See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267792387

Jelinski – Moranda Model for Software Reliability Prediction and its G.A.
based Optimised Simulation Trajectory

Article

CITATION READS

1 145

3 authors, including:

Sona Ahuja
Dayalbagh Educational Institute
14 PUBLICATIONS   11 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Sona Ahuja on 16 September 2017.

The user has requested enhancement of the downloaded file.

Jelinski – Moranda Model for Software Reliability Prediction and its G.A.
based Optimised Simulation Trajectory
Sona Ahuja, Guru Saran Mishra and Agam Prasad Tyagi
Abstract - Software reliability is the key
concern of many users and developers of
software. Although it is difficult to
measure the reliability of software before
its development is complete yet if data in
terms of inter-failure time are available,
predictions about its reliability can be
made. It is with this objective that Jelinski
– Moranda reliability model has been used
in the present paper. However, there are
other parameters like hazard rate, density
function, etc. that effect the prediction. A
real G.A. based evolutionary system using
Simulated Annealing Methodology has,
therefore, been developed to incorporate
effectiveness in predicted reliability
model. The model develops optimised
simulation trajectory using Musa-data set.
The application has been shown in this
context.
Index Terms – Software Reliability,
Jelinski-Moranda model, Hybrid
Stochastic Search Techniques, Simulated
Annealing.
I. INTRODUCTION
With the ever increasing role that
software is playing in our systems, concern
has steadily grown over the past few
decades on ‘software quality’
characterised mainly by reliability besides
other factors.
Sona Ahuja is with Faculty of Education,
Dayalbagh Educational Institute, Dayalbagh, Agra-
282005.
Guru Saran Mishra is with Faculty of
Engineering, Dayalbagh Educational Institute,
Dayalbagh, Agra-282005.
Agam Prasad Tyagi is with Faculty of Science,
Dayalbagh Educational Institute, Dayalbagh, Agra-
282005.
Software reliability is the
probability of failure free operation of a
computer program for a specified period of
time in a specified environment. There are
many reasons for software to fail but
usually these are attributed to design
problems resulting from new or changed
requirements, revisions or corrections, etc.
If existing problems, do not always create
failures immediately then they are
triggered only at certain states and inputs.
Reliability is defined in terms of
operational performance, something that
one cannot measure before the product is
finished. In order to provide reliability
indicators before the system is complete a
model is built of the factors that effect
reliability and then its predictions are
made based on one’s understanding of the
system while it is under development ([1],
[4]). A prediction system consists of a
mathematical model together with a set of
prediction procedures for determining
unknown parameters and interpreting
results ([1], [2], [3], [4]). A software
reliability model specifies the general form
of the dependence of the failure process on
the principle factors that affect it, namely
fault introduction, fault removal and the
environment reliability. Such modelling
has three broad stages viz., assessment,
model development and measurement [6].
Different software reliability
models can produce very different answers
when called upon to predict future
reliability in a reliability growth context
([8], [9]). Jelinski-Moranda reliability
model has been concentrated upon for the
prediction of the next time to failure in the
present paper. To analyse these prediction,
different measures have been used ([4],
[5], [7]). For the sake of further
convenience and effectiveness in
predictions, optimised simulation f (x i t i −1 ) = φ[N − (i − 1)] exp (− φ[N − (i − 1)]x i )
trajectory has been generated by ……(1)
evolutionary algorithm for predicted mean and cumulative density function be
time to failure. Fi (t i ) =1 − exp (− λ i t i ) ……(2)
and 1 φ [N − (i − 1)] = 1 λ i ……(3)
II. JELINSKI-MORANDA (J-M) MODEL
This model is credited with being where, λ i is hazard rate .
the first reliability model. It belongs to a
class of exponential order statistic model Since this exponential model belongs to
that assumes that fault detection and the binominal type, using equations (2),
correction begins when a program contains (3) one gets –
N faults and all the faults have the same µ (t ) = N(1 − exp(− φt )) ……(4)
rate φ. λ(t ) = Nφ(exp− (φt )) ……(5)
where µ(t ) is mean value function and
A. Assumptions and Data Requirement:
The basic assumptions of the λ(t ) is failure density function
model are: It is clearly a finite failure type model as –
1. The rate of fault detection is lim
t →∞
µ(t ) = lim
t →∞
N(1 − exp(− φt )) = N …(6)
proportional to the current fault content
of the software. C. Model Estimation and Reliability
2. The fault detection rate remains Prediction
constant over the intervals between The maximal likelihood estimates,
fault occurrence. MLE’s, calculated from the joint density
3. A fault is corrected instantaneously of the x i ’s, are the solutions to the
without introducing new faults into the
following equations:
software. ∧
n
4. The software is operated in a similar φ= ∧ n ……(7)
 x  − n (i − 1)x
manner as that in which reliability N ∑ i
 ∑ i
predictions are to be made. i =1 i =1

5. Every fault has the same chance of n 1 n

∑ ∧
=
N − (i − 1)
being encountered within a severity i =1   n
∧
 
class as any other fault in that class. N − n 1  ∑ (i − 1)x i 
6. The failures, when the faults are  ∑ x i  i −1 
detected, are independent.  i −1 
Time taken between successive ……(8)
failures is the data required for the A software program has been developed in
considered problem which has been taken C on Pentium III for the solution of
from the Musa data set [4] as given in equation (8). The solution so obtained is
Appendix-I. put in equation (7) for finding maximum
∧ ∧
likelihood estimates (MLE’s) N and φ .
B. Model Form Various reliability measures can then be
From the overview of the model ∧

and the assumptions, one can determine obtained by replacing the quantities N and
∧
that if the time-between-failure φ in the reliability function of interest. As
occurrences are xi = ti – ti-1, i = l, …., n,
an example, the mean time to failure
then the xi’s are independent exponentially
(MTTF) for the (n+1)th fault has its MLE
distributed random variables with mean.
as:
Let f(ti) be probability density function for
particular time ti such that –
 ∧
1
MTTF = ∧ ∧
……(9)
 
φ N − n  1.2
 
1
D. Algorithm
The algorithm for solving equation (8) is 0.8

given below: 0.6

1. min = 0.1 0.4
2. For N=3 to N=50 True Value
0.2
3. begin Predicted Value
4. pl = f(N) 0
5. if (p / < min) 0 0.05 0.13 0.2 0.31 0.48 0.61 0.69 0.88
6. begin Fig.2: Y Plot
7. min = p1
8. print min, N
9. end Further prequential likelihood function and
10. end Braun statistic concept has also been
Musa data set has been used in the utilised to deal with the noisiness in
above program. The value of N is system predictions. MTTF is therefore
initialized. The algorithm gives value of N introduced for such analysis. Braun
for which f(N) is minimum – statistic is a measure of variability and
where gives the quantitative inference that how
noisy a model is. Prequential likelihood
f (N ) = ∑
n
1 n
− function is a technique particularly using
i =1 N − (i − 1)  
 1  n narrow windows to detect changes in
N − n  ∑ (i − 1)x i  preferences and MTTF also helps in
 ∑ x i  i =1  analysing the noisiness of prediction
 i =1  system by showing a series of fluctuations
……(10) in its MTTF plot as depicted in Figure 3.

E. Predictive Accuracy
The concept of Kolmogorov – 3500

Smirnov distance ([1], [4]) has been used 3000

to determine and analyse the accuracy of ESTIMATED PREDICTION OPTIMISED TRAJECTORY
2500
prediction system. For this purpose U-plot
(Fig.1) and Y-plot (Fig.2) have been 2000
Mean Time------->

obtained as a measure of pessimistic or 1500

optimistic prediction as regards to its level
1000
of closeness with true distribution.
500

0
40
49
54
59
67
72
78
81
88
96
105
112
116
122
125
129
134

1.2
1.2 Next Failure------>
1 FIG.3: Mean Time to Failure for Estimated and Optimised Predictions
1
0.8
0.8
0.6
0.6
True
III. OPTIMISED SIMULATION
0.4 0.4 TRAJECTORY
Predicted
True
0.2 0.2
Predicted
0 0 It is prudent to make use of
0 0.18 0.52 0.7 00.81
0.050.13
0.910.20.95
0.31 0.48
0.990.61 1
0.690.88
multiple statistical model to predict
Fig.2: YPlot
Fig.1: U Plot reliability quality of any predictive
environment as in the case of software 4. Select the best child as the parent for the
reliability J-M model. next generation.
It would indeed be the most 5. Repeat Step 7 to Step 10 for each family
convenient for predicting reliability 6. Count=0
behaviour to generate optimised 7. Repeat Step 9 for each child; Goto Step 10
8. Increase count.
simulation in the predictive space defined 9. Acceptance number of the family is equal
by J-M model by applying Hybrid to count (A)
Stochastic Search Technique based on 10. Sum up the acceptance numbers of all the
Simulated Annealing in G.A. which have families (S)
proved to be effective as a solution tool for 11. For each family i, calculate the number of
complex optimised problems in number of children to be generated in the next
fields. generation according to the following
formula: m(i) = (T * A) / S
A. Problem Solving T = Total number of children generated by
all the families.
Objective function: 12. Decrease the temperature
13.Repeat Step 2 to Step 13 until a certain
N − i +1 number of iterations has been reached
Minimise MTTF= K
λ i (N − n )
where, K is the smoothing factor to bring The optimum solution has been shown in
unbiasedness and non-noisiness in the Table 1.
prediction accuracy.
Table 1: Optimised Solution
Subject to 0.002 ≤ λI ≤ .007 Population Size: 40
35 ≤ n ≤ 150 No. of Iteration: 50
1 ≤ i ≤ 100
1 ≤ N ≤ 200 N 40 50 60 70 80 90
MTTF 200 350 300 500 550 1100
90 ≤ K ≤ 100 (Model)
and N≥n MTTF 214.2 256.5 321.3 428.4 642.6 1285
≥i (HSST)

B. Hybrid Stochastic Search Technique

(HSST)
HSST is a hybrid approach, which
incorporates Simulated Annealing in the
selection process of GA. Thus, HSST provides
the advantages of both GA and SA. The HSST
algorithm can be expressed concisely in the
form of a pseudo-code as given below: ([10],
[11], [12], [13])
PSEUDO-CODE
Initialise
(i) Set initial temperatures
(ii) Randomly select N parent strings
(iii) Number of children to be generated by
each parent
1. For each parent i, generate m(i) children
using crossover.
2. Perform mutation with a probability pm
3. Find the best child for each parent based
on the fitness and constraints.
IV. CONCLUSION
Brawn Statistic, Prequential
likelihood function, U-plot and Y-plot
have been used to estimate the biasedness
and noiseness of the software reliability as
a measure of its quality. Kolomgrove-
Simrnov distance has been used wherever
necessary.
It has been observed that
Kolmogov - Simrnov distance of Y-plot is
less than of U-plot meaning thereby that
the predictions in Y-plot are closer to true
reliability for the adopted data set as
shown in Figures (1) and (2).
The analysis given by MTTF for
J-M Model infers that initially it is less
noisy but later becomes significantly noisy
as depicted in Figure 3.
GA based Hybrid Stochastic
Search Technique, has turned out to be
good tool for optimised simulated
trajectory for variables which are
important performance indicators like λ, φ
etc. to predict quality of reliability of the
predicted software failure. The simulated
trajectory is also shown in Figure (3).
The MTTF values have been found
to be much closer for predictions which
are less noisy. However, it can be inferred
that more experimental runs would try to
make better smoothing even at noisier
ranges by varying the bounds of correction
factor in the proposed Jelinski – Moranda
model.
Such experiments will be useful for
better reliability monitoring of softwares.
References
[1] Lyu.M(ed.), Handbook of Software
Reliability Engineering, IEEE
Computer Society Press, 1996.
[2] Abdel-Ghaly, A.A., Chan, P.Y. and
Littlewood, B., “Evaluation of
Competiting Software Reliability
Predictions,” IEEE Transactions on
software Engineering, SE-12(9), p.p.
950-67, 1986.
[3] Fenton, N.E., Pfleeger, S.L.,
Software Metrics, International
Thomson Computer Press, London,
1996.
[4] Musa, J., Lannino, A. and Okumoto,
K., Software reliability:
Measurement, Prediction and
Application, McGraw Hill, New
York, 1985.
[5] Littlewood, B., “How to measure
Software Reliability and how not
to,” IEEE Trans. Rel., Vol. R-28,
pp.103-110, June 1979.
[6] Lyu, M., Nikora, A., “Applying
Reliability Models more
effectively,” IEEE Software 9(4),
July 1992.
[7] Srimani Pradip K., Malaiya,
Yaswant K., “Steps to practical
Reliability measurement,” IEEE
Software 9(4), July 1992.
[8] Schick, G.J., Wolverton, R.W., “An
analysis of competing Software
Reliability Model”, IEEE
Transactions on Software
Engineering, Vol.SE-4, No.2, March
1978.
[9] Brocklehurst, S., Chan, P.Y.,
Littlewood, B. and Snell, J.,
“Recalibrating Software Reliability
Models,” IEEE Transactions on
Software Engineering, Vol.16, No.4,
April 1990.
[10] Goldberg D.E., and Richardson J.,
“Genetic algorithms with sharing for
multimodal function optimisation”.
Proc. of the 2nd Int. Conf. on Genetic
Algorithms, 1987, 41 – 49
[11] Golderg D.E., “Genetic algorithms in
search, optimisation, and machine
learning.” Reading : Addison Wesley,
1989.
[12] Dua, A., Mishra G.S., Sharma, R.S. and
Raj, K.H., “A hybrid Stochastic Search
Technique for solving constrained
optimization problems with multiple
objectives.” Proc. Of National
Conference on Emerging Convergent
Technologies and Systems, D.E.I.
Dayalbagh, Agra, 2002, pp. 399-404.
[13] Goldberg D.E., Deb K., and Clark J.H.,
"Genetic Algorithms, noise and the
sizing of population." Complex Syst.,
1992, 6, 333-362.
 Appendix-I

Musa Data Set

Execution times in seconds between successive failures (left to right)

3 30 113 81 115 9 2 91 112 15

138 50 77 24 108 88 670 120 26 114
325 55 242 68 422 180 10 1146 600 15
36 4 0 8 227 65 176 58 457 300
97 263 452 255 197 193 6 79 816 1351
148 21 233 134 357 193 236 31 369 748
0 232 330 365 1222 543 10 16 429 379
44 129 810 290 300 529 281 160 828 1011
445 296 1755 1064 1783 860 983 707 33 868
724 2323 2930 1461 843 12 261 1800 865 1435
30 143 108 0 3110 1247 943 700 875 245
729 1897 447 386 446 122 990 948 1082 22
75 482 5509 100 10 1071 371 790 6150 3321
1045 648 5485 1160 1864 4116
View publication stats