Applied Mathematical Modelling 36 (2012) 255269

Contents lists available at ScienceDirect

Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm

A batch arrival retrial queue with general retrial times under

Bernoulli vacation schedule for unreliable server and delaying repair
Gautam Choudhury a,, Jau-Chuan Ke b
a
Mathematical Sciences Division, Institute of Advanced Study in Science and Technology, Paschim Boragaon, Guwahati 781035, Assam, India
b
Department of Applied Statistics, National Taichung Institute of Technology, No. 129, Section 3, Sanmin Road, Taichung 404, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Article history: This paper deals with the steady state behaviour of an Mx/G/1 queue with general retrial
Received 4 June 2009 time and Bernoulli vacation schedule for an unreliable server, which consists of a break-
Received in revised form 18 May 2011 down period and delay period. Here we assume that customers arrive according to com-
Accepted 22 May 2011
pound Poisson processes. While the server is working with primary customers, it may
Available online 16 July 2011
breakdown at any instant and server will be down for short interval of time. Further con-
cept of the delay time is also introduced. The primary customer nding the server busy,
Keywords:
down or vacation are queued in the orbit in accordance with FCFS (rst come rst served)
Stationary system size distribution
Random breakdown
retrial policy. After the completion of a service, the server either goes for a vacation of ran-
Delay time dom length with probability p or may continue to serve for the next customer, if any with
Repair time probability (1  p). We carry out an extensive analysis of this model. Finally, we obtain
Retrial time some important performance measures and reliability indices of this model.
Reliability indices 2011 Elsevier Inc. All rights reserved.

1. Introduction

Retrial queues (or queues with repeated attempts) are characterized by the feature that a customer who nds the server
busy upon arrival is obliged to leave the service area and repeat his demand after some time called retrial time. Between
trials, the blocked customer joins a pool of unsatised customers called orbit. Queues in which customers are allowed to
conduct retrials have been widely used to model many practical problems in telephone switching systems, telecommunica-
tion networks and computers competing to gain service from a central processing unit. Moreover, retrial queues are also
used as mathematical models for several computer systems: Packet switching networks, shared bus local area networks
operating under the carrier-sense multiple access protocol and collision avoidance star local area networks, etc. For a review
of main results and methods, the reader is referred to the survey papers by Yang and Templeton [1], Falin [2], Kulkarni and
Liang [3] and the book by Falin and Templeton [4]. For more recent references see the bibliographical overviews in [57].
Further, a comprehensive comparison between retrial queues and their standard counterpart with classical waiting line
can be found in Artalejo and Falin [8].
Many of the queueing systems with repeated attempts operate under the classical retrial policy, where each block of cus-
tomers generate a stream of repeated attempts independently of the rest of the customers in the orbit. However, there is a
second kind of policy, called constant retrial policy, which arises naturally in problems where the server is required to search
for customers (e.g., see [9]) and in communication protocols of type carrier sense multiple access (CSMA). The latter disci-
pline was introduced by Fayolle [10], who investigated an M/M/1 retrial queue in which the repeat customers from a queue
and only the head customers of the orbit queue can request a service after an exponentially distributed retrial time with

Corresponding author.
E-mail addresses: gc.iasst@gmail.com (G. Choudhury), jauchuan@ntit.edu.tw (J.-C. Ke).

0307-904X/$ - see front matter 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2011.05.047
256 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

some parameter e (say), i.e., retrial rate is (1  d0,n)e (where di,j denotes Kroneckers delta), when number of units in the orbit
is n. Farahmand [11] called this discipline a retrial queue with FCFS orbit retrial policy. Choi et al. [12] generalized this retrial
policy by considering an M/M/1 retrial queue with general retrial times.
Retrial queueing systems with general service times and non-exponential retrial times have also received considerable
attention during last decade. The rst investigation with general retrial time done by Kapyrin [13], who assumed that each
customer in the orbit generates a stream of repeated attempts that are independent of the customer in the orbit and state of
the server. However, this methodology was found to be incorrect by Falin [14]. Subsequently, Yang et al.[15] have developed
an approximation method to obtain the steady state performance for the model of Kapyrin. Later, Gomez-Corral [51] dis-
cussed extensively an M/G/1 retrial queue with FCFS discipline and general retrial times. In recent years, several retrial mod-
els have been analyzed with general retrial times, details of which may be found in [9,1720].
A wide class of retrial policies for governing the vacation mechanism has also been discussed in the literature. Most of the
analysis for retrial queues concerns the exhaustive service schedule [21], gated service policy [22] and recently modied
vacation policy [23]. Keilson and Servi [24] introduced Bernoulli vacation schedule: if the queue is empty after service com-
pletion then the server becomes inactive, i.e., begins a vacation period. If the queue is not empty then service begins with
specic probability q or a vacation period begins with probability p = 1  q. At the end of a vacation period service begins
if a customer is present in the queue. Other wise, the server waits for the rst customer to arrive. A number of papers
[13,2527] have recently appeared in the queueing literature in which concept of general retrial times have been introduced
along with Bernoulli vacation schedule under the FCFS orbit retrial policy. Such type of queueing models occur in many real
life situations where the server may be used for other secondary jobs, for instance to serve customers in other systems.
Allowing server to take vacations makes the queueing model more realistic and exible in studying real word queueing sit-
uations. Applications arise naturally in call centers with multi task employees, customized manufacturing, telecommunica-
tion and computer networks, maintenance activities, production and quality control problems, etc.
The study of queueing models with service interruptions goes back to the 1950s. Among some early papers on service
interruptions, we refer the readers to see those by Gaver [28], Avi-ltzhak and Naor [29], Thirurengadan [30] and Mitrany
and Avi-ltzhak [31] for some fundamental works. While Li et al. [32], Sengupta [33], Takin and Sengupta [34] and Tang
[35], among others have studied some queueing systems with interruptions wherein one of the underlying assumption is
that, the service channel undergoes repair instantaneously, as soon as it fails. On the other hand retrial queues that take into
account servers failures and repairs were introduced by Aissani [36] and Kulkarni and Choi [37]. Wang et al. [38] studied a
repairable M/G/1 retrial queueing model from the viewpoint of reliability for the rst time. Recently, Choudhury and Deka
[39] investigate such a repairable M/G/1 retrial queueing model with two phases of service.
On the other hand the batch arrival retrial queue was rst considered by Falin [40], who assumed the following operating
rule: If the server is busy at the arrival epoch, then whole batch joins the retrial group, whereas if the server is free, then one of the
arriving units starts its service and the rest join the retrial group. In this connection, its applicability is connected with the per-
formance evaluation of Local Area Networks operating under transmission protocols like the CSMA/CD (carrier sense multiple
access with collision detection), see Choi et al. [12]. In such a context, messages of variable length arrive at the stations and, then
they are divided into a number of packets in order to be transmitted to the destination station. If the transmission median (i.e., a
bus in the engineering terminology) is idle, then one packet is selected to be transmitted automatically and the rest is stored in a
buffer (i.e., the retrial group). On the other hand, if the bus is busy then the entire packet must be stored in the buffer and the
station will retry the transmission later on. A more complete description of this mechanism can be found in Yang and Temple-
ton [1]. Some recent papers [4143,26] discussed more complicated queueing situations with retrials and batch arrivals, where
concept of repeated attempts with Bernoulli vacation schedule is also introduced under different retrial policy.
Although some aspects have been discussed separately on queueing systems with service interruptions, Bernoulli sche-
dule vacation, repeated attempts with general retrial time, however, no work have been found that combine these features
together for unreliable server queueing systems with batch arrivals, even in the most recent studies. Moreover, another
important characteristic for considering the retrial model with general retrial times is that we always obtain analytical solu-
tion in term of closed form expression. Hence to ll up to this gap, in this article an attempt has been made to study an MX/G/
1 retrial queue with general retrial times and Bernoulli vacation schedule for an unreliable server, which consists of a servers
breakdown period and delayed period. To this end, the methodology will be based on inclusion of supplementary variables.
The rest of the paper is organized as follows. In Section 2, we give a brief description of the mathematical model. Section 3
deals with the derivations of the stability criteria for existence of stationary regime and the steady state joint distribution of
the server state and the number of customers in the system. Existence of the Stochastic decomposition property is also dem-
onstrated in Section 4. Section 5 deals with the system size distribution at a departure epoch. Some important particular
cases of this model are discussed briey in Section 6. Some important performance measures and reliability indices of this
model are derived in Section 7. The cost effectiveness maximization of this model is discussed with an application to Wire-
less Network System in Section 8.

2. The mathematical model

We consider an MX/G/1 queueing system, where the number of individual primary customers arrives to the system
according to a compound Poisson process with arrival rate k. The size of successive arriving batches is X1, X2, . . . where
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 257

X1, X2, . . . are identically and independently distributed (i.i.d.) random variables with probability mass function (p.m.f)
an = Pr{X = n};n P 1, probability generating function (PGF) a(z) = E[zX] and rst two factorial moments a[1] and a[2].The service
times {Bn; n P 1} of the customers are i.i.d. random variables with probability distribution function (d.f.) B(x), Laplace Stielt-
jes Transform (LST) b(h) = E[ehB] and nite kth moment bk, for k P 1. While the server is serving the primary customers, it
may breakdown at any time and the server will be down for a short interval of time, i.e., service channel is interrupted for a
short interval of times. The service interruption, i.e., servers life times are generated by exogenous Poisson process with rates
a, which we may call some short of disaster during serving the customers. As soon as breakdown occurs it is sent for repair
during which the server stops providing service to the primary customers and waits for repair to start, which we may refer to
as waiting period of the server. We dene this waiting time as delay time. The delay times {Dn; n P 1} of the server are i.i.d
random variable with d.f. D(y), LST D(h) = E[ehD] and kth nite moment dk; for k P 1. The repair times fGn ; n P 1g of the
server are also assumed to be i.i.d random variables with d.f. G(y), LST G(h) = E[ehG] and nite kth moment gk. Immediately
after the broken server is repaired (i.e., server is xed), the server is ready to start its remaining service to the primary cus-
tomers and in this case the service times are cumulative, i.e., we are considering preemptive resume for service time, which
we may referred to as generalized service times. After each service completion the server takes Bernoulli vacation, i.e., after
each service completion the server may go for a vacation of random length V with probability p or with probability q he may
serve the next unit; if any, where p + q = 1. The vacation times fV n ; n P 1g of the server are assumed to be i.i.d random vari-
ables with d.f. V(y), LST c(h) = E[ehV] and kth nite moment ck; for k P 1.This type of queueing model is known as MX/G/1
Bernoulli vacation queue with unreliable server during which a breakdown period and a delay period occur. The model of
this nature without a delay period and single unit arrival case was rst investigated by Li et al. [32].Now further develop-
ment for such a type of model we may further introduce the concept of repeated attempts with FCFS orbit retrial policy, where
if a primary customer who nds the server busy, on vacation or down, then he/she leaves the service area and joins a group
of unsatised customers, i.e., in a orbit in order to seek service again and again until it nds the server free. The time between
successive repeated attempts, i.e., retrial times {Rn; n P 1} of the customers are assumed to be a sequence of i.i.d random
variable with d.f. C(x) and LST C(h) = E[ehR]. Further, we assume that input process, servers life times, servers delay times,
servers repair times, service times and retrial time random variables are mutually independent of each others.

3. Joint distribution of number in the orbit and state of the server

We rst investigate the necessary and sufcient condition for the system to be stable of our model.

Theorem 3.1. The inequality a[1](1  C(k)) + qH < 1 is a necessary and sufcient condition for the system to be stable, where
qH = ka[1]{b1(1 + a(g1 + d1)) + pc1}.

Proof. It is easy to see that {Xn; n 2 Z+} is an irreducible and a periodic Markov chain. To prove the positive recurrent we may
use the Fosters criterion, which states that an irreducible and aperiodic Markov chain is positive recurrent if there exists a
non-negative function f(s), s 2 Z+ and e > 0 such that the mean drift us = E[f(Xn+1)  f(Xn)jXn = s] is nite for all s 2 Z+ and
us 6 e for all s 2 Z+ except perhaps a nite number. In our case, we take f(s) = s to obtain
(
ka1 b1 f1 ag 1 d1 g pka1 c1  1; j 0;
uj
ka1 b1 f1 ad1 g 1 g kpa1 c1 a1 1  C  k  1; j 1; 2; . . .

Obviously, a[1](1  C(k)) + ka[1]{b1(1 + a(g1 + d1)) + pc1} = a[1](1  C(k)) + qH < 1 is a sufcient condition for ergodicity
The same inequality is also necessary for ergodicity. As noted in Sennott et al. [44], we can guarantee nonerodicity if the
Markov chain {Xn; n 2 Z+} satises Kaplins condition, namely uj < 1 for all j P 0 and there exists j0 2 Z+ such that uj P 0 for
j P j0. Notice that in our case, Kaplins condition is satised because there is a suchk such that for pi,j = 0 for j < i  k and
i > 0,where P = (pi,j) is the one step transition matrix of {Xn; n 2 Z+}. Then a[1](1  C(k)) + qH P 1 implies the nonergodicity of
the Markov chain. Hence if a[1](1  C(k)) + qH < 1 and G( ), D( ), V( ), C( ) and B( ) satisfy the regularity conditions, then the
system is stable. h
Since the arrival process is a Poisson process, it can be shown from Burkes theorem (e.g., see Cooper [45], pp. 187188)
that the steady state probabilities of our bivariate Markov process Z(t) = {N(t), X(t)} as t ? 1 exist and positive under the
same condition {Xn, n 2 Z+}, i.e., if and only if a[1](1  C(k)) + qH < 1.
From the mean draft uj = ka[1]{1 + a(g1 + d1)} + pka[1]c1 + a[1](1  C(k))  1 for j P 1, we have reasonable conclusion that
the term k a[1]{1 + a(g1 + d1)} + pka[1]c1 represents a batch arrival during service time, together with a batch arrival when the
server is on vacation both leaving for the orbit. The term a[1] (1  C(k))  1 refers to the contribution to the orbit size due to
a batch arrival during the retrial time excluding the arbitrary customer of the arriving batch whose service commences so
that he/she no longer belongs to the orbit. Similar interpretation can be provided for j = 0. The condition uj < 0 ensure that
the orbit size does not grow indenitely in course of time.
Next, attempts have been made to obtain the PGF of the joint distribution of the state of the server and number in the
system (by system size here we mean including one is being served, if any) by treating elapsed retrial time, elapsed service
time, of the customer and elapsed vacation time, elapsed delay time and elapsed repair time of the server as supplementary
258 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

variables. Assuming the system is in steady state conditions. Let N(t) be the orbit size (i.e., number of customers in the retrial
group) at time t, R0(t), B0(t) be the elapsed retrial time and service time of the customers, respectively at timet. In addition, let
D0(t), H0(t) and V0(t), be the elapsed delay time, respectively elapsed repair time of the server during which breakdown occurs
in the system and elapsed vacation time of the server at time t, Further, we introduced the following random variable:
8
> 0; if the ser v er is idle with zero customer in the system at time t;
>
>
>
> 1; if the ser v er is idle during retrial time at time t;
>
>
>
< 2; if the ser v er is busy at time t;
Yt
>
>
> 3; if the ser v er is on v acation at time t;
>
>
>
>
> 4; if the ser v er is waiting for repair at time t;
:
5; if the ser v er is under repair at time t:
So that the supplementary variables R0(t), B0(t), V0(t), D0(t) and H0(t) are introduced in order to obtain a bivariate Markov
process Z(t) = {N(t), X(t)}, where X(t) = 0 if Y(t) = 0, X(t) = R0(t) if Y(t) = 1, X(t) = B0(t) if Y(t) = 2, X(t) = V0(t) if Y(t) = 3, X(t) = D0(t)
if Y(t) = 4, and X(t) = H0(t) if Y(t) = 5. Now we dene following limiting probabilities:

P0;0 lim Pr fNt 0; Xt 0g;

t!1

wn xdx lim Pr fNt n; Xt R0 t; x < R0 t 6 x dxg; x > 0; n P 1;

t!1

Pn xdx lim Pr fNt n; Xt B0 t; x < B0 t 6 x dxg; x > 0; n P 1;

t!1

Q n ydy lim Pr fNt n; Xt V 0 t; y < V 0 t 6 y dyg; y > 0; nP0

t!1

and for xed values of x and n P 1

Rn x; ydy lim Pr fNt n; Xt D0 t; y < D0 t 6 y dyjB0 t xg; x; y > 0;

t!1

K n x; ydy lim Pr fNt n; Xt H0 t; y < H0 t 6 y dyjB0 t xg; x; y > 0:

t!1

Further, it is assumed that C(0) = 0, C(1) = 1, B(0) = 0, B(1) = 1, V(0) = 0, V(1) = 1, D(0) = 0, D(1) = 1, G(0) = 0, G(1) = 1 f and
that C(x), B(x) is continuous at x = 0 and V(y), D(y) and G(y) are continuous at y = 0 respectively, so that
dCx dBx dVy dDy dGy
gxdx ; lxdx ; cydy ; fydy and nydy
1  Cx 1  Bx 1  Vy 1  Dy 1  Gy
are the rst order differential (hazard rate) functions of C( ), B( ), V( ), D( ) and G( ), respectively.

3.1. The steady state equations

The Kolmogorov forward equations to govern the system under steady state conditions (e.g., see Cox [46]) can be written
as follows:

d
w x k gxwn x 0; n P 1; 3:1
dx n

Xn Z 1
d
Pn x k a lxP n x k ak Pnk x fyRn x; ydy; n P 1; 3:2
dx k1 0

d Xn
Q n y k cyQ n y k1  dn;0 ak Q nk y; n P 0; 3:3
dy k1

d Xn
Rn x; y k fyRn x; y k ak Rnk x; y; n P 1; 3:4
dy k1

d Xn
K n x; y k nyK n x; y k ak K nk x; y; n P 1; 3:5
dy k1
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 259

Z 1 Z 1
kP0;0 cyQ 0 ydy q lxP1 xdx; 3:6
0 0

where P0(x) = 0, R0(x, y) = 0 = K0(x, y) occurring in Eqs. (3.2), (3.4) and (3.5) respectively.
These set of equations are to be solved under the boundary conditions at x = 0:
Z 1 Z 1
wn 0 cyQ n ydy q lxPn1 xdx; n P 1; 3:7
0 0

X
n Z 1 Z 1
Pn 0 k ai wni xdx kan P0;0 gxwn xdx; n P 1 3:8
i1 0 0

and at y = 0:
Z 1
Q n 0 p lxPn1 xdx; n P 0; 3:9
0

also at y = 0 and xed values of x

Rn x; 0 aPn x; x > 0; n P 1; 3:10
Z 1
K n x; 0 fyRn x; ydy; x > 0; n P 1; 3:11
0
R1
where 0
w0 xdx 0 occurring in Eq. (3.8). Also, the normalizing condition is given by
1 Z
X 1 1 Z
X 1 Z 1 Z 1 Z 1 Z 1 Z 1 
P0;0 Q n ydy Pi;n xdx wn xdx Rn x; ydx dy K n x; ydx dy 1:
n0 0 n1 0 0 0 0 0 0

3:12

3.2. The model solution

To solve the system Eqs. (3.1)(3.11) let us dene following PGFs for jzj < 1:
X
1 X
1
Rx; y; z zn Rn x; y; Rx; 0; z zn Rn x; 0;
n1 n1

X
1 X
1
Kx; y; z zn K n x; y; Kx; 0; z zn K n x; 0;
n1 n1

X
1 X
1
Q y; z zn Q n y; Q 0; z zn Q n 0;
n0 n0

X
1 X
1
Px; z zn P n x; P0; z zn Pn 0;
n1 n1

X
1 X
1 X
1
wx; z zn wn x; w0; z zn wn 0 and az zn a n :
n1 n1 n1

Let b(z) = k(1  a(z)), then proceeding in usual manner with Eqs. (3.1)(3.5), we get a set of differential equations of
Lagrangian type whose solutions are given by:
wx; z w0; z1  Cx expfkxg; 3:13:1

Px; z P0; z1  Bx expfAzxg; 3:13:2

Q y; z Q0; z1  Vy expfbzyg; 3:13:3

Rx; y; z Rx; 0; z1  Dy expfbzyg; 3:13:4

and Kx; y; z Kx; 0; z1  Gy expfbzyg; 3:13:5

where A(z) = b(z) + a(1  D (b(z))G (b(z))).
Again multiplying both sides of Eq. (3.7) by zn and then taking summation over all possible values of n P 1, we get
260 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

w0; z Q 0; zc bz qz1 P0; zb Az  kP0;0 : 3:14

Similarly from Eqs. (3.8) and (3.9), we get

zP0; z kazP0;0 w0; zaz1  C  k C  k 3:15

and Q 0; z z1 pP0; zb Az: 3:16

Again from Eqs. (3.10) and (3.11), we have

Rx; 0; z aP0; z1  Bx expfAzxg 3:17

and Kx; 0; z Rx; 0; zD bz: 3:18

Now from Eqs. (3.16)(3.18), we get

P0;0 zbzC  k
P0; z ; 3:19
BzfC k az1  C  kg  z


P 0;0 kz  azBz
w0; z 3:20
BzfC  k az1  C  kg  z

P0;0 pbzb AzC  k

and Q 0; z ; 3:21
BzfC  k az1  C  kg  z

where B(z) = {q + pc(b(z))}b(A(z)).

Let z ? 1 in (3.19), we obtain the following equation by the L Hospitals rule

ka1 C  kP0;0
P0; 1 : 3:22
1  a1 1  C  k  qH 

This gives
ka1 C  kP0;0 1  Bx
Px; 1 ; 3:23
1  a1 1  C  k  qH 

pka1 C  kP 0;0 1  Vy

Qy; 1 ; 3:24
1  a1 1  C  k  qH 

ka1 C  kP0;0 1  Bx1  Dy

Rx; y; 1 3:25
1  a1 1  C  k  qH 

ka1 C  kP0;0 1  Bx1  Gy

and Kx; y; 1 : 3:26
1  a 1  C  k  qH 
Hence from normalizing condition (3.12), we get
1  a1 1  C  k  qH 
P0;0 : 3:27
C  k
Note that the Eq. (3.27) represents steady state probability that the server is idle but available in the system. Thus we sum-
marize our results in the following Theorem 3.2.

Theorem 3.2. Under the stability condition a[1](1  C(k)) + qH < 1, the joint distribution of the state of the server and the size of
the orbit has the following partial PGFs as
k1  a1 1  C  z  qH z  azBz1  Cx expfkxg
wx; z ; 3:28
C  kBzfC  k az1  C  kg  z

z1  a1 1  C  k  qH bz1  Bx expfAzxg

Px; z ; 3:29
BzfC  k az1  C  kg  z

p1  a1 1  C  k  qH bz1  Vy expfbzyg

Qy; z ; 3:30
BzfC  k az1  C  kg  z
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 261

az1  a1 1  C  k  qH bz1  Bx expfAzxg  1  Dy expfbzyg

Rx; y; z ; 3:31
BzfC  k az1  C  kg  z

az1  a1 1  C  k  qH D bzbz1  Bx expfAzxg  1  Gy expfbzyg

Kx; y; z ; 3:32
BzfC  k az1  C  kg  z
where b(z) = k(1  a(z)), A(z) = b(z) + a(1  D(b(z))G(b(z))) and B(z) = b(A(z))  {q + pc(b(z))}. h

Next we are interested to investigate marginal orbit size distribution due to system state of the server.

Theorem 3.3. Under the stability condition a[1](1  C(k)) + qH < 1, the marginal PGFs of the servers state orbit size distribution
are given by
1  a1 1  C  k  qH z  azBz1  C  k
wz ; 3:33
C  kBzfC  k az1  C  kg  z

1  a1 1  C  k  qH zbz1  b Az

Pz ; 3:34
AzBzfC  k az1  C  kg  z

p1  a1 1  C  k  qH b Az1  c bz

Q z ; 3:35
BzfC  k az1  C  kg  z

az1  a1 1  C  k  qH 1  b Az1  D bz

Rz 3:36
AzBzfC  k az1  C  kg  z

az1  a1 1  C  k  qH 1  b AzD bz1  G bz

and Rz : 3:37
AzBzfC  k az1  C  kg  z

Proof. Integrating (3.28) and (3.29) with respect to x and using the well known result of renewal theory
Z 1
1  f  s
esx 1  Fxdx ; 3:38
0 s
where F( ) is distribution function of a random variable and f( ) represents LST of it, we get formulae (3.33) and (3.34), respec-
tively. Similarly, integrating (3.30) with respect to y, we get formula (3.35).
Again, integrating Eq. (3.31) and (3.32) with respect to y, we get
Z 1
aPx; z1  D bz
Rx; z Rx; y; zdy ; 3:39
0 bz
Z 1
aPx; zD bz1  G bz
Kx; z Kx; y; zdy : 3:40
0 bz
Further integrating (3.39) and (3.40) with respect to x, we claim in formula (3.36) and (3.37), respectively. h

Corollary 3.1. Let Uj and Xj be the stationary distribution of the number of customers in the system and orbit respectively under
P P1 j
the steady state condition, then its corresponding PGFs Uz 1 j
j0 z Uj and Xz j0 z Xj are given by

1  a1 1  C  k  qH 1  zq pc bzb Az

Uz 3:41
fq pc bzgfC  k az1  C  kgb Az  z

1  a1 1  C  k  H 1  z
q
and Xz  bzgfC  k az1  C  kgb Az
; 3:42
fq pc z
respectively.

Proof. The result follows directly from Theorem 3.2. With the help of PGFs w(z), P(z), Q(z), R(z), K(z) and P0,0, we get the dis-
tribution of the number of customers in the system at a random epoch having PGF
Uz P0;0 wz Pz zQz Rz Kz:
By direct calculation we can obtain (3.41). Further utilizing the relationship between orbit size distribution and system size
distribution for an MX/G/1 type retrial queue, we may write
262 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

XzBz Uz: 3:43

Now utilizing (3.41) in (3.43), we obtain (3.42).
It should be noted here that for a = 0 above expression (3.42) is consistent with the expression (4.2) of Krishna Kumar
et al. [26]. h

4. Stochastic decomposition

In this section we study the stochastic decomposition property of the system size distribution of our model. The literature on
vacation models recognizes this property one of the most interesting features in this mater, e.g., see Doshi [47], and Fuhrmann
and Cooper [48]. Stochastic decomposition for the retrial models has also been found in Yang et al. [15] and Yang and Tampleton
[1]. The existence of the stochastic decomposition property for our model can be demonstrated easily by showing that
Uz U0 zvz; 4:1
X
where U0(z) is the PGF of the system size distribution of an M /G/1 queue with unreliable server and delaying repair under
Bernoulli vacation schedule, which can obtained from formula (3.40) by putting C(k) = 1 in the formula (3.40). Thus we have
1  qH 1  zq pc bzb Az
U0 z : 4:2
fq pc bzgb Az  z
Note that for a = 0 (i.e., for reliable server) the above result is consistent with the result obtained by Choudhury and Madan
[49] and v(z) is the PGF of the conditional distribution of the number of customers in the orbit given that the system is
P 0;0 wz
idle P0;0 w1
which is equal to
1  a1 1  C  k  qH fq pc bzgb Az  z
vz :
1  qH fq pc bzgb AzfC  k az1  C  kg  z
More specically, we may call it additional system size distribution due to retrial time.
The expected number of customers in the orbit during the retrial time, i.e., mean of additional system size due to retrials
is found to be

qH 1  C  ka1 ka2 pc2 2c1 b1 f1 ad1 g 1 g1  C  ka1

Lv v0 1
1  a1 1  c& k  qH  21  qH 1  a1 1  c& k  qH 
ka1 2 b2 f1 ad1 g 1 g2 ab1 fd2 2d1 g 1 g 2 g1  C  ka1
: 
21  qH 1  a1 1  c& k  qH 

Remark 4.1. From expression (4.1), we observe that the system size distribution at a departure epoch of the unreliable MX/
G/1 retrial queue with delaying repair and generalized repeated attempts under Bernoulli vacation schedule can be
decomposed into distributions of two independent random variables viz.

1. The system size distribution of an MX/G/1 queue with unreliable server and delaying repair under Bernoulli vacation sche-
dule [represented by rst term of right hand side of expression (4.1)] and
2. The additional system size distribution due to retrials [represented by the second term of right hand side of expression
(4.1)]. h

5. Distribution of number of customers in the system at a departure epoch

In this section rst we derive the PGF of the stationary system size distribution a departure epoch. The key result of this
section is now stated in Theorem 5.1.

Theorem 5.1. Under the steady state condition a[1](1  C(k)) + qH < 1, the PGF of the steady state system size distribution at a
departure epoch for this type of model is given by

1  a1 1  C  k  qH 1  azfq pc bzgb Az

pz : 5:1
a1 fq pc bzgb Az  z

Proof. Let sn be the time of nth total service completion epoch, i.e., we consider the epochs at which the required service
time requested by an unit expires. Then the sequence Nn = N(sn+) form a Markov chain which is the embedded Markov
renewal process of the continuous time Markov process Z(t) as t ? 1. Next we dene the following limiting probabilities
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 263

pj n!1
lim P r fNn jg; j P 0:

Now since the arrival process is compound Poisson process, it follows from Burkes theorem (e.g., Cooper [45], pp. 187
188) that the departure point probabilities {pj} exist and positive under the same condition a[1](1  C(k)) + qH < 1.
Then following argument of PASTA we note that a departing customer will see j units in the system just after a departure
if and only if there were (j + 1) units in the system just before the departure and therefore we may write
Z 1 Z 1
pj B0 q lxPj1 xdx B0 cxQ j xdx; j P 0; 5:2
0 0

where B0 is the normalizing constant.

P
Let pz 1 j j
j0 z pj be the PGF of {pj; j P 0}, then multiplying Eq. (5.2) by both sides of z and then taking summation over
j P 0 and utilizing Eqs. (3.29) and (3.30), we get on simplication

B0 1  a1 1  C  k  qH bzfq pc bzgb Az

pz : 5:3
fq pc bzgb Az  z

Utilizing normalizing condition p(1) = 1, we get

B0 ka1 1 : 5:4

Now inserting (5.4) in (5.3), we get required expression (5.1). h

Remark 5.1. If we compare expression (3.40) with expression (5.1), then we have
1  az
pz /z Hz/z; as expected; 5:5
a1 1  z
where Hz a=1az
11z
is the PGF of the number of customers placed before an arbitrary test customer (tagged customer) in an
admission batch in which the tagged customer arrives. This number is given as backward recurrence time in the discrete
time renewal process where renewal points are generated by the arrival size random variable and a[1] = a0 (1).
This is the relation ship between the system size distributions at a random point of time and at departure point of time for
MX/G/1 type of retrial queue with unreliable server. Now form above expression (5.5) it is observed that the PGF of system
size distribution at a departure epoch of this type of model is the convolution of distributions of three independent random
variables, one of which is the number of customers placed before a tagged customer in a batch in which the tagged customer
arrives. This is due to randomness property of the arriving size of the batches. The interpretation of the other two random
variables is provided in Remark 4.1. h

Next, we are interested about the distribution for the number of customers in the system immediately after those time
points at which a customer departs or an idle period is ended.

Theorem 5.2. Let W(z) be the PGF of the number of customers in the system at immediately after those time points at which a
customer departs or an idle period is terminated of an Mx/G/1 retrial queue with Bernoulli vacation schedule and unreliable server,
then under steady state condition a[1](1  C(k)) + qH < 1 we have
1  a1 1  C  k  qH fq pc bzgb Az  zaz
Wz : 5:6
1 a1  qH fq pc bzgfC  k az1  C  kgb Az  z

Proof. The result follows directly by utilizing stochastic decomposition property for an Mx/G/1 retrial queue with Bernoulli
vacation schedule and unreliable server viz (e.g., Section 4)

Wz W0 zvz; 5:7
where W0(z) is the PGF of the stationary queue size distribution at which a customer departs or at termination epoch of an
idle period for an Mx/G/1 type of retrial queue with Bernoulli vacation schedule and unreliable server. This can be obtained
easily by replacing the original service time distribution by our generalized service time distribution or equivalently
B(z) = {q + pc(b(z))}b(A(z)), i.e., the PGF of a batch of customers who arrived during our generalized service time, in the well-
known result of Gross and Harris [16] which is of the form [see Section 5.1.9, p. 237]:
C 0 Bz  zaz
W0 z ; 5:8
Bz  z
where C0 is a constant to be evaluated.
Now utilizing B(z) in (5.8), we may write
264 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

C 0 fq pc bzgb Az  zaz

W0 z : 5:9
fq pc bzgb Az  z
Since W0(1) = 1, we get

1  qH
C0 : 5:10
1 a1  qH
Hence formula (5.6) follows by inserting (5.9), (5.10) and (4.1) in (5.7). h

6. Some particular case

In this section we discuss briey some important particular cases of this type of model, which are consistent with existing
literature. Suppose that retrial time distribution is exponential with parameter m > 0, then we have C(k) = m(k + m)1 and
therefore from Eqs. (3.40), (3.41) and (5.1) yields

k m1  qH  a1 k1  zfq pc bzgb Az

Uz ; 6:1
fm kazgfq pc bzgb Az  zk m

k m1  qH  a1 k1  z
Xz 6:2
fm kazgfq pc bzgb Az  zk m

k m1  qH  a1 k1  azfq pc bzgb Az

and pz : 6:3
a1 fm kazgfq pc bzgb Az  zk m
These are the PGFs of the system size distribution at a random point of time, orbit size distribution and the system size
distribution at a departure epoch of the unreliable MX/G/1 retrial queue with delaying repair and constant repeated attempts
under Bernoulli vacation schedule. Note that for p = 0 (i.e., there is no Bernoulli vacation in the system) and a = 0 (i.e., there is
no breakdown in the system) and a(z) = z (i.e., for single unit arrival case) the above expression (6.1) is consistent with
expression (2.13) of Choi et al. [50].
Again if we take z = c(b(z)) in the Eqs. (3.40), (3.41) and (5.1), then we get on simplication

q  a1 1  C  k  qH 1  zq pzb Az
Uz ; 6:4
fq pzgfC  k az1  C  kgb Az  z

q  a1 1  C  k  qH 1  z
Xz 6:5
fq pzgfC  k az1  C  kgb Az  z

q  a1 1  C  k  qH 1  azq pzb Az

and pz ; respectively: 6:6
a1 fq pzgfC  k az1  C  kgb Az  z
These are the PGFs of the number of customers in the system at a service completion epoch, the number of customers in the
orbit and number of customers in the system at a departure epoch respectively service completion epoch for the unreliable
MX/G/1 retrial queue with Bernoulli feedback mechanism and generalized repeated attempts.
Further, if we take p = 0 (i.e., there is no Bernoulli feedback in the system), then qH = ka[1]b1{1 + a(g1 + d1)} < 1 and expres-
sion (6.4)(6.6) yields

1  a1 1  C  k  qH 1  zb Az
Uz ; 6:7
fC  k az1  C  kgb Az  z

1  a1 1  C  k  qH 1  z
Xz 6:8
fC k az1  C  kgb Az  z


1  a1 1  C  k  qH 1  azb Az
and pz ; respectively: 6:9
a1 fC  k az1  C  kgb Az  z
These are the corresponding PGFs of the system size distributions and orbit size distribution of the unreliable MX/G/1 retrial
queue with delaying repair and generalized repeated attempts. Note that a = 0 (i.e., for reliable server) and a(z) = z (i.e., for
single unit arrival case), the above expressions are consistent with the result obtained by Gomez-Corral [51]. It should be
noted here this type of model was considered by Krishna kumar et al. [26].
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 265

7. System performance measures

Our next object is to provide explicit expression for the system state probabilities and some important performance mea-
sures of the system. First of all we derive the system state probabilities and results are summaries in the following theorem.

Theorem 7.1. If the system is in steady state conditions, then we have

1a 1C  kq 

1 H
(i) the probability that the server is idle and system is empty is PE C  k
;
1a1 1C  kqH 
(ii) the probability that the server is idle but system is nonempty is P NE C  k
qH
(iii) the probability that the server is busy is PB = ka[1]b1
(iv) the probability that the server is waiting for repair is PW = aka[1]b1d1
(v) the probability that the server is under repair is PR = aka[1]b1g1
(vi) the probability that the server is on vacation is PV = pka[1]c1

Proof. Note that

PNE limz!1 wz; PB limz!1 Pz; P W limz!1 Rz; P R limz!1 Kz, and P V limz!1 Q z and P E 1  fPNE P B
P V PW P R g, the stated formulae follow by direct calculation. h

Secondly, we derive the mean orbit size and the mean system size of this model under stability conditions.

Theorem 7.2. Let E(NO), E(NS) and E(ND) be the expected number of units in the orbit and in the system at stationary point of time
and at a departure epoch respectively, then under the steady state condition we have
qH 1  C  ka1 ka1 b2 f1 ad1 g 1 g2 ab1 fd2 2d1 g 1 g 2 g
ENO 
1  a1 1  C k  qH  21  a1 1  C  k  qH 
pka1 2 c2 2c1 b1 f1 ad1 g 1 g a1 1  C  k qH 
 aR ; 7:1
21  a1 1  C k  qH  1  a1 1  C  k  qH 

ENS ENO qH 7:2

and END ENS aR ; 7:3

a2
where aR 2a1
is the mean residual batch size.

Proof. The results follow directly by differentiating Eqs. (3.40), (3.41) and (5.1) with respect to z and then taking limit z ? 1
by using the L Hospitals rule. h

Remark 7.1. Once we have obtained the expected number of units in the orbit, it is trivial to obtain the other related per-
formance measure viz mean waiting time. The mean waiting time of W in the steady state can be easily obtained with the
help of Littles formula viz L0 = ka[1]W. h

Next we derive the mean busy period and expected length of a busy cycle under the steady state condition.

Theorem 7.3. Let Tb and Tc be the length of a busy period and length of a busy cycle, respectively, then under the steady state
conditions, we have
1  C  ka1  1 b1 f1 ad1 g 1 g pc1
ET b 7:4
ka1 1  a1 1  C  k  qH  1  a1 1  C  k  qH  1  a1 1  C  k  qH 

C  k
and ET c : 7:5
ka1 1  a1 1  C  k  qH 

Proof. The results follow directly by applying the argument of alternating renewal process which lead to well known result
1
ET b P0;0 1  1; 7:6
ka1

and ET c ka1 P0;0 1 : 7:7

Inserting (3.26) in (7.6) and (7.7), we get (7.4) and (7.5), respectively. h
266 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

Finally, we consider two reliability indices of the system viz the system availability and failure frequency under the
steady state conditions. Let AV(t) be the system availability at time t, that is, the probability that the server is either working
for a customer or in an idle period, such that the steady state availability of the server will be AV = limt?1AV(t).

Theorem 7.4. The availability of the server and failure frequency of the server under the steady state condition are given by
1  a1 f1  C  k kb1 C  k 1 ag 1 d1 pkc1 g
AV 7:8
C  k

and Mf aka1 b1 : 7:9

.

Proof. The result follows directly by considering the following equations

Z 1
AV P0;0 Px; 1dx;
0

Z 1
Mf a Px; 1dx:
0
R1 R1
Now since 0
1  Bxdx 0
xdBx b1 ; therefore from Eqs. (3.22) and (3.26), we get (7.8) and (7.9), respectively. h

8. A real world application with the cost effectiveness maximization model

In this section, we present a possible application and some numerical examples in some situations to explain that our
model can represent the possible application reasonably well. In wireless network, access point is used to connect between
wireless and wired networks. Wireless access point provides services to a WLAN. It serves as an interface point and bridges
between wireless devices and wired network, so the wireless devices can access the resources of the wired network. It fur-
ther provides wireless linkage between wireless devices that may be out of range of each other. Typically, wireless devices
connection requests arrive at the access point following the Poisson stream. When requests arrive at the access point, one
request is selected to be served and the rest of the requests will join the buffer and retry again after a random period. In the
buffer, each message waits a certain amount of time and requires the service again. There is a daemon program implemented
at access point to manage the connection requests from buffer. Each time it tries but fails, it will wait another amount of time
and then try again. At the completion of a service, access point may perform one of maintenance activities such as virus scan
to keep the access point functioning well or serve the next connection request. In addition, in practice the access point may
fail and can be repaired. The repair time depends on the degree of failure of access point. Because the system performance
may be heavily affected by access point breakdown, it is well worth to investigate such system from the queueing theory
viewpoint, as well as reliability point of view. In this scenario, access point, buffer in the access point, retransmission policy,
and maintenance activities correspond to the server, the orbit, the retrial discipline, and the vacation policy, respectively, in
the queueing terminology.
For access point, we develop the cost effectiveness dened as (availability)/(the expected out of pocket cost rate) to be an
alternative cost criterion, which reects the efciency per dollar outlay. This criterion is useful for the effective use of avail-
able money. This criterion is helpful in the situation that the benets obtained from the investment are not reducible to mon-
etary terms.
Let Cr be the expected cost rate per busy cycle, then
Cs
Cr ;
ET c
where Cs is out of pocket cost per cycle. Further, the cost effectiveness is dened as
AV
Ce :
ET c
We now present some numerical examples to study the effect of various parameters k (request arrival rate), a (retrial rate of
requests in the buffer) andp(the probability of access point performing maintenance) on the cost rate and cost effectiveness
listed above. For convenience, the settings of systems parameters are given by:

 Cs = 100.
 Geometric batch size with mean a[1] = 1.0 for the request arrival.
 Exponential service time with a mean b1 = 0.25 for requests.
 Exponential delayed time with a mean d1 = 1 for access point.
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 267

40 1

Cost Effectness
Cost Rate

20 CrExp 0.5
CeExp
CrErl
CeErl
CrHyp
CeHyp

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Fig. 1. Cost rate and cost effectiveness for various values of k.

40 0.04

30
Cost Effectness
Cost Rate

CrExp
20
CeExp
CrErl
CeErl
CrHyp
10
CeHyp

0.035
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Fig. 2. Cost rate and cost effectiveness for various values of a.

 Exponential repair time with a mean g1 = 1 for access point.

 Exponential vacation time with a mean c1 = 0.5 for access point.
268 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269

40 0.04

Cost Effectness
Cost Rate

20 0.03
CrExp
CeExp
CrErl
CeErl
CrHyp
CeHyp

0 0.02
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
p

Fig. 3. Cost rate and cost effectiveness for various values of p.

For illustrative purpose, Exponential (denoted by Exp), 2-stage Erlang (Erl) and 2-stage Hyper-exponential (Hyp) are con-
sidered as retrial time, respectively, with retrial rate is 2. The results of cost rate and effectiveness are shown, respectively, in
Figs. 13 for the following three cases.

(I) We choose a = 0.05, p = 0.5, and vary the values of k from 0.0 to 1.0 for different retrial time distribution.
(II) We choose k = 0.6, p = 0.5, and vary the values of a from 0.0 to 1.0 for different retrial time distribution.
(III) We choose k = 0.6, a = 0.05, and vary the values of p from 0.0 to 1.0 for different retrial time distribution.

As shown in Fig. 1, the cost rate increases rst and then decreases, while the cost effectiveness decreases rst and then in-
creases, when the arrival rate for requests increases from 0.0 to 1.0. The reason is that access point has to take more computing
power to deal with the arriving requests in the initial period. Along with more and more requests being served, new arriving
requests with the same destination can be served quickly. Figs. 2 and 3 reveal that the effect of various retrial rates and vacation
probabilities on the cost rate and the cost effectiveness. From the gures, we can see that the cost rate decreases as retrial rate or
vacation probability increases from 0.0 to 1.0. Moreover, we also observe that the cost effectiveness increases as retrial rate or
vacation probability increases from 0.0 to 1.0. This is because that the increasing of retrial rate or vacation probability repre-
sents more connection requests are waiting in the buffer. From the viewpoint of access point, more connection requests can
be treated using the same computing power no matter the connection requests being served or waiting in the buffer.

9. Conclusions

In this paper we analyzed an MX/G/1 queue with general retrial time and Bernoulli vacation schedule for an unreliable
server, which consists of a breakdown period and delay period. The distributions of some important system characteristics
for such a system were derived. A cost effectiveness maximization model was also developed to demonstrate the efciency
per dollar outlay. Some extensive numerical computations were performed to study the effect of the system parameters on
the system characteristics. This research presents an extension of the vacation model theory and the analysis of the model
will provide a useful performance evaluation tool for more general situations arising in practical applications, for example,
access point is used to connect between wireless and wired networks in wireless network (presented in Section 8). Future
investigation can try to incorporate a more complex system (such as non-Markovian input) with different failure criteria. In
addition, the effect of balking/reneging on a service system or more complex system may also be a problem of interest.

Acknowledgement

We thank Editor, Professor Cross, and referees for constructive suggestions to improve its presentation.
 G. Choudhury, J.-C. Ke / Applied Mathematical Modelling 36 (2012) 255269 269

References

[1] T. Yang, J.G.C. Templeton, A survey on retrial queue, Queueing Syst. 2 (1987) 201233.
[2] G.I. Falin, A survey of retrial queues, Queueing Syst. 7 (1990) 127168.
[3] V.G. Kulkarni, H.M. Liang, Retrial queues revisited, in: J.H. Dshalalow (Ed.), Frontiers in Queueing, CRC, New York, 1997, pp. 1934.
[4] G.I. Falin, J.G.C. Templeton, Retrial Queues, Chapman and Hall, London, 1997.
[5] J.R. Artalejo, Accessible bibliography on retrial queues, Math. Comput. Model. 30 (1999) 16.
[6] J.R. Artalejo, A classical bibliography of research on retrial queues: Progress in 19901999, TOP 7 (1999) 187211.
[7] A. Gomez Corral, A bibliographical guide to analysis of retrial queues through matrix analytic techniques, Ann. Oper. Res. 141 (2006) 163191.
[8] J.R. Artalejo, G. Falin, Standard and retrial queueing systems: A comparative analysis, Rev. Math. Comput. 15 (2002) 101129.
[9] P. Moren, An M/G/1 retrial queue with recurrent customers and general retrial times, Appl. Math. Comput. 159 (2004) 651666.
[10] G. Fayolle, A simple telephone exchange with delayed feedback, teletrafc analysis and computer performance evaluation, in: O.J. Boxma, J.W. Cohen,
H.C. Tijms (Eds.), Elsevier, Amsterdam, 1986, pp. 245253.
[11] K. Farahmand, Single line queue with repeated attempts, Queueing Syst. 6 (1990) 223228.
[12] B.D. Choi, Y.W. Shin, W.C. Ahn, Retrial queues with collision arising from unslottted CDMA/CD protocol, Queueing Syst. 11 (1992) 335356.
[13] V.A. Kapyrim, A study of stationary distributions of a queueing system with recurring demands, Cybernatics 13 (1977) 548590.
[14] G.I. Falin, Single line repeated orders queueing systems, Optimization 17 (1986) 649677.
[15] T. Yang, M.J.M. Posner, J.G.C. Templeton, H. Li, An approximation method for the M/G/1 retrial queue with general retrial times, Eur. J. Oper. Res. 76
(1994) 552562.
[16] D. Gross, C.M. Harris, Fundamentals of Queueing Theory, 2nd ed., John Willey and Sons, New York, 1985.
[17] I. Atencia, P. Moreno, A single server retrial queue with general retrial time and Bernoulli schedule, Appl. Math. Comput. 162 (2005) 855880.
[18] G. Choudhury, An M/G/1 retrial queue with an additional phase of second service and general retrial time, Int. J. Inform. Manage. Sci. 20 (2009) 114.
[19] B. Krishna Kumar, A. Vijay Kumar, D. Arivudainambi, An M/G/1 retrial queueing system with two phase service and preemptive resume, Ann. Oper. Res.
113 (2002) 6179.
[20] X. We, P. Brill, M. Hlyanka, J. Wang, An M/G/1 retrial queue with balking and retrials during service, Int. J. Oper. Res. 1 (2005) 3051.
[21] J.R. Artalejo, Analysis of an M/G/1 queue with constant repeated attempts and server vacations, Comput. Oper. Res. 24 (1997) 493504.
[22] G. Langaries, Gated pooling models with customers in orbit, Math. Comput. Model. 30 (1999) 171187.
[23] J.-C. Ke, F.-M. Chang, Modied vacation policy for M/G/1 retrial queue with balking and feedback, Comput. Indus. Eng. 57 (2009) 433443.
[24] J. Keilson, L.D. Servi, Oscillating random walk models for GI/G/1 vacation systems with Bernoulli schedule, J. Appl. Probab. 23 (1986) 790802.
[25] B. Krishna Kumar, D. Arivudainambi, The M/G/1 retrial queue with Bernoulli schedule and general retrial time, Comput. Math. Appl. 43 (2002) 1530.
[26] B. Krishna Kumar, D. Arivudainambi, A. Vijayakumar, On the MX/G/1retrial queue with Bernoulli schedule and general retrial times, Asia-Pacic J. Oper.
Res. 19 (2002) 177194.
[27] Z. Wenhui, Analysis of a single server retrial queue with FCFS orbit and Bernoulli vacation, Appl. Math. Comput. 161 (2005) 353364.
[28] D.P. Gaver, A waiting line with interrupted service including priorities, J. R. Stat. Soc., Ser.-B 24 (1962) 7390.
[29] B. Avi-ltzhak, P. Naor, Some queueing problems with the service station subject to breakdowns, Oper. Res. 11 (1963) 303320.
[30] K. Thirurengadan, Queueing with breakdowns, Oper. Res. 11 (1963) 6271.
[31] I.L. Mitrany, B. Avi-Itzhak, A many server queue with service interruptions, Oper. Res. 16 (1968) 628638.
[32] W. Li, D. Shi, X. Chao, Reliability analysis of M/G/1 queueing system with server breakdowns and vacations, J. Appl. Probab. 34 (1997) 546555.
[33] B. Sengupta, A queue with service interruptions in an alternating random environment, Oper. Res. 38 (1990) 308318.
[34] T. Takine, B. Sengupta, A single server queue with service interruptions, Queueing Syst. 26 (1998) 285300.
[35] Y. Tang, A single server M/G/1 queueing system subject to breakdowns Some reliability and queueing problems, Microelectron. Reliab. 37 (1997)
315321.
[36] A. Aissani, On the M/G/1/1 queueing system with repeated orders and unreliable, J. Technol. 6 (1988) 93123.
[37] V.G. Kulkarni, B.D. Choi, Retrial queues with server subject to breakdowns and repairs, Queueing Syst. 7 (1990) 191208.
[38] J. Wang, J. Cao, Q. Li, Reliability analysis of the retrial queue with server breakdowns and repairs, Queueing Syst. 38 (2001) 363380.
[39] G. Choudhury, K. Deka, An M/G/1 retrial queueing system with two phases of service subject to the server breakdown and repair, Perform. Eval. 65
(2008) 714724.
[40] G.I. Falin, Aggregate arrival of customers in one line systems with repeated calls, Ukrain Math. J. 28 (1976) 337340.
[41] G. Choudhury, K. Deka, A batch arrival retrial queue with two phases of service and Bernoulli vacation schedule, Acta Math. Appl. Sin., forthcoming.
[42] G. Choudhury, A two phase batch arrival retrial queueing system with Bernoulli vacation schedule, Appl. Math. Comput. 188 (2007) 14551466.
[43] J.-C. Ke, F.-M. Chang, M[x]/(G1,G2)/1 retrial queue under Bernoulli vacation schedules with general repeated attempts and starting failures, Appl. Math.
Model. 33 (2009) 31863196.
[44] L.I. Sennot, P.A. Humblet, R.L. Tweedie, Mean drifts and the non-ergodicity of Markov chain, Oper. Res. 31 (1983) 783789.
[45] R.B. Cooper, Introduction to Queueing Theory, Elsevier, Amsterdam, 1981.
[46] D.R. Cox, The analysis of non Markovian stochastic process by the inclusion of supplementary variables, Proc. Cambridge Philos. Soc. 51 (1955) 433
441.
[47] B.T. Doshi, Queueing systems with vacations-A survey, Queueing Syst. 1 (1986) 2966.
[48] S.W. Fuhrmann, R.B. Cooper, Stochastic decomposition in M/G/1queue with generalized vacations, Oper. Res. 33 (1985) 11171129.
[49] G. Choudhury, K.C. Madan, A batch arrival queue with Bernoulli vacation schedule under multiple vacation policy, Int. J. Manage. Sci. 12 (2) (2006) 1
18.
[50] B.D. Choi, K.H. Rhee, K.K. Park, The M/G/1 retrial queue with retrial rate control policy, Prob. Eng. Inform. Sci. 7 (1993) 2946.
[51] A. Gomez-Corral, Stochastic analysis of a single server retrial queue with general retrial time, Naval Res. Log. 46 (1999) 561581.