptable. Compared with the
model with the assumption of a change-point, the new model gives very
promising results where the MSE measures for description and prediction
are 12.81 and 5.36 respectively.
Table-3
Comparison of descriptive and predictive power of imperfect debugging
NHPP models (simulated data)
Our
Model
With
change-point
No change-point
Data Set -1 MSE (fit)
MSE (predict)
4.91
1.01
5.01
1.21
9.88
1.63
Data Set -2 MSE (fit)
MSE (predict)
10.77
4.00
12.81
5.36
519.18
9.45
8. Genetic Algorithms:
We apply a genetic algorithms based optimizer to slove the formulated
mathematical models of the parameter estimation for software reliability.
Genetic algorithms (GAs) have been used extensively for dealing with
optimization problems. GAs is based on the biological evolution process, and
was firstly introduced by Holland (1975) in the 1970s. GAs is useful (Goldberg
1989) where the search space is large, nonlinear and noisy, and solutions are
ill-defined a priori.
The proposed model using the data set collection from the System T1 in Musa
(1979) are also examined. These data sets include 136 faults found in the test
phase. Table-4 summarizes the MSE (mean squared error) values investigated
models. It shows that the MSE (fit) and MSE (predict) values for the model are
smaller than the other one.

77
Table-4
Our
Model
With
change-point
No change-point
MSE (fit)
MSE (predict)
25.01
0.821
27.734
0.838
59.471
4.688
9. Conclusion:
The developed NHPP SRGM is unique in that it allows for analysis of
software failure data with change point, imperfect debugging, and various fault
types. From tables 1, 2, 3 and 4 we say that our proposed model is rated as
better than the other consider models with respect to all the conditions are
chosen. They for genetic algorithm are more suitable to our model with
minimum disabances than comparative models.
10.References:
[1]. Chang, I.P., 1997. An analysis of software reliability with change-point
Models. NSC 85-2121-M031-003, National Science Council, Taiwan.
[2 ].Fenton, N.E., Pfleeger, S.L., 1997. Software Metrics: A Rigorous and
Practical Approach. PWS Publishing Company, Boston.
[3]. Goel, A.L., Okumoto, K., 1979. Time-dependent error-detection rate
model for software reliability and other performance measures.
IEEE Trans. Reliab. R-28, 206–211.
[4]. Goldberg, D.E., 1989, Genetic Algorithms in Search, Optimization, and
Machine Learning, (Addison-Wesley).
[5]. Hinkley, D.V., 1970. Inference about the change-point in a sequence of
Random variables. Biometrika 57, 1–16.

78
[6]. Huan jyh shyur, 2003. A stochastic software reliability model with
imperfect-debugging and change-point, Elsevier, the journal of systems and
software.135-141.
[7]. Huan jyh shyur, Mu-chen chen,Analyzing software reliability Growth
model with imperfect-debugging and change-point by Genetic Algrothms.
[8]. Jelinski, Z., Moranda, P.B., 1972. In: Software Reliability Research,
Statistical Computer Performance Evaluation. Academic Press,
New York, p. 465.
[9]. Misra, P.N., 1983. Software reliability analysis. IBM Syst. J. 22, 262–
270.
[10]. Musa, J.D., 1979. Software reliability measures applied to system
Engineering. In: Proceedings CONPCON.
[11]. Musa, J.D., Iannino, A., Okumoto, K., 1987. Software Reliability:
Measurement Prediction Application. McGraw-Hill, New York.
[12]. Ohba, M., 1984. Software reliability analysis model. IBM J. Res.
Develop. 28, 428–443.
[13]. Pham, H., 1993. Software reliability assessment: Imperfect debugging
and multiple failure types in software development. EG&GRAMM-
10737, Idaho National Engineering Laboratory.
[14]. Pham, H., Nordmann, L., Zhang, X., 1999. A general imperfectsoftware-
Debugging model with s-shaped fault-detection rate. IEEE
Trans. Reliab. 48, 169–175.
[15].Yamada, S., Ohba, M., Osaki, S., 1983. S-shaped reliability growth
Modeling for software error detection. IEEE Trans. Reliab. 12,
475–484.
[16]. Zhao, M., 1993, Change-point problems in software and hardware
reliability.Commun. Statist – Theory Meth., 22, 757-768

79
Authors
Dr. R. Satya Prasad Received Ph.D. degree in Computer Science
in the faculty of Engineering in 2007 from Acharya Nagarjuna
University, Guntur, Andhra Pradesh, India. He have a
satisfactory consistent academic track of record and received
gold medal from Acharya Nagarjuna University for his out
standing performance in a first rank in Masters Degree. He is
currently working as Associative Professor and Head of the
Department, in the Department of Computer Science &
Engineering, Acharya Nagarjuna University. He has occupied
various academic responsibilities like practical examiner, project
adjudicator, external member of board of examiners for various
Universities and Colleges in and around in Andhra Pradesh. His
current research is focused on Software Engineering, Image
Processing & Database Management System. He has published
several papers in National & International Journals.
O.NagaRaju received the Masters Degree in Computer Science
& Engineering from Acharya Nagarjuna University, Guntur, Andhra
Pradesh, India. He is currently pursuing Ph.D., at Department of
Computer Science and Engineering, Acharya Nagarjuna University,
Guntur, Andhra Pradesh, India. His research interests include Software
Engineering, Network Computing, and Image Processing.

SRGM with Imperfect Debugging by Genetic Algorithms

More Related Content

Similar to SRGM with Imperfect Debugging by Genetic Algorithms (20)

Recently uploaded (20)

SRGM with Imperfect Debugging by Genetic Algorithms