You are on page 1of 13

International Journal of Systems Science, 2013

http://dx.doi.org/10.1080/00207721.2013.791001

A structure-based software reliability allocation using fuzzy analytic hierarchy process


Subhashis Chatterjee , Jeetendra B. Singh and Arunava Roy
Department of Applied Mathematics, Indian School of Mines, Dhanbad, India
(Received 27 August 2012; final version received 2 March 2013)

During the design phase of a software, it is often required to evaluate the reliability of the software system. At this stage of
development, one crucial question arises how to achieve a target reliability of the software? Reliability allocation methods
can be used to set reliability goals for individual components. In this paper, a software reliability allocation model has been
proposed incorporating the user view point about various functions of a software. Proposed reliability allocation method
attempts to answer the question how reliable should the system components be? The proposed model will be useful for
determining the reliability goal at the planning and design phase of a software project, hence making reliability a singular
measure for performance evaluation. Proposed model requires a systematic formulation of user requirements and preference
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

into the technical design and reliability of the software. To accomplish this task, a system hierarchy has been established, which
combines the users view of the system with that of the software manager and the programmer. Fuzzy analytic hierarchy process
(FAHP) has been used to derive the required model parameters from the hierarchy. Sensitivity analysis has also been carried
out in this paper. Finally, an example has been given to illustrate the effectiveness and feasibility of the proposed method.
Keywords: reliability; software reliability; reliability allocation; system hierarchy; fuzzy analytic hierarchy process (FAHP)

1. Introduction 2012; Kapur, Pham, Aggarwal, and Kaur 2012; Pievatolo,


Computer systems have gradually become an important at- Ruggeri, and Soyer 2012) have been done to achieve the
tribute in many aspects of our daily life and it is used in reliability goal of software. To apply these models, it is
many areas like: nuclear power plant, safety critical sys- necessary to know how well the models suit an actual fail-
tems, air traffic control system, communication, defence, ure data. Accurate software reliability estimation requires
medical science, etc. Failure of these systems can lead to a large number of failure data, which are not usually avail-
big losses. Many examples of spectacular software fail- able until the system has been tested for a longer period of
ures are available in literature (Pham 2006) which have time. Software reliability engineers are more interested in
caused calamitous loss of life and property. Some of them estimating the reliability of a software as early as possible
are: the crash of a Boeing 727 Mexicana airlines, over- (Junhong, Xiaozong, and Hongwei 2005).
dose given to cancer patients by the massive Therac-25 In recent years, a major concern of software developers
radiation machine in Marietta, Apollo mission problems is how to use an effective, controlled and planned software
in the 1970s, AT & T long-distance telephone switching development process to produce highly reliable software.
network failure in 1990, etc. Therefore, it is necessary for Reliability allocation technique plays an important role
a software system to work with an acceptable degree of during the design phase of software in fixing and allocating
reliability. a target reliability. Reliability allocation is a method
The standard definition of software reliability is the of allocating a target reliability among subsystem and
probability of execution of software without failure under components. Since, software development is a major cost
specified environment for a specified period of time (Pham element of computerised application systems, obtaining
2006). Software failures are the manifestation of software an accurate reliability allocation during the early phase of
errors, which are introduced into the system by the soft- development is an important requirement for achieving an
ware developers during its development phase. In the liter- optimal system reliability goal. Software managers must
ature, various software reliability growth models (SRGMs) be able to estimate the resource required for alternative
(Xie 1991; Yamada, Tokuno, and Osaki 1992; Lyu 1996; system designs during the design phase of the project.
Pham 1996, 2006; Chatterjee, Misra, and Alam 1997; Failure to do so may result in a significant waste of resource
Kapur, Garg, and Kuamr 1999; Zhang and Pham 2000) are and perhaps even in the failure to deliver a completed
available. Some recent studies (Ishii, Dohi, and Okamura system (Aggarwal and Singh 1995).


Corresponding author. Email: chatterjee_subhshis@rediffmail.com


C 2013 Taylor & Francis
2 S. Chatterjee et al.

Initially, research related to the reliability allocation the concept of fuzzy set theory is applicable in developing
problem was confined to the hardware system (Tillman, software reliability models. Cai, Wen, and Zhang (1991,
Hwang, and Kuo 1977; Nakagawa and Miyazaki 1981; 1993) have discussed about the applicability of fuzzy set
Painton and Campbell 1995). Later, software reliability al- theory in software reliability modelling. Software reliability
location has received attention (Kubat 1989; Zahedi and allocation problem is a decision-making and optimisation
Ashrafi 1991; Lee 1991; Poore, Mills, and Mutchler 1993; problem in nature. There are many factors which influence
Poore and Trammell 1995; Leung 1997; Lyu, Rangarajan, the reliability allocation. These factors include many un-
and van Moorsel 1997; Helander, Zhao, and Ohlsson 1998; certain fuzzy characteristics, and it is difficult to establish
Leung 2004). Poore et al. (1993) and Poore and Trammell by specific mathematical model (Yi and Chengwen 2009).
(1995) used a spreadsheet approach to consider various In this paper AHP with the fuzzy decision (FAHP) has been
strategies for allocating reliability to software modules. Za- applied to optimise reliability allocation problem for soft-
hedi and Ashrafi (1991) adopted analytic hierarchy process ware system. Proposed model provides a unified approach
(AHP) for modelling the software architecture with cost in which the users requirements and preferences are for-
as the constraints and proposed a method for the system mally integrated with the technical structure of the soft-
reliability maximisation. For reliability allocation problem, ware, its modules and programme reliabilities. Proposed
Leung (1997) has used the operational profile to define model determines reliability goal at the planning and de-
a software utility function, that reflects a weighted sum sign phase of a software project. Hence, making reliability
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

of reliability-like measures based on the same AHP pro- a singular measure for performance evaluation and project
cess used by Zahedi and Ashrafi. Kubat (1989) presented control. Reliability allocation makes it possible to plan for
a stochastic model to minimise cost subject to an over- system reliability growth realistically. It helps in designing
all system failure intensity goal. Helander et al. (1998) schemes for achieving the target reliability, and it provides
described a reliability allocation model called RCCM a guideline for the allocation of programming talents and
(Reliability Constrained Cost Minimisation), which is used control scrutiny in the implementation stage. During the
to assign the reliability. Rani and Misra (2005) proposed an testing stage, the reliability allocations provide the target
economic model to allocate reliabilities during the design value against which the software performance should be
phase by minimising a cost function, which depends on tested. In this proposed model, a hierarchy has been con-
fixed development costs and on a previously experienced structed, which links users view about reliability to the
failure decrease cost. Some architecture-based approach software managers and programmers view of the software.
(Guan, Wang, and Chen 2009; Pietrantuono, Russo, and The ultimate goal of this structure is to derive parameters
Trivedi 2010) for software reliability allocation has also needed for the formulation of a reliability allocation model.
been proposed in literature. In 2009, Guan et al. formu- The proposed method will be best suited for the reliability
lated an architecture-based model to optimise software re- allocation problem of those softwares which can be decom-
liability. They have used dynamic programming algorithm posed into hierarchical way. The proposed FAHP approach
to allocate the reliability to each component such that the has been compared with AHP using consistency ratio. Also,
software attains the desired reliability with minimum design the performance of the proposed model has been tested us-
cost. In 2010, Pietrantuono et al. suggested an architecture- ing sensitivity analysis.
based approach to allocate software reliability and test- The remaining part of this paper is organised as fol-
ing time. For modelling the architecture, they have used a lows: Section 2 describes some basic definition about the
Discrete Time Markov Chain (DTMC). Kubat (1989) fuzzy set and fuzzy number. In Section 3, the FAHP has
adopted a software reliability model to characterise the been discussed. Section 4 describes the proposed reliability
software operational profile and formulated the problem allocation model based on a hierarchical structure. Section
of minimising the total development cost subject to a given 5 illustrates model validation. Section 6 describes the com-
requirement on the software failure rate. Tamura and Ya- parative study between the proposed method and AHP, as
mada (2009) formulated the maintenance effort model to well as sensitivity analysis. Finally, Section 7 concludes the
control the software development process in terms of re- paper.
liability, development effort and version-upgrade time for
open source software. Also, some optimal resource alloca- 2. Fuzzy sets and fuzzy number
tion problems (Huang and Lyu 2005; Huang and Lo 2006) The concept of fuzzy set theory has been introduced by
have been discussed to minimise the cost of software de- Zadeh (1965) to deal with the uncertainty due to impreci-
velopment when the fixed amount of testing-effort and a sion in data or system. Fuzzy set theory includes fuzzy set,
desired reliability objective are given. membership function and fuzzy number to change vague
Starting from requirement analysis to design, coding data into useful data efficiently.
and maintenance activities of a software development pro-
cess are performed by human being and human behaviour Definition 2.1 (Fuzzy set): Let X be a universe of dis-
is fuzzy. Also, the environment of software is fuzzy. Hence, course, A is a fuzzy subset of X for all x X. There is a
International Journal of Systems Science 3

constituent parts. The traditional AHP requires crisp judge-


ment. However, due to the complexity and uncertainty in-
volved in real-world decision problems, a decision-maker
may sometimes feel more confident to provide fuzzy judge-
ments than crisp comparisons (Van Laarhoven and Pedrycz
1983; Buckely 1985; Chang 1996; Wang, Chu, and Wu
2007). The fuzzy AHP methodology extends Saatys AHP
by combining it with fuzzy set theory. The procedures of
the fuzzy AHP involve three essential steps (Buckely 1985;
Figure 1. A triangular fuzzy number. Chang 1996):
(i) Define your goal and decompose the unstructured
number A (x) [0, 1] which is assigned to represent the problem into a hierarchical structure with decision
membership of x to A and A (x) is called the membership elements (criteria and alternatives).
function of A (Zadeh 1965). (ii) Employ pairwise comparisons among decision ele-
ments to establish the fuzzy judgement matrix and
Definition 2.2 (Fuzzy number): A fuzzy number A is a a fuzzy weight vector.
normal and convex fuzzy subset of X . Here, the convex (iii) Aggregate the relative weights of decision elements
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

set implies that to obtain an overall rating for the alternatives and
x1 , x2 X and [0, 1], select the optimal one.
A (x1 + (1 )x2 ) min(A (x1 ), A (x2 )) (Zadeh
1965). There are different fuzzy AHP approaches proposed by
various authors for alternative selection and justification
Definition 2.3 (Triangular fuzzy number): A triangular problem (Van Laarhoven and Pedrycz 1983; Buckely 1985;
fuzzy number (TFN), defined to be a normal and convex Chang 1996; Mikhailov 2003; Mikhailov 2004; Wang, El-
fuzzy subset of X and denoted as A = (a, b, c), has the hag, and Hua 2006). Chang (1996) extent analysis method
following membership function (graphically represented in has been used in this paper since, the steps of this ap-
Figure 1) (Kaufmann and Gupta 1991): proach are similar to the conventional AHP and rela-
tively easier than the other fuzzy approaches (Lee 2010).
xa Various applications of FAHP based on Changs extent

, a x b,

ba analysis in different areas are available in literature (Ce-


lik, Er, and Ozok 2009; Lee 2010; Mikaeil, Yousefi,
A (x) = cx (1)
, bxc and Ataei 2011; Reza, Mohammad, and Reza 2011). In

cb

this section, the outline of Chang (1996) extent analy-


sis method on fuzzy AHP has been described. Let X
0, otherwise
= {x1 , x2 , . . . , xn } be an object set, and U = {u1 ,
u2 , . . . , um } be a goal set. According to the princi-
The operation laws on a TFN are as follows ples of Changs extent analysis, each object is consid-
(Zimmermann 1991): ered correspondingly and extent analysis for each of
the goal gi is executed. It implies that it is possi-
(a1 , b1 , c1 ) (a2 , b2 , c2 ) = (a1 + a2 , b1 + b2 , c1 + c2 ) ble to obtain the values of m extent analysis that can
(2) be demonstrated as Mg1i , Mg2i , . . . , Mgmi , i = 1, 2, . . . , n,
j
where Mgi (j = 1, 2, . . . , m) are TFNs. The steps of
(a1 , b1 , c1 ) (a2 , b2 , c2 ) = (a1 a2 , b1 b2 , c1 c2 ) Changs extent analysis are as follows:
(3) Step 1. The value of the fuzzy synthetic extent with respect
  to the ith object is defined as
1 1 1 1
(a, b, c)1 = , , (4) m 
n m
a b c
Si = Mgji Mgji (5)
j =1 i=1 j =1

3. Fuzzy analytic hierarchy process (FAHP) j


AHP was first described by Thomas Saaty (1980) and has To obtain m j =1 Mgi , perform the fuzzy addition operation
since been used in numerous applications as a decision- of extent values for a particular matrix such that

making framework that organises the dimensions of a com- m m 
m 
m
plex system into a hierarchical structure. It provides a sys- Mgji = lj , mj , uj (6)
tematic procedure to decompose a problem into smaller j =1 j =1 j =1 j =1
4 S. Chatterjee et al.
 n m j 1
To obtain i=1 j =1 Mgi , perform the fuzzy addition Assume that d (Ai ) = minV (Si Sk ) for k = 1, 2, . . . , n;
operation of
j
Mgi , j = 1, 2, , m such that k = i, then the weight vector is given by

 n  W = (d (A1 ), d (A2 ), . . . , d (An ))T , (12)



n 
m  
n 
n
Mgji = lj , mj , uj (7)
i=1 j =1 i=1 i=1 i=1 where Ai (i = 1, 2, . . . , n) are n elements.
Step 4. Via normalisation, the normalised weight vectors
Compute the inverse of the vector in Equation (7) such that
are
1

n 
m  
1 1 1 W = (d(A1 ), d(A2 ), . . . , d(An ))T , (13)
Mgji = n , n , n
i=1 j =1 i=1 ui i=1 mi i=1 li
where W is a non-fuzzy number.
(8)
Step 2. The degree of possibility of M2 = (l2 , m2 , n2 ) =
M1 = (l1 , m1 , n1 ) is defined as 4. Software reliability allocation model
User is the ultimate judge of systems quality and reliability.
V (M2 M1 ) = sup
min(M1 (x), M2 (y))
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

(9) Their opinion and perceptions about reliability of various


yx
functions must be used as a keystone before developing
any scheme. Users opinion may be different from that of
and can be equivalently expressed as follows:
software manager and programmer. To achieve the goal, one
V (M2 M1 ) = hgt(M1 M2 ) = M2 (d)
must find a way to combine all view points for allocating
the reliability values to various modules, programmes and
1 if m2 m 1

functions of the software.



0 if l2 u1
= (10)



l1 u2
4.1. Hierarchy of software system

, otherwise In this subsection, a structure of the hierarchy has been set
(m2 u2 ) (m1 l1 )
up, which links the users view about reliability to the soft-
Figure 2 illustrates Equation (10), where d is the ordinate ware managers and programmers view of the software.
of the highest intersection point D between M1 and M2 . The ultimate goal of this structure is to derive parame-
To compare M1 and M2 , both the values V(M1 M2 ) and ters needed for the formulation of the reliability allocation
V(M1 M2 ) are required. model. In AHP, there is a top-down approach using which
the problem is structured into a hierarchy. A bottom-up ap-
Step 3. The degree of possibility for a convex fuzzy number proach mathematically combines the levels of the hierarchy
to be greater than k convex fuzzy numbers Mi (i = 1, 2, . . . , (Zahedi and Ashrafi 1991; Lee 1993; Aggarwal and Singh
k) can be defined by 1995). In the proposed model, the hierarchy starts from the
top with the users view, which has been defined as the over-
V (M M1 , M2 , . . . , Mk ) = minV (M Mi ), all software reliability or system reliability goal, denoted by
i = 1, 2, . . . , k. (11) R.
The user bases his assessment on the functionality and
attributes of the software, which are represented at the sec-
ond level of the hierarchy. The user expects the software to
perform a set of functions and produce the desired result.
This is the users view about software. Assume that a typical
user enumerates f such function and denotes them by F1 ,
F2 , . . . , Ff .
The third level of the hierarchy represents the computer
programme written to accommodate the functions specified
by the user. This level denotes the software engineers (SE)
view of the software. There is no reason to believe that each
user-specified function would be programmed into only one
programme, i.e., a one-to-one mapping from function to
programmes could be inefficient and unrealistic (Zahedi
Figure 2. The intersection between M1 and M2 . and Ashrafi 1991). Therefore, each function may use more
International Journal of Systems Science 5

(ii) Employ pairwise comparisons among decision ele-


ments and establish the fuzzy judgement matrix and
a fuzzy weight vector.
(iii) Aggregate the relative weights of decision elements
to obtain an overall rating for the alternatives and
select the optimal one.

The first step is already completed by Figure 3. In the


second step, collect the data in the form of a pairwise com-
parison fuzzy judgement matrix expressed as

(1, 1, 1) (l12 , m12 , u12 ) (l1n , m1n , u1n )
(l21 , m21 , u21 ) (1, 1, 1) (l2n , m2n , u2n )

Figure 3. The hierarchy of software reliability. .. .. ..
. . .
than one programme as shown in Figure 3. Denote the (ln1 , mn1 , un1 ) (ln2 , mn2 , un2 ) (1, 1, 1)
software programmes at the third level by P1 , P2 , . . . , Pn . (14)
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

The fourth level of hierarchy contains the independent


 
modules of which the programmes are composed. In the where aij = (lij , mij , nij ) = 1/uj i , 1/mj i , 1/ lj i for i, j =
formulation, it is assumed that SE adheres strictly to the 1, 2, n and n = j. To calculate a priority vector of the above
concept of structured programming, which has become an triangular comparison matrix, Chang (1996) suggested an
inevitable programming approach for medium and large extent analysis method, which has already been discussed
systems, which in turn, are prime candidates for using in Section 2.
a reliability allocation model (Zahedi and Ashrafi 1991; At the second level, users asked to compare the func-
Aggarwal and Singh 1995). In this paper, it is assumed tions (Fi s) pairwise. This is required to determine relative
that the modules are independent units which themselves importance of the functions in the total assessment of the
may have submodules, but each submodule belongs to only software reliability goal. The relative importance of each
one module. Furthermore, assume that the software uses function can be evaluated by pairwise comparisons. The ba-
m modules denoted by M1 , M2 , . . . , Mm . Stop hierarchical sic premise of pairwise comparisons is that two alternatives
structure at the level of the independent modules. are compared at a time until all combinations of comparison
are considered. Thus, the total number of pairwise evalu-
ation is 2(ff !2)! , where f is the total number of functions
4.2. Relative reliability weights involved in the assessment of the software. Input matrix is
In this subsection, a methodology has been proposed to constructed by asking users reliability requirements. User
show how one can determine the relative weights of the might ask that function F1 is more important than F2 and
modules using FAHP. Here the goal is to identify the relative extremely important than F3 , the answer is based on TFN
weights of each programme Pi at the third level and each (Kaufmann and Gupta 1991). The results of all pairwise
module Mi at the fourth level of the hierarchy, from the comparisons are stored in an input matrix Af = [afij ] a
knowledge of the users reliability goal R at the first level. f f matrix. The element afij states the importance of alter-
Obviously, one is unable to ask directly the user to express native i compared to alternative j. Obviously, the diagonal
his preference for each programme and module because of the matrix is always 1, and the elements of the lower
his view of the software is external one, that ends at the triangle are the inverse of the corresponding elements at
functional level (Fi s) of the hierarchy. The programmer too the upper triangle of the matrix. The relative importance of
has only a partial view of the system because he can see each function in determining the software reliability goals
only the third and fourth levels ( Pi s and Mi s). Hence, to is elicited using Changs extent analysis method (Chang
achieve the goal, one must find a way to combine the two 1996). A vector Wi of relative weight has been derived
views for identifying the relative reliability weights of the using extent analysis method. Since there are f such func-
programme and their modules with the users assessment tions at second level the number of Wi s will be f. These
about the software. To accomplish this task FAHP method are called relative weights for functions and represented
has been used hare. The procedures of the FAHP involve as (WF1 , WF2 , . . . , WFf ). Next step is to combine these
three essential steps: weights to arrive at the relative reliability requirements of
programmes and achieve the system reliability goal. Input
(i) Define your goal and decompose the unstructured matrix is constructed by asking users reliability require-
problem into a hierarchical structure with decision ments. Input matrices Aip , Aim are constructed by pairwise
elements (criteria and alternatives). questioning process and relative weights are calculated. A
6 S. Chatterjee et al.

matrix of relative weights for the programmes may look programmes, use input matrix Am j where j signifies that
like: input matrix is for programme Pj where j = 1, 2, . . . , n.
For programme j, weights are calculated using the

WP 11 WP 12 WP 1n method described in Section 2 as follows:
WP 21 WP 22 1
WP n

.. .. .. .. . j
. . . . WMj = (WM k )j = 1, 2, . . . , n
f f
WP 1 WP 2 WP fn
k Q {1, 2, . . . , m| for all modules connected to Pj },
WPi s are the relative weights at the programme level. where k takes values in a set Q which is a subset of {1, 2,
A superscript on a weight vector indicates that it is a lo- . . . , m}.
cal relative weight of the programme j corresponding to Hence, allocated reliability will be
function i. Allocated reliability RFi , for function i will be
RF i = (R)W Fi provided all the functions are executed al- j
RM k = (RP j )WM k j = 1, 2, . . . , n
j
ways.
j
For third level one must produce f matrices of pairwise RM k = Maximum(RM k ) k = 1, 2, . . . , m
comparisons. This is because, the programmes may play
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

different roles in accomplishing functions F1 through Ff .


j Y {1, 2, . . . , n| for all programmes in which the module
Take function Fi , one input matrix of comparisons of pro-
k is called}, where j takes values in a set Y which is the subset
grammes P1 , P2 , . . . , Pn in accomplishing the mission of
of {1, 2, . . . , n}. RMk is the reliability of module k, k = 1,
the function Fi . This procedure produces a n n matrix
2, . . . , m. Hence, RFi , RPj , RMk are the allocated reliability
Ai , the superscript i signifies that the input matrix is for
values.
accomplishing function Fi which is at one higher level.
The reliability goal for the programme level will be
RFi for the programmes which are connected to function
i. To accomplish function i, an input matrix Ap i using the 5. Example of proposed software reliability model
superscript i is available to signify that the input matrix is In this section, the application of the proposed method in
for accomplishing function Fi , which is at one level higher. reliability allocation problem has been demonstrated with
Relative reliability weights are calculated using method a simple example. The hierarchical structure of software
discussed in Section 2. For function i, relative weights of reliability has been shown in Figure 4.
the programme which belong to function i are Assume that a software has been developed with three
function F1 , F2 , F3 for various applications. Furthermore, it
WP i = (WP ij ) i = 1, 2, . . . , f is assumed that users requirement of system reliability goal
is 0.95. The users have been asked to compare the functions
Fi s pairwise. Since, there are three alternatives at this level,
j S {1, 2, . . . , n| for the programme connected to Fi }
the total number of pairwise comparisons are three.
where j takes the values in a set S which is a subset of {1,
Fuzzy scale for pairwise comparisons of one attribute
2, . . . , n}. Therefore, the allocated reliability will be
over another has been shown in Table 1. The fuzzy scale
has been used when decision-makers make pairwise com-
RP ij = (RF i )WP j i = 1, 2, . . . , f
i
parisons. While collecting the experts opinions instead of
crisp ratio, linguistic statement has been used for pairwise
j S {1, 2, . . .n| for programmes connected to Fi }. comparison. In this study, TFNs have been used as the
The programme j will have different values of reliabil- membership function.
ity corresponding to different functions. If the programme The relative importance of the function in the software
j is connected to three functions, then three values of reli- reliability goal can be determined by pairwise comparison.
abilities with respect to each function are available. To be The matrix obtained from pairwise questioning process may
on safest side, highest value of reliability is chosen. be as follows:

RP j = Maximum (RP ij ) j = 1, 2, . . . , n F1 F2 F3

F1 (1, 1, 1) (3/2, 2, 5/2)
(3/2, 2, 5/2)
i X {1, 2, . . . , f| for all functions in which programme F2 (2/5, 1/2, 2/3) (1, 1, 1) (2/3, 1, 2) .
j is called RPj is the reliability of programme j j = 1, F3 (2/5, 1/2, 2/3) (1/2, 1, 3/2) (1, 1, 1)
2, . . . , n. (15)
At the next level, similar approach is used for the cal- Based on the method discussed in Section 2, Si can be
culation of reliability of various modules. For developing obtained using Equation (5) as follows:
International Journal of Systems Science 7
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

Figure 4. The software hierarchy of the simple project.

  0.31 0.46
1 1 1 V (SF 2 SF 1 ) = = 0.38
SF 1 = (4, 5, 6)  , , (0.25 0.46) (0.5 0.31)
12.84 10 7.97
0.14 0.46
= (0.31, 0.50, 0.75) V (SF 2 SF 3 ) = =1
  (0.25 0.46) (0.25 0.14)
1 1 1
SF 2 = (2.07, 2.5, 3.67)  , , 0.31 0.40
12.84 10 7.97 V (SF 3 SF 1 = 0.26
(0.25 0.46) (0.5 0.31)
= (0.16, 0.25, 0.46)
  V (SF 3 SF 2 ) = 1.
1 1 1
SF 3 = (1.9, 2.5, 3.17)  , ,
12.84 10 7.97 Finally, using the formula given in Equation (11)
= (0.14, 0.25, 0.40).
d (F1 ) = V (SF 1 SF 2 , SF 3 ) = min (1, 1) = 1
The degree of possibility (Si Sj ) (i = j) can be determined d (F2 ) = V (SF 2 SF 1 , SF 3 ) = min (0.38, 1) = 0.38
by Equations (9) and (10) as follows:
d (F3 ) = V (SF 3 SF 1 , SF 2 ) = min (0.26, 1) = 0.26.
V (SF 1 SF 2 ) = 1
Therefore
V (SF 1 SF 3 ) = 1
W (F ) = (1, 0.38, 0.26).
Table 1. Fuzzy scale.
Via normalisation W(F) will be as follows:
Triangular Triangular fuzzy
Linguistic scale scale reciprocal scale W (F ) = (0.61, 0.23, 0.16).
Just equal (1,1,1) (1,1,1)
Equally important (1/2,1,3/2) (2/3,1,2) The reliability allocation RF i = R W Fi (i = 1, 2, . . . , f ).
Weakly important (1,3/2,2) (1/2,2/3,1 ) Therefore, allocated reliability to the function F1 , F2 , F3
Moderately more important (3/2,2,5/2) (2/5,1/2,2/3) will be
More important (2,5/2,3) (1/3,2/5,1/2)
Strongly more important (5/2,3,7/2) (2/7,1/3,2/5)
RF 1 = (R)W F1 = (0.95)0.61 = 0.969
8 S. Chatterjee et al.

RF 2 = (R)W F2 = (0.95)0.23 = 0.988 = Maximum (0.978, 0.995) = 0.995


RF 3 = (R)W F3 = (0.95)0.16 = 0.991. RP 2 = Maximum (RP 12 , RP 22 , RP 32 )
= Maximum (0.989, 0.988, 0.995) = 0.995.
Relative weights of the programmes for reliability require-
ments are computed for accommodating the needs of each The relative reliability requirement of modules in the de-
function. velopment of each programme is as follows.
For function F1 , programme P1 and programme P2 are For developing programme 1, the pairwise comparison
required as shown in Figure 4. To accomplish function 1 matrix may be

M1 M2 M3

M1 (1, 1, 1) (3/2, 2, 5/2) (2/7, 1/3, 2/5)
AM1 = M2 (2/5, 1/2, 2/3) (1, 1, 1) (5/2, 3, 7/2) (19)
M3 (5/2, 3, 7/2) (2/7, 1/3, 2/5) (1, 1, 1)

The estimated relative weights according to this matrix are


the comparison matrix may be as follows:
WM1 = (WM 11 , WM 12 , WM 13 ) = (0.20, 0.42, 0.38)
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

P1 P2
  RM 11 = (RP 1 )WM 1 = (0.995)0.20 = 0.999
1

P (1, 1, 1) (1, 3/2, 2)


AP1 = 1 (16)
RM 12 = (RP 1 )WM 2 = (0.995)0.42 = 0.997
1
P2 (1/2, 2/3, 1) (1, 1, 1)
RM 13 = (RP 1 )WM 3 = (0.995)0.38 = 0.998.
1

According to the method discussed in Section 2, the esti-


mated relative weights using the matrix given in Equation
(16) is as follows: For developing programme 2, the pairwise matrix may be
WP 1 = (WP 11 , WP 12 ) = (0.68, 0.32). Allocated relia-
bility will be
M1 M2 M3

M1 (1, 1, 1) (2/3, 1, 2) (2/3, 1, 2)
RP 11 = (0.969)0.68 = 0.978
AM2 = M2 (1/2, 1, 3/2) (1, 1, 1)

(3/2, 2, 5/2)
RP 12 = (0.969)0.32 = 0.989. M3 (1/2, 1, 3/2) (2/5, 1/2, 2/3) (1, 1, 1)
(20)
For accomplishing function 2, pairwise comparison matrix The estimated relative weights according to this matrix are
may be
P2 WM2 = (WM 21 , WM 22 , WM 23 ) = (0.34, 0.42, 0.24)
 
AP2 = P2 (1, 1, 1) (17)
RM 21 = (RP 2 )WM 1 = (0.995)0.34 = 0.998
2

WP 2 = (WP 22 ) = 1
RM 22 = (RP 2 )WM 2 = (0.995)0.42 = 0.997
2

WP 22
RP 22 = (RF 2 ) = (0.988) = 0.988.
1

RM 23 = (RP 2 )WM 3 = (0.995)0.24 = 0.999.


2

For accomplishing function 3, pairwise comparison matrix


may be
Therefore
P1 P2
  RM 1 = Maximum (RM 11 , RM 21 )
P (1, 1, 1) (1/2, 1, 3/2)
AP3 = 1 (18)
P2 (2/3, 1, 2) (1, 1, 1) = Maximum (0.999, 0.998) = 0.999

The estimated relative weights according to this matrix are RM 2 = Maximum (RM 12 , RM 22 )
= Maximum (0.997, 0.997) = 0.997
WP 3 = (WP 31 , WP 32 ) = (0.5, 0.5)
RP 31 = (RF 3 )WP 1 = (0.991)0.5 = 0.995
3
RM 3 = Maximum (RM 13 , RM 23 )
WP 32 = Maximum (0.998, 0.999) = 0.999.
RP 32 = (RF 3 ) = (0.991)0.5 = 0.995.

Hence Using this allocation method the reliability goals are set for
all component of the software before designing the actual
RP 1 = Maximum (RP 11 , RP 31 ) system. For these goals to be realistic and meaningful, one
International Journal of Systems Science 9

must encompass the typical user requirements and specifi- Table 2. Randomly generated consistency index for different
cation about the reliability and the structure of the software. size of matrix.
This method improves communication among users, soft- n 1 2 3 4 5 6 7 8 9 10
ware managers and programmers for software reliability
allocation. RI 0 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45 1.49

6. Comparison
linguistic weight variable to respond to attributes of soft-
In this subsection, a comparison has been carried out with ware systems and expectation of the decision-maker.
the method used in Zahedi and Ashrafi (1991) and Aggar-
wal and Singh (1995). In Zahedi and Ashrafi (1991) and 6.1. Numerical example
Aggarwal and Singh (1995), AHP method has been used
for reliability allocation. AHP enables the decision-makers In this section, a numerical example has been given, which
to structure a complex problem in the form of a simple demonstrates the comparison between FAHP and AHP. To
hierarchy and to evaluate a large number of quantitative compare between these two approaches consistency index
as well as qualitative factors in a systematic manner under proposed by Saaty (1980) has been used here. Saaty (1980)
conflicting multiple criteria. The application of AHP to the measure of consistency is given as consistency index (CI),
n
complex problem usually involves four major steps (Cheng, where CI = max n1
. Consistency ratio = consistency index
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

Yang, and Hwang 1999): / randomly generated consistency index. Randomly gener-
ated CI represents the average CI over numerous random
entries of same order reciprocal matrices. Table 2 shows
(1) Break down the complex problem into a number of
different randomly generated CI of matrices having orders
small constituent elements and then structure the
n n.
elements in a hierarchical form.
The acceptable CR range varies according to the size
(2) Make a series of pairwise comparisons among the
of matrix, i.e., 0.05 for a 3 3 matrix, 0.08 for a 4 4 ma-
elements according to a ratio scale 1, 3, 5, 7 and 9.
trix and 0.1 for all larger matrices, n 5. If the value of
(3) Use the eigenvalue method to estimate the relative
CR is equal to or less than the above values, it implies that
weights of the elements.
the evaluation within the matrix is acceptable or indicates
(4) Aggregate these relative weights and synthesise
a good level of consistency in the comparative judgements
them for the final measurement of given decision
represented in that matrix. In contrast, if CR is more than the
alternatives.
acceptable value then, inconsistency of judgements within
that matrix has occurred and the evaluation process should
However, in complex systems, the experiences and therefore be reviewed, reconsidered and improved. An ac-
judgements of humans are represented by linguistic and ceptable consistency ratio helps to ensure decision-maker
vague patterns, and it is not quantitatively defined. There- reliability in determining the priorities of a set of criteria
fore, by fuzzy set theory, one can give a much better repre- (Saaty 1980).
sentation of these linguistic data. Case A: Let, for the software cited in Section 5, each
The AHP has some shortcomings as shown below module has five quality criteria, viz., the development cost
(Cheng et al. 1999): of a module (DC), demand of a module (DM), degree of
importance of a module (DI), last date of use of a mod-
(1) The AHP method is mainly used in nearly crisp ule (LM) and effectiveness of a module (EM). The fuzzy
(non-fuzzy) decision applications. comparison or the judgement matrix among those criteria
(2) The AHP method creates and deals with a very is given in Table 3.
unbalanced scale of judgement. To find the inconsistency of the information stored in
(3) The AHP method does not take into account the the judgement matrix, its consistency ratio (CR) has to
uncertainty associated with the mapping of ones be calculated. The largest eigenvalue corresponding to the
judgement to a number. matrix given in Table 3 is max = 5.33 (approximately). Its
(4) Ranking of the AHP method is rather imprecise. Consistency Index, CI = 0.0825. The randomly generated
(5) The subjective judgement, selection and preference CI RI = 1.12 given in Table 2. Hence, the Consistency
of decision-makers have great influence on the AHP Ratio, CR = 0.074. As CR < 0.1 the level of inconsistency
method. present in the information stored in comparison matrix is
satisfactory.
Therefore, in the proposed model of allocating reliabil-  
ity at design phase, AHP based on linguistic variable weight 1 1 1
SDC = (4.05, 6.59, 11.2)  , ,
method has been used to overcome some of the above short- 47.46 31.2 28.8
comings. Proposed method combines the spirit of AHP with = (0.09, 0.23, 0.58)
10 S. Chatterjee et al.

Table 3. Aggregated fuzzy comparison matrix of the attributes for fuzzy AHP model.

Attributes DC DM DI LM EM

DC (1,1,1) (0.8,1.6,2.3) (0.65,1.09,1.9) (0.8,1.5,2.8) (0.8,1.4,3.2)


DM (0.44,0.6,1.12) (1,1,1) (2.02,3.1,4.7) (0.8,1,1.5) (1.2,2.36,4.5)
DI (3.23,4.4,6.4) (0.22,0.34,0.6) (1,1,1) (0.68,1.11,1.7) (0.8,1,1.72)
LM (0.36,0.7,1.21) (0.68,1,1.3) (0.6,0.9,1.5) (1,1,1) (0.76,0.93,1.25)
EM (0.31,0.73,1.3) (0.22,0.42,0.49) (0.6,1,1.26) (0.8,1.1,1.3) (1,1,1)

 
1 1 1 The minimum degree of possibility of superiority of each
SDM = (5.46, 8.06, 12.82)  , , criterion over another is obtained. This further decides the
47.46 31.2 28.8
= (0.11, 0.26, 0.45) weight vectors of the criteria. Therefore, the weight vector
  is given as:
1 1 1
SDI = (5.93, 7.85, 11.42)  , ,
47.46 31.2 28.8
W = (0.86, 1, 0.96, 0.41, 0.41),
= (0.123, 0.252, 0.396)
 
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

1 1 1 and the normalised weight vector can be given as:


SLM = (3.4, 4.53, 6.26)  , ,
47.46 31.2 28.8
= (0.072, 0.145, 0.217) W = (0.260, 0.285, 0.155, 0.158, 0.142).
 
1 1 1
SEM = (2.93, 4.25, 5.76)  , ,
47.46 31.2 28.8 The global weight of each criterion is 0.2.
= (0.062, 0.132, 0.2). Case B:The procedure for the AHP is given as follows:
Let, (aij )n n be the aggregated comparison or a judge-

The degree of possibility of superiority of DC can be cal- ment matrix with entries aij = 1/aji . Then, yk = i aij .
culated as follows: The geometric mean of the elements of each row can be
1
constructed bk = [(ak1 ).(ak2 ) . . . (akn )] n and Xk = bkbk .
k
V (SDC SDM ) = 0.90, V (SDC SDI ) = 0.0.86, Hence, the normalised weight vector is given as W = (X1 ,
V (SDC SLM ) = 1, and V (SDC SEM ) = 1. X2 , . . .Xn ).
The aggregated pairwise crisp comparison matrix in
The degree of possibility of superiority of DM can be cal- case of AHP is given in Table 4.
culated as follows: The maximum value of eigenvector for the above ma-
n)
trix is max = 5.0527. Consistency index, CI = (max n1
=
V (SDM SDI ) = 1, V (SDM SLM ) = 1, 0.013. Random Index for the matrix of order 5 is RI = 1.12.
V (SDM SEM ) = 1, and V (SDM SDC ) = 1. Consistency Ratio, CR = CI/RI = 0.012.CR < 0.1, i.e., the
level of inconsistency present in the information stored in
The degree of possibility of superiority of DI can be calcu- comparison matrix is satisfactory (Saaty 1980). Now, y1 =
lated as follows: 3.713, y2 = 3.901, y3 = 6.938, y4 = 5.79 and y5 = 7.857.
Hence
V (SDI SDC ) = 1, V (SDI SDM ) = 0.96,
b1 = [1 1.58 1.076 1.626 1.843]1/5 = 1.385
V (SDI SLM ) = 1, and V (SDI SEM ) = 1.
b2 = [0.625 1 3.294 1.104 2.784]1/5 = 1.445
The degree of possibility of superiority of LM can be cal-
culated as follows:
Table 4. Aggregated comparison matrix of the attributes for
V (SLM SDC ) = 0.6, V (SLM SDM ) = 0.41, AHP model.
V (SLM SDI ) = 0.71, and V (SLM SEM ) = 1. Attributes DC DM DI LM EM Weight

The degree of possibility of superiority of EM can be cal- DC 1 1.58 1.076 1.626 1.843 0.271
culated as follows: DM 0.625 1 3.294 1.104 2.784 0.283
DI 0.93 0.242 1 1.116 1.17 0.153
LM 0.615 0.72 0.713 1 1.06 0.157
V (SEM SDC ) = 0.53, V (SEM SDM ) = 0.41, EM 0.543 0.359 0.855 0.944 1 0.135
V (SEM SDI ) = 0.55, and V (SEM SLM ) = 0.90.
International Journal of Systems Science 11

b3 = [0.93 0.242 1 1.116 1.17]1/5 = 0.783 7. Conclusion


In this paper, the software reliability allocation problem at
b4 = [0.615 0.72 0.713 1 1.06]1/5 = 0.803
the planning and design stage of a software has been dis-
b5 = [0.543 0.359 0.855 0.944 1]1/5 = 0.691. cussed. The users views about the software functions and
their relative importance to the user have been integrated
Using the above values of bi , i = 1, 2, 3, 4, 5, the com- with the software engineers and programmers view about
puted values of Xi are: X1 = 1.385/5.107 = 0.271, X2 = software in terms of programmes and module. A hierarchi-
1.445/5.107 = 0.283, X3 = 0.783/5.107 = 0.153, X4 = cal structure has been set up to break down the problem
0.803/5.107 = 0.157, X5 = 0.691/5.107 = 0.135. The nor- from top to bottom. Then the fuzzy analytic hierarchy pro-
malised weight vector is given as: cess (FAHP) has been applied to derive the required pa-
rameters from this hierarchy. A comparison between AHP
W = (0.271, 0.283, 0.153, 0.157, 0.135). and FAHP has been carried out based on consistency ratio.
Also, sensitivity analysis has been performed to demon-
strate the performance of the proposed approach. The re-
The weights of the criteria represent the ratio of how
liability allocation model presented in this paper guaran-
much more important one criterion is than another, with
tees about the achievement of software system reliability
respect to the goal or criterion at a higher level.
goal. This method improves communication among users,
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

software engineers and programmers for reliability allo-


6.1.1. Decision cation. For the software having complex architecture, the
proposed method can be modified and used for its reliability
Comparing the weight vectors of the FAHP and AHP, it has allocation.
been found that DM is the component of highest priority in
both the cases. The result depicts that the classical and the
fuzzy methods are not the competitors to each others under
the same conditions. Most importantly, if the information is Acknowledgements
certain, i.e., the consistency ratio (CR) or the level of incon- The authors acknowledge Indian School of Mines, Dhanbad, In-
sistency present in the information stored in the comparison dia, for providing necessary facilities for completion of this work.
matrix is less, classical method should be preferred. If the Also, the authors are very much thankful to the reviewers and
level of inconsistency present in the information stored in Editor-in-Chief for their valuable comments in improving the
paper.
comparison matrix is high, i.e., the consistency ration is
large, fuzzy method should be preferred.

Notes on contributors
6.1.2. Sensitivity analysis Subhashis Chaterjee obtained his BSc
(Mathematics) from TDB College,
From the numerical example given in Section 6.1, it is Raniganj, The University of Burdwan,
clear that the quality of a module is highly dependent on India. He obtained MSc (Mathematics)
the different given quality criteria. Small changes in the and PhD from IIT Kharagpur, India. His
relative weights can therefore cause major changes in the area of research is software reliability
final ranking. Since, these weights are usually based on modelling. Presently Dr. Chatterjee is
working as Associate Professor, in the
highly subjective judgements, the stability of the ranking Dept. of Applied Mathematics, Indian
under varying criteria weights has to be tested. For this School of Mines (ISM), Dhanbad, India. He has served GIET,
purpose, sensitivity analysis can be performed based on Orissa and SMIT, Sikkim, India, as a faculty. He has total 11 years
scenarios that reflect alternative future developments or of teaching and research experience. He has quite a good number
different views on the relative importance of the criteria. of international and national publications. He has reviewed
papers for various national and international journals. His areas
From the normalised weight vector of different criteria of of interest are software reliability, web software reliability, or,
a module of the software proposed in the example, i.e., W stochastic process and fuzzy set.
= (0.260, 0.285, 0.155, 0.158, 0.142), it can be observed
that the priority of the second criterion DM is highest. If its Jeetendra B. Singh received his MSc de-
weight increases by 30%, the global weight of each module gree in Mathematics in 2007 from Banaras
will increase from 0.2 to 0.2168. It indicates that when the Hindu University, Varanasi, India. He has
demand of the modules increases, their global weight also obtained his PhD degree in Applied Mathe-
increases. Similarly, if the weight of DC is increased by matics from Indian School of Mines, Dhan-
bad, India. His research interests include
30%, the global weight increases from 0.2 to 0.2156, which software reliability modelling, time series,
signifies that, if the development cost increases, the global fuzzy time series, genetic algorithm and ar-
weight of the modules increases. tificial neural network.
12 S. Chatterjee et al.

Arunava Roy is currently working as a Se- Kapur, P.K., Pham, H., Aggarwal, A.G., and Kaur, G. (2012),
nior Research Fellow in the Department Two Dimensional Multi-Release Software Reliability Mod-
of Applied Mathematics, Indian school of eling and Optimal Release Planning, IEEE Transactions on
Mines, Dhanbad. He did his MSc in Mathe- Reliability, 61(3), 758768.
matics and Computing from Indian School Kaufmann, A., and Gupta, M.M. (1991), Introduction to Fuzzy
of Mines Dhanbad in 2010. He was a Arithmetic Theory and Applications, New York: Van Nostrand
gold medallist in MSc. His areas of in- Reinhold.
terest are web software reliability, algo- Kubat, P. (1989), Assessing Reliability of Modular Software,
rithm design and analysis, data structure and Operations Research Letters, 8, 3541.
programming. Lee, H. (1993), A Structure Methodology for Software Develop-
ment Effort Prediction Using the Analytic Hierarchy Process,
Journal of System and Software, 21, 179186.
Lee, S.H. (2010), Using Fuzzy AHP to Develop Intellectual Cap-
References ital Evaluation Model for Assessing Their Performance Con-
Aggarwal, K.K., and Singh, Y. (1995), Software Reliability Ap- tribution in a University, Expert Systems with Applications,
portionment Using the Analytic Hierarchy Process, ACM 37, 49414947.
SIGSOFT Software Engineering Notes, 20(5), 5661. Leung, Y.W. (1997), Software Reliability Allocation Under an
Buckley, J.J. (1985), Fuzzy Hierarchical Analysis, Fuzzy Sets Uncertain Operational Profile, Journal of the Operational
and Systems, 17, 233247. Research Society, 48, 401411.
Cai, K.Y., Wen, C.Y., and Zhang, M.L. (1991), A Critical Review Leung, Y.W. (2004), Optimal Reliability Allocation for Modu-
of Software Reliability Modeling, Reliability Engineering lar Software Systems Designed for Multiple Customers, IE-
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

and System Safety, 32, 357371. ICE Transactions on Information and Systems, 12(79), 1655
Cai, K.Y., Wen, C.Y., and Zhang, M.L. (1993), A Novel Approach 1662.
to Software Reliability Modeling, Microelectronic and Reli- Lyu, M.R. (1996), Handbook of Software Reliability Engineering,
ability, 33, 22562267. New York: IEEE Computer Society Press McGraw Hill.
Celik, M., Er, I.D., and Ozok, A.F. (2009), Application of Fuzzy Lyu, M.R., Rangarajan, S., and van Moorsel, A.P.A. (1997). Op-
Extended AHP Methodology on Shipping Registry Selection: timization of Reliability Allocation and Testing Schedule
The Case of Turkish Maritime Industry, Expert Systems with for Software Systems, in Proceedings of the 1997 Interna-
Applications, 36, 190198. tional Symposium on Software Reliability Engineering, Nov.,
Chang, D.Y. (1996), Applications of Extent Analysis Method on pp. 336346.
Fuzzy AHP, European Journal of Operational Research, 95, Mikaeil, R., Yousefi, R., and Ataei, M. (2011), Sawability Rank-
649655. ing of Carbonate Rock Using Fuzzy Analytic Hierarchy Pro-
Chatterjee, S., Misra, R.B., and Alam, S.S. (1997), Prediction cess and TOPSIS Approaches, Scientia Iranica, 18(5), 1106
of Software Reliability Using an Auto Regressive Process, 1115.
International Journal of Systems Science, 28, 205211. Mikhailov, L. (2003), Deriving Priorities From Fuzzy Pairwise
Cheng, C.H., Yang, K.L., and Hwang, C.L. (1999), Evaluating Comparison Judgements, Fuzzy Sets and Systems, 134, 365
Attack Helicopters by AHP Based on Linguistic Variable 385.
Weight, European Journal of Operational Research, 116, Mikhailov, L. (2004), A Fuzzy Approach to Deriving Priorities
423435. From Interval Pairwise Comparison Judgments, European
Guan, H., Wang, T., and Chen, W. (2009), Exploring Architecture- Journal of Operational Research, 159(3), 687704.
Based Software Reliability Allocation Using a Dynamic Pro- Nakagawa, Y., and Miyazaki, S. (1981), Surrogate Constraints
gramming Algorithm, in Proceedings of the Second Sym- Algorithm for Reliability Optimization Problems With Two
posium International Computer Science and Computational Constraints, IEEE Transactions on Reliability, 30(2), 175
Technology (ISCSCT 09), Huangshan, P. R. China, pp. 106 181.
109. Painton L., and Campbell, J. (1995), Genetic Algorithms in Op-
Helander, M.E., Zhao, M., and Ohlsson, N. (1998), Planning timization of System Reliability, IEEE Transactions on Re-
Models for Software Reliability and Cost, IEEE Transactions liability, 44(2), 172178.
on Software Engineering, 24(6), 424434. Pham, H. (1996), A Software Cost Model With Imperfect De-
Huang, C.Y., and Lo, J.H. (2006), Optimal Resource Allocation bugging, Random Life Cycle and Penalty Cost, International
for Cost and Reliability of Modular Software Systems in the Journal of Systems Science, 27(5), 455463.
Testing Phase, Journal of Systems and Software, 79(5), 653 Pham, H. (2006), System Software Reliability, London: Springer.
664. Pietrantuono, R., Russo, S., and Trivedi, K.S. (2010), Software
Huang, C.Y., and Lyu, M.R. (2005), Optimal Testing Resource Reliability and Testing Time Allocation: An Architecture-
Allocation and Sensitivity Analysis in Software Develop- Based Approach, IEEE Transactions on Software Engineer-
ment, IEEE Transactions on Reliability, 54(4), 592603. ing, 36(3), 323337.
Ishii, T., Dohi, T., and Okamura, H. (2012), Software Reliabil- Pievatolo, A., Ruggeri, F., and Soyer, R. (2012), A Bayesian
ity Prediction Based on Least Squares Estimation, Quality Hidden Markov Model for Imperfect Debugging, Reliability
Technology and Quantitative Management, 9(3), 243264. Engineering & System Safety, 103, 1121.
Junhong, G., Xiaozong Y., and Hongwei, L. (2005), Software Poore, J.H., Mills, H.D., and Mutchler, D. (1993), Planning
Reliability Nonlinear Modeling and Its Fuzzy Evaluation, in and Certifying Software System Reliability, IEEE Software,
4th WSEAS International Conference on Non-Linear Analy- 10(1), 8899.
sis, Non-Linear Systems and Chaos, Sofia, Bulgaria, October Poore, J.H., and Trammell, C.J. (1995), Cleanroom Software En-
2729, 2005, pp. 4954. gineering: A Reader, Oxford: Blackwell.
Kapur, P.K., Garg, R.B., and Kuamr, S. (1999), Contribution to Rani, I., and Misra, R.B. (2005), Economic Allocation of Tar-
Hardware and Software reliability, Singapore: World Scien- get Reliability in Modular Software Systems, in Proceed-
tific Press.
International Journal of Systems Science 13

ings of the Annual Reliability and Maintainability Symposium, Analytic Hierarchy Process, Fuzzy Sets and Systems, 157,
pp. 428432. 30553071.
Reza, M., Mohammad, A., and Reza, Y. (2011), Application of Xie, M. (1991), Software Reliability Modelling, Singapore: World
a Fuzzy Analytical Hierarchy Process to the Prediction of Scientific Press.
Vibration During Rock Sawing, Mining Science and Tech- Yamada, S., Tokuno, K., and Osaki, S. (1992), Imperfect Debug-
nology (China), 21, 611619. ging Models With Fault Introduction Rate for Software Relia-
Saaty, T.L. (1980), The Analytic Hierarchy Process: Planning, Pri- bility Assessment, International Journal of Systems Science,
ority Setting, Resources Allocation, London: McGraw-Hill. 23(12), 22412252.
Tamura, Y., and Yamada, S. (2009), Optimisation Analysis for Re- Yi, W., and Chengwen, W.U. ( 2009), Optimal Reliability Alloca-
liability Assessment Based on Stochastic Differential Equa- tion for Modular Software Systems Basis on Support Vector
tion Modelling for Open Source Software, International Clustering and Fuzzy Decision, in International Conference
Journal of Systems Science, 40(4), 429438. on Artificial Intelligence and Computational Intelligence, pp.
Tillman, F.A., Hwang, C.L., and Kuo, W. (1977), Determining 515519.
Component Reliability and Redundancy for Optimum System Zadeh, L.A. (1965), Fuzzy Sets, Information Control, 8, 338
Reliability, IEEE Transactions on Reliability, 26, 162165. 353.
Van Laarhoven, P.J.M., and Pedrycz, W. (1983), A Fuzzy Ex- Zahedi, F., and Ashrafi, N. (1991), Software Reliability Allo-
tension of Saatys Priority Theory, Fuzzy Sets and Systems, cation Based on Structure, Utility, Price, and Cost, IEEE
11(13), 229241. Transactions on Software Engineering, 17(4), 345356.
Wang, L., Chu, J., and Wu, J. (2007), Selection of Optimum Main- Zhang, X., and Pham, H. (2000), Comparisons of Nonhomo-
tenance Strategies Based on a Fuzzy Analytic Hierarchy Pro- geneous Poisson Process Software Reliability Models and
cess, International Journal of Production Economics, 107, Its Applications, International Journal of Systems Science,
Downloaded by [Indian School of Mines] at 06:38 03 June 2013

151163. 31(9), 11151123.


Wang, Y.M., Elhag, T.M.S., and Hua, Z.S. (2006), A Mod- Zimmermann, H.J. (1991), Fuzzy Set Theory and Its Applications,
ified Fuzzy Logarithmic Least Squares Method for Fuzzy Boston, MA: Kluwer Academic Publishers.

View publication stats

You might also like