Applied Mathematics and Computation 180 (2006) 217–232

www.elsevier.com/locate/amc

Parametric programming approach for batch arrival queues

with vacation policies and fuzzy parameters
a,*
Jau-Chuan Ke , Hsin-I Huang b, Chuen-Horng Lin b

a
Department of Statistics, National Taichung Institute of Technology, Taichung 404, Taiwan, ROC
b
Department of Information Management, National Taichung Institute of Technology, Taichung 404, Taiwan, ROC

Abstract

This work constructs the membership functions of the system characteristics of a batch arrival queuing model with
vacation policies and fuzzy batch arrival, customer service and server vacation rates. The a-cut approach is used to trans-
form a fuzzy queue into a family of conventional crisp queues in this context. By means of the membership functions of the
system characteristics, a set of parametric nonlinear programs is developed to describe the family of crisp queues with a
vacationing server. Two numerical examples are solved successfully to illustrate the validity of the proposed approach.
Because the system characteristics are expressed and governed by the membership functions, more information is provided
for use by management. By extending this model to the fuzzy environment, fuzzy queues with a vacationing server are rep-
resented more accurately and the analytic results are more useful for system designers and practitioners.
Ó 2006 Published by Elsevier Inc.

Keywords: Fuzzy sets; Multiple vacation; Parametric nonlinear programming; Single vacation

1. Introduction

Vacation queuing systems in which the server is unavailable during non-deterministic intervals of time have
received considerable attention in literature. Levy and Yechiali [1] first proposed an M/G/1 queuing system
under a single vacation policy—i.e., under the assumption that the server takes exactly one vacation when
the system is empty. A study of the variations and extensions of this single vacation model was presented
by Doshi [2] and Takagi [3] and included numerous applications in the study of maintenance problems in pro-
duction/inventory schedules, computer networks and digital communication systems. The generality and flex-
ibility of these vacation models are useful in modeling many real-life situations (see [4]). For example, in most
computer systems, the processor is shared among various types of jobs and hence is not available to each job
type at all times. From the perspective of one job type, the processor alternates between handling its job type

*
Corresponding author.
E-mail address: jauchuan@ntit.edu.tw (J.-C. Ke).

0096-3003/$ - see front matter Ó 2006 Published by Elsevier Inc.

doi:10.1016/j.amc.2005.11.144
218 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

and other job types. To reflect the occasional unavailability of the processor in queuing systems, the server is
modeled as taking vacations (see [5]).
For queuing systems with batch inputs, Choudhury [6] successfully modeled a M[x]/G/1 queuing system
(where [x] represents random batch arrival size) with a single vacation policy, extending the results of Levy
and Yechiali [1] and Doshi [2]. The M[x]/G/1 queuing system with multiple vacations, where the server goes
on vacations repeatedly until it finds at least one waiting customer at the end of a vacation, was first studied
by Baba [7]. He derived the expected queue length, waiting time and busy period distributions through a sup-
plementary variable technique. Lee and Srinivasan [8] examined the control operating policy of Baba’s [7]
model using a general approach and presented applications in production/inventory systems and other areas.
Later, Lee et al. [9,10] analyzed in detail Lee and Srinivasan’s [8] system with single and multiple vacation pol-
icies. They also provided a probabilistic interpretation of the single (multiple) vacation system with a threshold
policy.
In the literature described above, the batch inter-arrival times, customer service times and server vacation
times are required to follow certain probability distributions. However, in many real-world applications, the
parameter distributions may only be characterized subjectively; that is, the arrival, service and vacation pat-
terns are typically described in everyday language summaries of central tendency, such as ‘‘the mean arrival
rate is around 5 per day’’, ‘‘the mean service rate is about 10 per hour’’ or ‘‘the mean vacation rate is approx-
imately 2 per week’’, rather than with complete probability distributions. In other words, these system param-
eters are both possibilistic and probabilistic. Thus, fuzzy queues are potentially much more useful and realistic
than the commonly used crisp queues (see [11,12]). By extending the usual crisp queues to fuzzy queues in the
context of a vacationing server, these queuing models become appropriate for a wider range of applications.
Li and Lee [11] investigated the analytical results for two typical fuzzy queues (denoted M/F/1/1 and FM/
FM/1/1, where F represents fuzzy time and FM represents fuzzified exponential distributions) using a general
approach based on Zadeh’s extension principle (see also [13,14]), the possibility concept and fuzzy Markov
chains (see [15]). A useful modeling and inferential technique would be to apply their approach to general
fuzzy queuing problems. However, their approach is complicated and not suitable for computational pur-
poses; moreover, it cannot easily be used to derive analytic results for other complicated queuing systems
(see [16]). In particular, it is very difficult to apply this approach to fuzzy queues with a vacationing server.
Negi and Lee [16] proposed a procedure using a-cuts and two-variable simulation to analyze fuzzy queues
(see also [17]). Unfortunately, their approach provides only crisp solutions; i.e., it does not fully describe
the membership functions of the system characteristics. Using parametric programming, Kao et al. [18] con-
structed the membership functions of the system characteristics for fuzzy queues and successfully applied them
to four simple fuzzy queue models: M/F/1/1, F/M/1/1, F/F/1/1 and FM/FM/1/1. Recently, Chen [19,20]
developed FM/FM/1/L and FM/FM[K]/1/1 fuzzy systems using the same approach.
All previous research on fuzzy queuing models is focused on ordinary queues with one or two fuzzy vari-
ables. In this paper, we develop an approach that provides system characteristics for queues with a vaca-
tioning server and three fuzzy variables: fuzzified exponential arrival, service and vacation rates. Through
a-cuts and Zadeh’s extension principle, we transform the fuzzy queues to a family of crisp queues. As a
varies, the family of crisp queues is described and solved using parametric nonlinear programming
(NLP). The NLP solutions completely and successfully yield the membership functions of the system char-
acteristics, including the expected waiting time in the queue, the expected number of customers in the system
and the expected lengths of time the server is idle and busy. Although an explicit closed-form expression for
the membership function is very difficult to obtain in the case of three fuzzy variables, we develop a char-
acterization that yields closed-form expressions when interval limits are invertible. Furthermore, this paper
extends the analysis of system characteristics to encompass other system indices that are useful in more real-
istic systems.
The remainder of this paper is organized as follows. Section 2 presents the system characteristics of stan-
dard and fuzzy queuing models with a vacationing server. In Section 3, a mathematical programming
approach is developed to derive the membership functions of these system characteristics. To demonstrate
the validity of the proposed approach, two realistic numerical examples are described and solved. Discussion
is provided in Section 4, and conclusions are drawn in Section 5. For notational convenience, our model in this
paper is hereafter denoted FM[x]/FM/1/FV, where FV represents the fuzzified exponential vacation rate.
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 219

2. Fuzzy queues with a vacationing server

2.1. M[x]/M/1/V queues

We consider a queuing system with two different vacation policies. Under Policy I (multiple vacations), the
server takes consecutive vacations when there are no customers queued for service, while under Policy II (sin-
gle vacation), the server takes only a single vacation. It is assumed that customers arrive in batches according
to a compound Poisson process with group P1arrival rate k. Each batch size, A, has a probability mass function
{rk : rk = Pr(A = k), k P 1}, and E½A ¼ k¼1 krk denotes the expectation of A. Customers arriving in batches
at the server form a single-file queue and are served in order. The service time is exponentially distributed with
rate l. When there are no customers in the queue, the server takes a vacation with vacation length exponen-
tially distributed with rate h. Define: Wq, the expected waiting time in the queue, Ns, the expected number of
customers in the system, E[I], the expected length of time the server is idle (including vacations), E[B], the
expected length of time the server is busy.
From the results in Takagi [3] and Choudhury [6], we can easily derive the system characteristics for Policies
I and II in terms of the system parameters:
q E½AðA  1Þ 1
Policy I: Wq ¼ þ þ ; ð1Þ
lð1  qÞ 2lE½Að1  qÞ h
 
q2 kE½AðA  1Þ 1 1
Ns ¼ þ þ kE½A þ ; ð2Þ
ð1  qÞ 2lð1  qÞ h l
1 1
E½I ¼ þ ; ð3Þ
k h
 
q 1 1
E½B ¼ þ ; ð4Þ
ð1  qÞ k h
q E½AðA  1Þ kðk þ hÞ
Policy II: Wq ¼ þ þ ; ð5Þ
lð1  qÞ 2lE½Að1  qÞ hðh2 þ kh þ k2 Þ
q2 kE½AðA  1Þ k2 E½Aðk þ hÞ kE½A
Ns ¼ þ þ 2 2
þ ; ð6Þ
ð1  qÞ 2lð1  qÞ hðh þ kh þ k Þ l
h 1
E½I ¼ þ ; ð7Þ
kðk þ hÞ h
 
E½A h k
E½B ¼ þ ; ð8Þ
lð1  qÞ k þ h h
where q = kE[A]/l. In steady-state, it is necessary that we have 0 < hð1qÞ
kþh
< 1 and 0 < hðkþhÞð1qÞ
kðkþhÞþh2
< 1 for Policies
I and II, respectively.

2.2. FM[x]/FM/1/FV queues

To extend the applicability of the standard queuing model with a vacationing server, we allow for fuzzy
specification of system parameters. Suppose the arrival rate k, service rate l and vacation rate h are approx-
imately known and can be represented by the fuzzy sets ~k, l ~ and ~h, respectively. Let g~k ðxÞ, gl~ ðyÞ and g~h ðvÞ
denote the membership functions of ~ ~ and ~
k, l h, respectively. We then have the following fuzzy sets:
~
k ¼ fðx; g~k ðxÞÞjx 2 X g; ð9aÞ
~ ¼ fðy; gl~ ðyÞÞjy 2 Y g;
l ð9bÞ
~
h ¼ fðv; g~h ðvÞÞjv 2 V g; ð9cÞ
where X, Y and V are the crisp universal sets of the arrival, service and vacation rates, respectively.
220 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

Let f(x, y, v) denote the system characteristic of interest. Since ~k, l

~ and ~h are fuzzy numbers, f ð~k; l
~; ~hÞ is also
a fuzzy number. Following Zadeh’s extension principle (see [12,13]), the membership function of the system
characteristic f ð~ ~; ~
k; l hÞ is defined as
gf ð~k;~l;~hÞ ðzÞ ¼ sup minfg~k ðxÞ; gl~ ðyÞ; g~h ðvÞjz ¼ f ðx; y; vÞg; ð10Þ
X
n  o

where the supremum is taken over the set X ¼ x 2 X ; y 2 Y ; v 2 V 0 < vðyxE½AÞ < 1 for Policy I and
n  o yðxþvÞ
 vðxþvÞðyxE½AÞ
X ¼ x 2 X ; y 2 Y ; v 2 V 0 < y½xðxþvÞþv2  < 1 for Policy II.
Assume that the system characteristic of interest is Wq, the expected waiting time in the queue. It follows
from (1) that the expected waiting time under Policy I is
xE½A E½AðA  1Þ 1
f ðx; y; vÞ ¼ þ þ . ð11Þ
yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v

The membership function for the expected waiting time in the queue under Policy I is
  
 xE½A E½AðA  1Þ 1

gW~ q ðzÞ ¼ sup min g~k ðxÞ; gl~ ðyÞ; g~h ðvÞz ¼ þ þ . ð12aÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
Likewise, membership functions for the expected number of customers in the system and the expected lengths
of time the server is idle and busy are
(  )
 ðxE½AÞ
2
xE½AðA  1Þ xE½A

gN~ s ðzÞ ¼ sup min g~k ðxÞ; gl~ ðyÞ; g~h ðvÞz ¼ þ þ ; ð12bÞ
X  yðy  xE½AÞ 2ðy  xE½AÞ v
  
 1 1
~ ðzÞ ¼ sup min g~
gE½I ðxÞ; g ðyÞ; g ðvÞ z ¼ þ ; ð12cÞ
X
k l
~ ~
h  x v
and
  
 ðx þ vÞE½A
~ ðzÞ ¼ sup min g~
gE½B ðxÞ; g ðyÞ; g ðvÞ  z ¼ . ð12dÞ
X
k l
~ ~
h  vðy  xE½AÞ
The membership functions of the system characteristics under Policy II can be expressed in a similar manner.
Unfortunately, these membership functions are not expressed in the usual forms, making it very difficult to
imagine their shapes. In this paper we approach the representation problem using a mathematical program-
ming technique. Parametric NLPs are developed to find the a-cuts of f ð~k; l ~; ~hÞ based on the extension
principle.

3. The solution procedure

~ q (for Policy I) in an understandable and usable form,

To re-express the membership function gW~ q ðzÞ of W
we adopt Zadeh’s approach, which relies on a-cuts of W ~ q . Definitions for the a-cuts of ~k, l
~ and ~h as crisp inter-
vals are as follows:
 
kðaÞ ¼ xLa ; xU
a ¼ minfxjg~
k ðxÞ P ag; maxfxjg~
k ðxÞ P ag ; ð13aÞ
x2X x2X

 
lðaÞ ¼ y La ; y U
a ¼ minfyjgl
~ ðyÞ P ag; maxfyjgl
~ ðyÞ P ag ; ð13bÞ
y2Y y2Y

 
hðaÞ ¼ vLa ; vU
a ¼ minfvjg~
h ðvÞ P ag; maxfvjg~
h ðvÞ P ag . ð13cÞ
v2V v2V

The constant arrival, service and vacation rates are shown as intervals when the membership functions are no
less than a given possibility level for a. As a result, the bounds of these intervals can be described as functions
of a and can be obtained as: xLa ¼ min g1 ~
k
ðaÞ, xUa ¼ max g~
1
k
ðaÞ, y La ¼ min gl1 U 1
~ ðaÞ, y a ¼ max gl
~ ðaÞ,
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 221

vLa ¼ min g1 ðaÞ and vU 1 ~ q to construct its membership

~
h a ¼ max g~ h
ðaÞ. Therefore, we can use the a-cuts of W
function since the membership function defined in (12a) is parameterized by a.
Using Zadeh’s extension principle, gW~ q ðzÞ is the minimum of g~k ðxÞ, gl~ ðyÞ and g~h ðvÞ. To derive the member-
xE½A E½AðA1Þ 1
ship function gW~ q ðzÞ, we need at least one of the following cases to hold such that z ¼ yðyxE½AÞ þ 2ðyE½AxE 2 ½AÞ þ v

satisfies gW~ q ðzÞ ¼ a:

Case ðiÞ : ðg~k ðxÞ ¼ a; gl~ ðyÞ P a; g~h ðvÞ P aÞ;
Case ðiiÞ : ðg~k ðxÞ P a; gl~ ðyÞ ¼ a; g~h ðvÞ P aÞ;
Case ðiiiÞ : ðg~k ðxÞ P a; gl~ ðyÞ P a; g~h ðvÞ ¼ aÞ.
This can be accomplished using parametric NLP techniques. The NLP to find the lower and upper bounds of
the a-cut of gW~ q for Case (i) are
 
L1 xE½A E½AðA  1Þ 1
ðW q Þa ¼ min þ þ ; ð14aÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
 
U1 xE½A E½AðA  1Þ 1
ðW q Þa ¼ max þ þ ; ð14bÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
for Case (ii) are
 
L2 xE½A E½AðA  1Þ 1
ðW q Þa ¼ min þ þ ; ð14cÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
 
U2 xE½A E½AðA  1Þ 1
ðW q Þa ¼ max þ þ ; ð14dÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
and for Case (iii) are
 
L3 xE½A E½AðA  1Þ 1
ðW q Þa ¼ min þ þ ; ð14eÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
 
U xE½A E½AðA  1Þ 1
ðW q Þa 3 ¼ max þ þ . ð14fÞ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
From the definitions of k(a), l(a) and h(a) in (13), x 2 k(a), y 2 l(a) and v 2 h(a) can be replaced by
   L U  
x 2 xLa ; xU
a , y 2 ya ; ya and v 2 vLa ; vU
a , respectively. The ha-cuts i
formh a nested
i structure
h i with
h respect
i to a
L U L U L U L U
h(see [21,22]);
i h i.e.,i given 0 < a2 < a1 6 1, we have xa1 ; xa1  xa2 ; xa2 , y a1 ; y a1  y a2 ; y a2 and
vLa1 ; vU L U
a1  va2 ; va2 . Therefore, (14a), (14c) and (14e) have the same smallest element and (14b), (14d) and

(14f) have the same largest element. To find the membership function gW~ q , it suffices to find the left and right
L U
shape functions of gW~ q , which is equivalent to finding the lower bound ðW q Þa and upper bound ðW q Þa of the
~
a-cuts of W q , which, based on (12a), can be rewritten as
 
L xE½A E½AðA  1Þ 1
ðW q Þa ¼ min þ þ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
s.t. xLa 6 x 6 xU L U L U
a ; y a 6 y 6 y a and va 6 v 6 va ; ð15aÞ
 
U xE½A E½AðA  1Þ 1
ðW q Þa ¼ max þ þ
X yðy  xE½AÞ 2ðyE½A  xE2 ½AÞ v
s.t. xLa 6 x 6 xU L U L U
a ; y a 6 y 6 y a and va 6 v 6 va . ð15bÞ
At least one of x, y or v must hit the boundaries of their a -cuts to satisfy gW~ q ðzÞ ¼ a. This model is a set of
mathematical programs with boundary constraints and lends itself to the systematic study of how the optimal
solutions change with xLa , xU L U L U
a , y a , y a , va and va as a varies over (0, 1]. The model is a special case of parametric
NLPs (see [23]).
222 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

L U ~ q . Again, by applying the

The crisp interval ½ðW q Þa ; ðW q Þa  obtained from (15) represents the a-cuts of W
~ L L
results of Zimmermann [21] and Kaufmann [22] and convexity properties to W q , we have ðW q Þa1 P ðW q Þa2
U U L U
and ðW q Þa1 6 ðW q Þa2 , where 0 < a2 < a1 6 1. In other words, ðW q Þa increases and ðW q Þa decreases as a
increases. Consequently, the membership function gW~ q ðzÞ can be found from (15).
L 1
If both ðW q ÞLa and ðW q ÞU
a in (15) are invertible with respect to a, then a left shape function LðzÞ ¼ ½ðW q Þa 
U 1
and a right shape function RðzÞ ¼ ½ðW q Þa  can be derived, from which the membership function gW~ q is
constructed:
8 L L
< LðzÞ; ðW q Þa¼0 6 z 6 ðW q Þa¼1 ;
>
L U
gW~ q ðzÞ ¼ 1; ðW q Þa¼1 6 z 6 ðW q Þa¼1 ; ð16Þ
>
: U U
RðzÞ; ðW q Þa¼1 6 z 6 ðW q Þa¼0 .
L U
In most cases, the values of ðW q Þa and ðW q Þa cannot be solved analytically. Consequently, a closed-form
membership function for W ~ q cannot be obtained. However, the numerical solutions for ðW q ÞL and ðW q ÞU
a a
at different possibility levels can be collected to approximate the shapes of L(z) and R(z). That is, the set of
L U
intervals f½ðW q Þa ; ðW q Þa  j a 2 ½0; 1g shows the shape of gW~ q , although the exact function is not known
explicitly.
Note that the membership functions for other system characteristics under Policy I, such as the expected
number of customers, the expected lengths of time the server is idle and busy, and the system characteristics
under Policy II system can be derived in a similar manner.

4. Numerical examples

We consider two examples motivated by real-life systems to demonstrate the practical use of the proposed
approach.
Example 1 (The system under Policy I). Consider a local postal route in a central mail handling system. A
postal worker collects mail from mailboxes with fixed pick-up times and delivers the mail to the local post
office. The workers at the local post office collect mail up to a certain point and then send batches to the
central mail handling office. The number of parcels sent each time follows a geometric distribution with
parameter p = 0.5; i.e., the probability that A parcels are sent is Pr(A = k) = 0.5k, k = 1, 2, . . . One of the
workers at the central mail handling office sorts the parcels into categories including registered mail, ordinary
mail in a long flat envelope and ordinary mail in a standard envelope. This worker can be viewed as a server
and he or she can take consecutive vacations when there are no parcels to be sorted. Concerned with system
efficiency, management wants to know the system characteristics, including the expected waiting time in the
queue for a parcel, the expected number of parcels in the system and the expected lengths of time the server is
idle and busy.

~ q)
4.1. The fuzzy expected waiting time in the queue for a parcel (W

~
Suppose the arrival, service and vacation rates are trapezoidal fuzzy numbers represented  L U by k ¼ ½3; 4; 5; 6,
~ ¼ ½19; ~
l  20; 21; 22 and h ¼ ½1;  L8; 15;  22. First, it is easy to find that xa ; xa ¼U½3 þ a; 6L a,
y La ; y U
a ¼ ½19 þ a; 22  a and v a ; v U
a ¼ ½1 þ 7a; 22  7a. Next, it is obvious that when x ¼ xa , y ¼ y a and
v ¼ vLa , the waiting time attains its maximum value, and when x ¼ xLa , y ¼ y U U
a and v ¼ va , the waiting time
attains its minimum value. According to (15), the a-cuts of W ~ q are

L 4a2  256a þ 968

ðW q Þa ¼ ;
21a3
þ 640a2  4268a þ 7744
U 4a2 þ 280a þ 164
ðW q Þa ¼ .
21a3 þ 451a2 þ 995a þ 133
L U
With the help of MATLABÒ 7.0.4, the inverse functions of ðW q Þa and ðW q Þa exist, yielding the membership
function:
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 223
8
>
> 1 236
> LðzÞ;
> 6z6 ;
>
> 8 1365
<
236 11
gW~ q ðzÞ ¼ 1; 6z6 ;
>
> 1365 40
>
>
>
> 11 164
: RðzÞ; 6z6 ;
40 133
where
pffiffiffi 2 1 pffiffiffi
ð1 þ 3iÞP 3 þ 8ð1 þ 160zÞP 3  4ð1 þ 3iÞð4 þ 5312z þ 35179z2 Þ
LðzÞ ¼ 1 ;
126zP 3
2 1
Q3  ð4 þ 451zÞQ3 þ 4ð4 þ 5312z þ 35179z2 Þ
RðzÞ ¼ 1 ;
63zQ3
with
P ¼ 64 þ 127; 488z þ 13; 021; 788z2 þ 50; 119; 264z3
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ 756z 6752  11; 466; 240z  16; 087; 527z2 þ 75; 394; 000z3  480; 067; 500z4 ;
Q ¼ 64  127; 488z  13; 021; 788z2  50; 119; 264z3
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ 756z 6752  11; 466; 240z  16; 087; 527z2 þ 75; 394; 000z3  480; 067; 500z4
as shown in Fig. 1. The overall shape turns out as expected.
The membership functions L(z) and R(z) have complex values with their imaginary parts approaching zero
when 18 6 z 6 1365
236
for L(z) and 11
40
6 z 6 164133
for R(z). Hence, the imaginary parts of these two functions have no
influence on the computational results and can be disregarded.
Next, we perform a-cuts of arrival, service and vacation rates and fuzzy expected waiting time in the queue
at eleven distinct a values: 0.0, 0.1, 0.2, . . . , 1.0. Crisp intervals for fuzzy waiting time at different possibilistic a
levels are presented in Table 1. The fuzzy expected waiting time W ~ q has two characteristics to be noted. First,
the support of W~ q ranges from 0.125 to 1.233; this indicates that, though the expected waiting time is fuzzy, it

Fig. 1. The membership function for fuzzy expected waiting time in the queue for a parcel in Example 1.
224 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

Table 1
a-Cuts of arrival, service and vacation rates and expected waiting time in the queue for a parcel in Example 1
a xLa xU
a y La yU
a vLa vU
a ðW q ÞLa ðW q ÞU
a

0.00 3.00 6.00 19.00 22.00 1.00 22.00 0.1250 1.2331

0.10 3.10 5.90 19.10 21.90 1.70 21.30 0.1287 0.8099
0.20 3.20 5.80 19.20 21.80 2.40 20.60 0.1325 0.6277
0.30 3.30 5.70 19.30 21.70 3.10 19.90 0.1366 0.5239
0.40 3.40 5.60 19.40 21.60 3.80 19.20 0.1409 0.4555
0.50 3.50 5.50 19.50 21.50 4.50 18.50 0.1455 0.4062
0.60 3.60 5.40 19.60 21.40 5.20 17.80 0.1503 0.3686
0.70 3.70 5.30 19.70 21.30 5.90 17.10 0.1554 0.3385
0.80 3.80 5.20 19.80 21.20 6.60 16.40 0.1609 0.3138
0.90 3.90 5.10 19.90 21.10 7.30 15.70 0.1667 0.2929
1.00 4.00 5.00 20.00 21.00 8.00 15.00 0.1729 0.2750

is impossible for its values to fall below 0.125 or exceed 1.233. Second, the a-cut at a = 1 contains the values
from 0.173 to 0.275, which are the most possible values for the expected waiting time in the queue for a parcel.

~ s)
4.2. The fuzzy expected number of parcels in the system (N

~ s are
By the same argument and Eq. (2), the a-cuts of N
L 8a3  536a2 þ 400a þ 5808
ðN s Þa ¼ ;
21a3 þ 640a2  4268a þ 7744
8a3  608a2 þ 3032a þ 1968
ðN s ÞU
a ¼ .
21a3 þ 451a2 þ 995a þ 133
The membership function is
8
>
> 3 1888
>
> LðzÞ; 6z6 ;
>
> 4 1365
>
<
1888 11
gN~ s ðzÞ ¼ 1; 6z6 ;
>
> 1365 4
>
>
>
> 11 1968
>
: RðzÞ; 6z6 ;
4 133
where
pffiffiffi 2 1 pffiffiffi
ð1 þ 3iÞP 3 þ 8ð67 þ 80zÞP 3  ð1 þ 3iÞð74; 224 þ 190; 828z þ 35; 179z2 Þ
LðzÞ ¼ 1 ;
3ð8 þ 21zÞP 3
2 1
2Q3  ð608 þ 451zÞQ3 þ 2ð74; 224 þ 190; 828z þ 35; 179z2 Þ
RðzÞ ¼ 1 ;
3ð8 þ 21zÞQ3
with
P ¼ 19; 586; 368 þ 81; 994; 344z þ 55; 861; 509z2 þ 6; 264; 908z3
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ð1800  4725zÞ 7; 805; 312  23; 072; 736z þ 930; 177z2  529; 276z3  192; 027z4 ;
Q ¼ 19; 586; 368  81; 994; 344z  55; 861; 509z2  6; 264; 908z3
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ð1800  4725zÞ 7; 805; 312  23; 072; 736z þ 930; 177z2  529; 276z3  192; 027z4
as shown in Fig. 2. Table 2 reports a-cuts of arrival, service and vacation rates and the fuzzy expected number
of parcels in the queue for the same eleven selected a values. From Table 2, we gain insight into the possible
expected number of a parcel in the system.
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 225

Fig. 2. The membership function for fuzzy expected number of parcels in the system in Example 1.

Table 2
a-Cuts of arrival, service and vacation rates and expected number of parcels in the system in Example 1
a xLa xU
a y La yU
a vLa vU
a ðN s ÞLa ðN s ÞU
a

0.00 3.00 6.00 19.00 22.00 1.00 22.00 0.7500 14.7970

0.10 3.10 5.90 19.10 21.90 1.70 21.30 0.7978 9.5563
0.20 3.20 5.80 19.20 21.80 2.40 20.60 0.8483 7.2818
0.30 3.30 5.70 19.30 21.70 3.10 19.90 0.9017 5.9728
0.40 3.40 5.60 19.40 21.60 3.80 19.20 0.9583 5.1018
0.50 3.50 5.50 19.50 21.50 4.50 18.50 1.0183 4.4686
0.60 3.60 5.40 19.60 21.40 5.20 17.80 1.0821 3.9804
0.70 3.70 5.30 19.70 21.30 5.90 17.10 1.1501 3.5882
0.80 3.80 5.20 19.80 21.20 6.60 16.40 1.2226 3.2633
0.90 3.90 5.10 19.90 21.10 7.30 15.70 1.3001 2.9878
1.00 4.00 5.00 20.00 21.00 8.00 15.00 1.3832 2.7500

~
4.3. The fuzzy expected length of time the server is idle (including vacations) (E½I)

~ are
Using the same argument and Eq. (3), the a-cuts of E½I
L 8a þ 28
ðE½IÞa ¼ ;
7a2  64a þ 132
8a þ 4
ðE½IÞU
a ¼ .
7a2 þ 22a þ 3
The membership function is
8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> ð4 þ 32zÞ  2 4  15z þ 25z2 7 4
>
> ; 6z6 ;
>
> 7z 33 15
<
4 3
~ ðzÞ ¼
gE½I 1; 6z6 ;
>
> 15 8
>
> pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
: ð4  11zÞ þ 2 4  15z þ 25z ;
2 3 4
6z6
7z 8 3
226 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

as shown in Fig. 3. Crisp intervals for the fuzzy expected length of time the server is idle at different possibi-
listic a levels are provided in Table 3.

~
4.4. The fuzzy expected length of time the server is busy (E½B)

~
As defined in Eq. (4), the a-cuts of E½B are

L 50  12a
ðE½BÞa ¼ ;
21a2  178a þ 352
12a þ 14
ðE½BÞU
a ¼ .
21a2 þ 52a þ 7

Fig. 3. The membership function for fuzzy expected length of time the server is idle in Example 1.

Table 3
a-Cuts of arrival, service and vacation rates and expected length of time the server is idle in Example 1
a xLa xU
a y La yU
a ðE½IÞLa ðE½IÞU
a

0.00 3.00 6.00 1.00 22.00 0.2121 1.3333

0.10 3.10 5.90 1.70 21.30 0.2164 0.9108
0.20 3.20 5.80 2.40 20.60 0.2210 0.7292
0.30 3.30 5.70 3.10 19.90 0.2257 0.6256
0.40 3.40 5.60 3.80 19.20 0.2307 0.5573
0.50 3.50 5.50 4.50 18.50 0.2359 0.5079
0.60 3.60 5.40 5.20 17.80 0.2414 0.4701
0.70 3.70 5.30 5.90 17.10 0.2472 0.4398
0.80 3.80 5.20 6.60 16.40 0.2533 0.4147
0.90 3.90 5.10 7.30 15.70 0.2598 0.3934
1.00 4.00 5.00 8.00 15.00 0.2667 0.3750
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 227

The membership function is

8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> ð6 þ 89zÞ  36  18z þ 529z2 50 38
>
> ; 6z6 ;
>
> 21z 352 195
<
38 13
~ ðzÞ ¼
gE½B 1; 6z6 ;
>
> 195 40
>
> pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
> ð6  26zÞ þ 36  18z þ 529z2
> 13
: ; 6z62
21z 40
as shown in Fig. 4. Crisp intervals for the fuzzy expected length of time the server is busy at different possi-
bilistic a levels are given in Table 4.
Example 2 (The system under Policy II). Continuing Example 1, sorted mail at the central office is sent to an
automated distribution machine which dispenses the parcels into slots according to the destination post office.

Fig. 4. The membership function for fuzzy expected length of time the server is busy in Example 1.

Table 4
a-Cuts of arrival, service and vacation rates and expected length of time the server is busy in Example 1
a xLa xU
a y La yU
a vLa vU
a ðE½BÞLa ðE½BÞU
a

0.00 3.00 6.00 19.00 22.00 1.00 22.00 0.1420 2.0000

0.10 3.10 5.90 19.10 21.90 1.70 21.30 0.1459 1.2248
0.20 3.20 5.80 19.20 21.80 2.40 20.60 0.1500 0.8991
0.30 3.30 5.70 19.30 21.70 3.10 19.90 0.1544 0.7187
0.40 3.40 5.60 19.40 21.60 3.80 19.20 0.1591 0.6033
0.50 3.50 5.50 19.50 21.50 4.50 18.50 0.1640 0.5229
0.60 3.60 5.40 19.60 21.40 5.20 17.80 0.1693 0.4633
0.70 3.70 5.30 19.70 21.30 5.90 17.10 0.1750 0.4172
0.80 3.80 5.20 19.80 21.20 6.60 16.40 0.1811 0.3804
0.90 3.90 5.10 19.90 21.10 7.30 15.70 0.1877 0.3502
1.00 4.00 5.00 20.00 21.00 8.00 15.00 0.1949 0.3250
228 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

Viewing the distribution machine as a server, suppose that it can take only one vacation (say for calibration)
when there are no parcels to be dispensed.
Following the solution procedure illustrated above, we obtain membership functions as shown in Figs. 5–8
and crisp intervals for the expected waiting time in the queue for a parcel, the expected number of parcels in

Fig. 5. The membership function for fuzzy expected waiting time in the queue for a parcel in Example 2.

Fig. 6. The membership function for fuzzy expected number of parcels in the system for Example 2.
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 229

Fig. 7. The membership function for fuzzy expected length of time the server is idle in Example 2.

Fig. 8. The membership function for fuzzy expected length of time the server is busy in Example 2.

the system and the expected lengths of time the server is idle and busy for different possibilistic a levels in
Tables 5–8.
The results of Examples 1 and 2 show that the ranges of the membership functions under Policy I are wider
than those under Policy II for the expected number of parcels and the expected lengths of time the server is idle
230 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

Table 5
a-Cuts of arrival, service and vacation rates and expected waiting time in the queue for a parcel in Example 2
a xLa xU
a y La yU
a vLa vU
a ðW q ÞLa ðW q ÞU
a

0.00 3.00 6.00 19.00 22.00 1.00 22.00 0.0856 1.2098

0.10 3.10 5.90 19.10 21.90 1.70 21.30 0.0884 0.7742
0.20 3.20 5.80 19.20 21.80 2.40 20.60 0.0914 0.5827
0.30 3.30 5.70 19.30 21.70 3.10 19.90 0.0945 0.4721
0.40 3.40 5.60 19.40 21.60 3.80 19.20 0.0978 0.3989
0.50 3.50 5.50 19.50 21.50 4.50 18.50 0.1013 0.3464
0.60 3.60 5.40 19.60 21.40 5.20 17.80 0.1051 0.3069
0.70 3.70 5.30 19.70 21.30 5.90 17.10 0.1091 0.2759
0.80 3.80 5.20 19.80 21.20 6.60 16.40 0.1134 0.2509
0.90 3.90 5.10 19.90 21.10 7.30 15.70 0.1181 0.2303
1.00 4.00 5.00 20.00 21.00 8.00 15.00 0.1231 0.2130

Table 6
a-Cuts of arrival, service and vacation rates and expected number of parcels in the system for Example 2
a xLa xU
a y La yU
a vLa vU
a ðN s ÞLa ðN s ÞU
a

0.00 3.00 6.00 19.00 22.00 1.00 22.00 0.5139 14.5179

0.10 3.10 5.90 19.10 21.90 1.70 21.30 0.5483 9.1360
0.20 3.20 5.80 19.20 21.80 2.40 20.60 0.5849 6.7597
0.30 3.30 5.70 19.30 21.70 3.10 19.90 0.6238 5.3816
0.40 3.40 5.60 19.40 21.60 3.80 19.20 0.6652 4.4673
0.50 3.50 5.50 19.50 21.50 4.50 18.50 0.7094 3.8108
0.60 3.60 5.40 19.60 21.40 5.20 17.80 0.7568 3.3141
0.70 3.70 5.30 19.70 21.30 5.90 17.10 0.8075 2.9241
0.80 3.80 5.20 19.80 21.20 6.60 16.40 0.8620 2.6091
0.90 3.90 5.10 19.90 21.10 7.30 15.70 0.9209 2.3488
1.00 4.00 5.00 20.00 21.00 8.00 15.00 0.9845 2.1298

Table 7
a-Cuts of arrival, service and vacation rates and expected length of time the server is idle in Example 2
a xLa xU
a y La yU
a ðE½IÞLa ðE½IÞU
a

0.00 3.00 6.00 1.00 22.00 0.1764 1.0833

0.10 3.10 5.90 1.70 21.30 0.1797 0.7025
0.20 3.20 5.80 2.40 20.60 0.1831 0.5506
0.30 3.30 5.70 3.10 19.90 0.1866 0.4694
0.40 3.40 5.60 3.80 19.20 0.1903 0.4184
0.50 3.50 5.50 4.50 18.50 0.1942 0.3829
0.60 3.60 5.40 5.20 17.80 0.1983 0.3564
0.70 3.70 5.30 5.90 17.10 0.2025 0.3356
0.80 3.80 5.20 6.60 16.40 0.2070 0.3185
0.90 3.90 5.10 7.30 15.70 0.2117 0.3041
1.00 4.00 5.00 8.00 15.00 0.2167 0.2917

and busy. In contrast, the range of the membership function for the expected waiting time for a parcel is wider
under Policy II.
 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232 231

Table 8
a-Cuts of arrival, service and vacation rates and expected length of time the server is busy in Example 2
a xLa xU
a y La yU
a vLa vU
a ðE½BÞLa ðE½BÞU
a

0.00 3.00 6.00 19.00 22.00 1.00 22.00 0.1270 1.7551

0.10 3.10 5.90 19.10 21.90 1.70 21.30 0.1297 1.0121
0.20 3.20 5.80 19.20 21.80 2.40 20.60 0.1326 0.7130
0.30 3.30 5.70 19.30 21.70 3.10 19.90 0.1356 0.5547
0.40 3.40 5.60 19.40 21.60 3.80 19.20 0.1387 0.4580
0.50 3.50 5.50 19.50 21.50 4.50 18.50 0.1421 0.3935
0.60 3.60 5.40 19.60 21.40 5.20 17.80 0.1456 0.3475
0.70 3.70 5.30 19.70 21.30 5.90 17.10 0.1494 0.3132
0.80 3.80 5.20 19.80 21.20 6.60 16.40 0.1535 0.2866
0.90 3.90 5.10 19.90 21.10 7.30 15.70 0.1578 0.2654
1.00 4.00 5.00 20.00 21.00 8.00 15.00 0.1625 0.2481

5. Conclusions

This paper applies the concepts of a-cuts and Zadeh’s extension principle to a batch arrival queuing model
with a vacationing server and constructs membership functions of the expected waiting time in the queue, the
expected number of customers in the system and the expected lengths of time the server is idle and busy using
paired NLP models. Following the proposed approach, a-cuts of the membership functions are found and
their interval limits inverted to attain explicit closed-form expressions for the system characteristics. Even
when the membership function intervals cannot be inverted explicitly, system management or designers can
also specify the system characteristics of interest, perform numerical results to examine the corresponding
a-cuts and then use this information to develop or improve system processes.
For example, in Example 1, a manager can set the range of waiting time for a parcel to be [0.1366, 0.5239] to
reflect the desired service and vacation rates and find that the corresponding a level is 0.30 with y La ¼ 19:3,
yU L U
a ¼ 21:7, va ¼ 3:1 and va ¼ 19:9. Similarly, a manager can set the range of the number of parcels to be
[1.2226, 3.2633] to reflect the desired service and vacation rates and find that the corresponding a level is
0.80 with y La ¼ 19:8, y U L U
a ¼ 21:2, va ¼ 6:6 and va ¼ 16:4. In other words, the manager can determine that
the service rate is between 19.8 and 21.2 and the vacation rate is between 6.6 and 16.4. As this example dem-
onstrates, the approach proposed in this paper provides practical information for system managers and
practitioners.

Acknowledgement

The authors wish to thank R. Gulati for polishing the paper which improved its readability.

References

[1] Y. Levy, U. Yechiali, Utilization of idle time in an M/G/1 queueing system, Management Science 22 (1975) 202–211.
[2] B.T. Doshi, Queueing systems with vacations—a survey, Queueing Systems 1 (1986) 29–66.
[3] H. Takagi, Queueing analysis: a foundation of performance evaluation, Vacation and Priority Systems, Part I, vol. I, North-Holland,
Amsterdam, 1991.
[4] E. Gelenbe, I. Mitrani, Analysis and Synthesis of Computer Systems, Academic Press, New York, 1980.
[5] S. Kasahara, Y. Takahashi, T. Hasegawa, Analysis of waiting time of M/G/1/K system with vacations under random scheduling and
LCFS, Performance Evaluation 21 (1995) 239–259.
[6] G. Choudhury, A batch arrival queue with a vacation time under single vacation policy, Computers and Operations Research 29
(2002) 1941–1955.
[7] Y. Baba, On the M[x]/G/1 queue with vacation time, Operations Research Letters 5 (1986) 93–98.
[8] H.S. Lee, M.M. Srinivasan, Control policies for the M[x]/G/1 queueing system, Management Science 35 (1989) 708–721.
[9] S.S. Lee, H.W. Lee, K.C. Chae, On a batch arrival queue with N policy and single vacation, Computers and Operations Research 22
(1995) 173–189.
[10] H.W. Lee, S.S. Lee, J.O. Park, K.C. Chae, Analysis of M[x]/G/1 queue with N policy and multiple vacations, Journal of Applied
Probability 31 (1994) 467–496.
232 J.-C. Ke et al. / Applied Mathematics and Computation 180 (2006) 217–232

[11] R.J. Li, E.S. Lee, Analysis of fuzzy queues, Computers and Mathematics with Applications 17 (1989) 1143–1147.
[12] L.A. Zadeh, Fuzzy sets as a basis for a theory of possibility, Fuzzy Sets and Systems 1 (1978) 3–28.
[13] H.M. Prade, An outline of fuzzy or possibilistic models for queueing systems, in: P.P. Wang, S.K. Chang (Eds.), Fuzzy Sets, Plenum
Press, New York, 1980.
[14] R.R. Yager, A characterization of the extension principle, Fuzzy Sets and Systems 18 (1986) 205–217.
[15] R.E. Stanford, The set of limiting distributions for a Markov chain with fuzzy transition probabilities, Fuzzy Sets and Systems 7
(1982) 71–78.
[16] D.S. Negi, E.S. Lee, Analysis and simulation of fuzzy queue, Fuzzy Sets and Systems 46 (1992) 321–330.
[17] S. Chanas, M. Nowakowski, Single value simulation of fuzzy variable, Fuzzy Sets and Systems 21 (1988) 43–57.
[18] C. Kao, C.C. Li, S.P. Chen, Parametric programming to the analysis of fuzzy queues, Fuzzy Sets and Systems 107 (1999) 93–100.
[19] S.P. Chen, Parametric nonlinear programming for analyzing fuzzy queues with finite capacity, European Journal of Operational
Research 157 (2004) 429–438.
[20] S.P. Chen, Parametric nonlinear programming approach to fuzzy queues with bulk service, European Journal of Operational
Research 163 (2005) 434–444.
[21] H.J. Zimmermann, Fuzzy Set Theory and Its Applications, fourth ed., Kluwer Academic, Boston, 2001.
[22] A. Kaufmann, Introduction to the Theory of Fuzzy Subsets, vol. 1, Academic Press, New York, 1975.
[23] T. Gal, Postoptimal Analysis, Parametric Programming, and Related Topics, McGraw-Hill, New York, 1979.