J Syst Sci Syst Eng (Sep 2010) 19(3): 306-320 ISSN: 1004-3756 (Paper) 1861-9576 (Online)

DOI: 10.1007/s11518-010-5140-z CN11-2983/N

A BATCH ARRIVAL RETRIAL QUEUE WITH STARTING FAILURES,

FEEDBACK AND ADMISSION CONTROL

Jinting WANG Peng-Feng ZHOU

Department of Mathematics, School of Science, Beijing Jiaotong University, Beijing , China
jtwang@bjtu.edu.cn ( )

Abstract
This paper is concerned with the analysis of a feedback M[X]/G/1 retrial queue with starting failures
and general retrial times. In a batch, each individual customer is subject to a control admission policy
upon arrival. If the server is idle, one of the customers admitted to the system may start its service and
the rest joins the retrial group, whereas all the admitted customers go to the retrial group when the
server is unavailable upon arrival. An arriving customer (primary or retrial) must turn-on the server,
which takes negligible time. If the server is started successfully (with a certain probability), the
customer gets service immediately. Otherwise, the repair for the server commences immediately and
the customer must leave for the orbit and make a retrial at a later time. It is assumed that the customers
who find the server unavailable are queued in the orbit in accordance with an FCFS discipline and only
the customer at the head of the queue is allowed for access to the server. The Markov chain underlying
the considered queueing system is studied and the necessary and sufficient condition for the system to
be stable is presented. Explicit formulae for the stationary distribution and some performance measures
of the system in steady-state are obtained. Finally, some numerical examples are presented to illustrate
the influence of the parameters on several performance characteristics.
Keywords: Batch arrival, feedback, repair, retrial queue, starting failure

1. Introduction telecommunication systems (see Yang &

Retrial queueing systems are characterized
by the feature that arriving customers finding all
the servers unavailable upon arrival are obliged
to leave the service area and repeat their requests
for service after some random time. All blocked
customer join a pool of unsatisfied customers
called orbit. Retrial queues are widely used as
mathematical models of several computer and
Templeton 1987). With the recent advancements
in mobile communications the significance of
retrials is becoming more recognized. There is
an extensive literature on the retrial queues, and
readers are referred to Yang & Templeton (1987),
Falin (1990), Falin & Templeton (1997),
Kulkarni & Liang (1997), Artalejo (1999) and a
recent book of Artalejo & Gomez-Corral (2008).

This work was supported by the National Natural Science Foundation of China (10871020).

Systems Engineering Society of China and Springer-Verlag Berlin Heidelberg 2010

Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng
Most of the papers on retrial queues have
considered the system with single arrivals,
which cannot cover many practical situations.
For instance, in digital communication systems,
messages which are transmitted could consist of
a random number of packets. But not all the
packets in a batch can be transmitted
successfully, especially in a wireless
communication environment. To model
congestion of sophisticated retrial queues, a type
of mechanism called Bernoulli admission
mechanism is introduced recently by Artalejo
& Atencia (2004) for the continuous time
queueing model and Artalejo et al. (2005) for the
discrete time queueing model. In such an
admission policy each individual blocked
customer is admitted to join the retrial group
with a probability ( 0 1 ) independently.
This consideration can be regarded as a first step
to extend the existing control mechanism for
admission of customers in the classical queues
to retrial queues.
Recent applications to communication
protocols and local area networks show that
there are queueing situations in which the retrial
rate is a constant whenever there is at least a
customer in the orbit. That is, the retrial rate is
(1 0,n ) , where i, j denotes Kronecker's
delta function and n is the number of repeated
customers. This constant retrial policy was
introduced by Fayolle (1986), who investigated
a telephone exchange model as an M/M/1 retrial
queue. In this model, the customers in the retrial
group form a queue and only the customer at the
head of the orbit queue can request service after
a retrial time exponentially distributed with rate
. Farahmand (1990) called this discipline a
retrial queue with FCFS orbit. Since Fayolle
307
(1986), there has been a quick growth in the
literature on retrial queues operating under the
constant retrial policy, see, for instance, Choi et
al. (1993), Martin & Artalejo (1995), Artalejo
(1997), Gomez-Corral (1999), Li & Zhao (2005),
Wang (2006) and Wang & Li (2009). This retrial
policy is a useful device for modeling the retrial
phenomenon in communication and computer
networks where repeated attempts are made by
processor units independently of the number of
messages stored in each node of the network.
Besides, this kind of retrial control policy is also
used for the stability of the ALOHA protocol
and unslotted CSMA/CD (Carrier Sense
Multiple Access with Collision Detection)
protocol in communication systems (see Choi et
al. 1992).
A more practical retrial queue with feedback
occurs in many practical situations: for instance,
multiple access telecommunication systems,
where unsuccessful transmitted messages are
sent again, can be modeled as a retrial queue
with feedback. Choi & Kulkarni (1992) studied
an M/G/1 retrial queue with Bernoulli feedback.
Falin (1990) discussed an M/M/1 retrial queue
with Bernoulli feedback and geometric loss.
Choi et al. (1998) investigated an M/M/C retrial
queue with geometric loss and Bernoulli
feedback. Krishna Kumar et al. (2002) treated
the M/G/1 retrial queues with Bernoulli
feedback and starting failures. Furthermore,
although it is natural in the real world, there are
only a few works take into consideration retrial
phenomena involving the unreliability of the
server. For related literature, interested readers
may refer to Kulkarni & Choi (1990), Yang & Li
(1995), Wang et al. (2001) where a single-server
queue with unreliable server was considered.
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
308
In this paper, we consider an M[X]/G/1 retrial
queue with Bernoulli feedback, Bernoulli
admission mechanism and a server subject to
starting failures. We adopt the constant retrial
policy. This model has potential applications in
the performance of Wireless Local Area
Networks (WLAN) operating under
transmission protocols like the CSMA/CD. In
such a context, messages of a variable length
arrive at the stations and then are divided into a
number of packets in order to be transmitted to
the destination station through wireless channels.
Before transmission, the availability of the
wireless channels will be detected according to
specific transmission protocols such as IEEE
802.11 (see Bianchi 2000). If the channel is
found available, one packet is selected to be
transmitted automatically and the rest are stored
in a buffer (i.e. the retrial orbit). Collisions may
occur in few situations due to imperfect wireless
channel conditions and hidden stations, and they
can be modeled as starting failures in this paper
(see Wang et al. 2007). On the other hand, if the
server is busy, all the packets must be stored in
the buffer and will retry their transmissions later.
Such a model also has potential applications
in transfer model of email system (see Ke &
Chang 2009). The simple mail transfer protocol
(SMTP) is commonly used to deliver the
messages between mail servers. When a mail
transfer program contacts a server on a remote
machine, it forms a TCP connection over which
it communicates. Once the connection is in place,
the two programs follow SMTP that allows the
sender to identify itself, specify a recipient, and
transfer an email message. By having sender
deposits the email in his own mail server, the
mail server can repeatedly try to send the contact
J Syst Sci Syst Eng
message to target server until the target server
becomes operational. It is reasonable to assume
that contacting messages arrive at the mail
server following a Poisson stream. When
messages arrive at the mail server, the mail
server usually employs a spam filter service to
filter the incoming messages to prevent the spam
mails clogging it. After that, one message is
selected to serve and the rest will join to the
buffer. In the buffer, each message waits a
certain amount of time and retries the service.
There is a daemon program implemented at mail
server to manage the service requests from
buffer. Each time it tries but fails, it will wait
another amount of time before trying again. The
target server is the same as sender's mail server
and the sending message will be possibly
feedback to the server to request the receiving
service. Meanwhile, the mail server in reality
could be starting failure due to software
unreliability or other reasons (for example,
virus). Once failed, the mail server will be
recovered (repaired) for a random amount of
time. In this scenario, the buffer in the sender
mail server, the receiver mail server, spam filter
server, the retransmission policy, software
unreliability and recover process correspond to
the retrial orbit, the server, admission control,
feedback, starting failure and repair process,
respectively in the queueing terminology.
The organization of the paper is as follows.
The model under consideration is described in
Section 2. A necessary and sufficient condition
for the system to be stable is obtained in Section
3. Section 4 discusses the steady-state
distribution of the server state and the orbit
length. In Section 5, we show some numerical
examples to illustrate the impact of the
Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng 309

parameters on the system performance. Finally,
in Section 6, conclusions and future work are
given.
2. Model Description
In this section, we describe the batch arrival
M/G/1/1 feedback retrial queue with a server
ak and ck , k=1, 2, 3, satisfy the relationship:
ak = k ck (3)
3) The service process: The service times
{ n , n 1} of the customers are assumed to be
i.i.d. random variables with the probability
distribution function

{ }
subject to starting failures and repairs, which x
will be analyzed in this paper. For any real B ( x ) = P { n x} = 1 exp ( v ) dv
0
number x [0,1], we denote x 1 x
throughout the paper. The model is described as density function b(x), Laplace-Stieltjes
follows. transform ( s ) and first three moments
*

1) The arrival process: The primary calls 1 , 2 and 3 , respectively.

arrive in bulk according to a Poisson process 4) The repair process: Upon arrival, if the
with rate . The number of individual primary server is idle, an arriving customer (primary or
customers arriving in a batch, Y, is equal to retrial) must start (turn-on) the server, which
k ( 1) with probability ck , that is, P {Y = k } takes negligible time. If the server is started
= ck , k=1, 2, . Its probability generating successfully (with a certain probability), the
function (PGF) is given by customer gets service immediately. Otherwise,
the repair for the server commences immediately
( )
C ( z ) = E z Y = n =1 cn z n and the customer must leave for the orbit and
make a retrial at a later time.
2) The admission control rule: Let
We assume that the probability of successful
(0,1] probability of admission for each
commencement of service is for a new
individual customer. Hence an , the probability
customer who finds the server is idle and sees no
that a group of size n 0 joins the system, is
other customer in the orbit (the counterpart of a
given as follows:
customer who starts a busy period in the
standard M/G/1 system) and is for all other
ck (1 ) ,
k
n=0
k =1 new and returning customers. The successive
an = (1)
repair times are i.i.d. random variables with the
c k n 1 k n ,
k n ( ) n 1
probability distribution function D(x), density
k =n
function d(x), Laplace-Stieltjes transform
From (1) we easily get the relationship * ( s ) and first three moments 1 , 2 and 3 ,
between the generating functions of the respectively.
sequences {ck }k =1 and {an }n =0 (see Artalejo 5) Retrial rule: The blocked customers enter
& Atencia (2004)): the retrial group (orbit) in accordance with an
a ( z ) = C ( z + 1 ) (2) FCFS discipline. That is, only the customer at
the head of the orbit queue is allowed for access
Hence, the corresponding factorial moments,
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
310
to the server. Customers only start the retrial
process when the server becomes free and
successive inter-retrial times of any customer are
governed by an arbitrary probability distribution
function T(x), with the corresponding density
function t(x) and its Laplace-Stieltjes transform
T * (s) .
6) Feedback rule: After the customer is
served completely, he will decide either to join
the retrial group again for another service with
probability p or to leave the system forever with
probability q 1 p .
Customers admitted to the system follow the
policy as follows: If the server is idle then one
customer starts its service time and the rest joins
the retrial group, whereas all the admitted
customers go to the retrial group when the server
is busy upon arrival.
The state of the system at time t can be
described by the Markov process {X(t); t 0 }=
{(C(t), N(t), 0 (t ) , 1 (t ) , 2 (t ) ), t 0 }, where
C(t) denotes the server state (0, 1 or 2 according
as the server being free, busy or down,
respectively) and N(t) corresponds to the number
of customers in orbit at time t. If C(t)=0 and
N(t)>0, then 0 (t ) represents the elapsed retrial
time; if C(t)=1, then 1 (t ) corresponds to the
elapsed time of the customer being served; if
C(t)=2 and N(t)>0, then 2 (t ) represents the
elapsed repair time at time t.
The functions ( x), ( x), ( x) are the
conditional completion rates (at time x) for
repeated attempts, for service and for repair,
respectively; i.e. ( x) = t ( x) / (1 T ( x) ) ,
( x) = b( x) / (1 B( x) ) , ( x) = d ( x) / (1 D ( x) ) .
In the following section, we will obtain an
ergodicity condition for the model under
consideration.
J Syst Sci Syst Eng
3. Ergodicity Condition
We first obtain a necessary and sufficient
condition for the system to be stable. To this end,
we discuss the ergodicity of the embedded
Markov chain at the departure/repair completion
epochs at first. Let { n , n N} be the sequence
of epochs at which either a service completion
occurs or a repair period ends. The sequence of
random vectors Qn = (C ( n + ), N ( n + )) forms a
Markov chain, which is the embedded Markov
chain for our queueing system. Its state space is
S= {0} {0, 1, 2, 3, }.
Theorem 1 Let Qn , n 1 , be the orbit length
at the time of either nth customer departure or
repair completion. Then { Qn , n 1 } is ergodic
if and only if
( a11 + p ) + (1 + a11 )
+
1 1
( )
a 1 T * ( 1 ) < 1 (4)
where 1 = (1 a0 ) denotes that there is at
least one customer that is accepted by the server.
Proof. It is not difficult to see that { Qn , n 1 }
is an irreducible and aperiodic Markov chain. To
prove ergodicity, we shall use Foster's criterion:
An irreducible and aperiodic Markov chain is
ergodic if there exists a non-negative function
f(j), j N and >0 such that the mean drift
[ f ( Qn +1 ) f ( Qn ) Qn = j ] is finite for
j = E
all j N and j for all j N , except
perhaps a finite number.
In our case, we consider the function f (j) = j.
Then we have
j = E f ( Qn +1 ) f ( Qn ) Qn = j
Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng 311

( a11 + p ) + (1 + a11 ) < x + dx}, t 0, x 0, n 1

By the method of supplementary variable
+ a1 1, j=0
1 technique, we obtain the following system of
=
( a11 + p ) + 1 (1 + a11 ) equations that govern the dynamics of the



(
+ a1 1 T * (1 ) )1
1, j 1
system behaviour:
dP0 (t )
= 1 P0 (t ) + q Q0 ( x, t ) ( x) dx (5)
Clearly, the inequality <1 is a sufficient dt 0

condition for ergodicity. The same inequality is

t + x + ( x) + 1 Pn ( x, t ) = 0, n = 1, 2,...
also necessary for ergodicity. As noted in
Sennott et al. (1983), we can guarantee (6)
non-ergodicity of the Markov chain { Qn , n 1 },
t + x + 1 + ( x) Qn ( x, t )
if it satisfies Kaplan's condition, namely
j < for all j N and there exits j0 N n
= (1 0,n ) ak Qn k ( x, t ), n = 0,1, 2,... (7)
such that j 0 for j j0 . Notice that, in k =1
our case, Kaplans condition is satisfied because
k N such that rij = 0 for j<ik and i >0, t + x + 1 + ( x) Rn ( x, t )

where R = (rij ) is the one step transition matrix n 1
of {Qn , n 1} . Then the inequality < 1 = (1 1,n ) ak Rn k ( x, t ), n = 1, 2,... (8)
k =1
implies the non-ergodicity of the Markov chain.
Equations (5)-(8) are to be solved under the
Since the arrival stream is a Poisson process, following boundary conditions:

it can be shown from Burkes theorem (see Pn ( 0, t ) = q Qn ( x, t ) ( x)dx
0
Cooper (1981)) that the steady state probabilities

of {(C(t), N(t)), t 0 } exist and are positive if + p Qn 1 ( x, t ) ( x)dx
0
and only if < 1 .
+ Rn ( x, t ) ( x)dx, n = 1, 2,... (9)
0

4. Steady State Distribution Qn ( 0, t ) = an +1 P0 (t ) + (1 0,n )

We define the probability
n
P0 (t ) = P {C (t ) = 0, N (t ) = 0} ak Pn k +1 ( x, t )dx
0
k =1
and the probability densities
+ Pn +1 ( x, t ) ( x)dx, n = 0,1, 2,...
Pn ( x, t ) dx = P{C (t ) = 0, N (t ) = n, x 0 (t ) 0
(10)
< x + dx}, t 0, x 0, n 1 Rn ( 0, t ) = an P0 (t ) + (1 1,n )
Qn ( x, t ) dx = P{C (t ) = 1, N (t ) = n, x 1 (t ) n 1
ak Pn k ( x, t )dx
< x + dx}, t 0, x 0, n 0 k =1
0

Rn ( x, t ) dx = P{C (t ) = 2, N (t ) = n, x 2 (t ) + Pn ( x, t ) ( x)dx, n = 1, 2,... (11)
0
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
312 J Syst Sci Syst Eng

with the normalizing equation: Pn ( 0 ) = q Qn ( x) ( x)dx + p Qn 1 ( x) ( x)dx
0 0

P0 (t ) + Pn ( x, t )dx + Qn ( x, t )dx
+ Rn ( x) ( x)dx, n = 1, 2,... (17)
0 0
n =1 n =0 0

+ Rn ( x, t )dx = 1 (12) Qn ( 0 ) = an +1 P0
0
n =1 n
+ (1 0,n ) ak Pn k +1 ( x)dx

where i , j is Kroneckers delta function. 0
k =1
We assume that the stability condition is
fulfilled in the sequel and so that we can set + Pn +1 ( x ) ( x)dx, n = 0,1, 2,... (18)
0
P0 = limt P0 (t ) and the limiting densities: Rn ( 0 ) = an P0
Pn ( x ) = limt Pn (t , x) , n 1 , x 0 ; n 1
+ (1 1,n ) ak Pn k ( x)dx

Qn ( x) = limt Qn (t , x) , n 0 , x 0 ; k =1
0

Rn ( x) = limt Rn (t , x) , n 1 , x 0 .
+ Pn ( x ) ( x)dx, n = 1, 2,... (19)
0
Let t in Equations (5)-(12), we have

and the normalizing equation is
1 P0 = q Q0 ( x) ( x)dx, (13)
P0 + Pn ( x)dx + Qn ( x)dx
0
0 0
d
Pn ( x) = ( ( x) + 1 ) Pn ( x), n = 1, 2,...
n =1 n =0
(14)
dx
d + Rn ( x)dx = 1 (20)
Qn ( x) = ( 1 + ( x) ) Qn ( x) n =1
0
dx
n To solve the above equations, we define the
+ (1 0,n ) ak Qn k ( x ), n = 0,1, 2,... generating functions for z < 1 :
k =1

(15)
P ( x, z ) = Pn ( x )z n , Q ( x, z ) = Qn ( x )z n ,
d
Rn ( x) = ( 1 + ( x) ) Rn ( x) n =1 n =0
dx
n 1 R ( x, z ) = Rn ( x )z n
+ (1 1, n ) ak Rn k ( x ), n = 1, 2,.... n =1
k =1
(16) The following theorem presents the steady
The steady state boundary conditions are state distribution of thesystem.

Theorem 2 If < 1 then the joint steady state distribution of { N n ; n 1 } is obtained as

P( z ) = P( x, z )dx = P0 [1 T * (1 )]
0

( pz + q ) ( a ( z ) a0 ) ( a ( z ) ) + z ( pz + q ) ( a ( z ) a0 ) ( a ( z ) ) z
* *

1 1
(21)
z ( pz + q ) * ( a ( z ) ) + z * ( a ( z ) ) F ( z )

1 * ( a ( z ) )
Q ( x, z )dx = P0

Q( z) =
0 1 a ( z)
Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng 313

( a ( z ) a0 ) + F ( z ) ( ) ( a ( z ) a0 ) ( a ( z ) )
*
1 (22)

z ( pz + q ) ( a ( z ) ) + z ( a ( z ) ) F ( z )
* *


1 * ( a ( z ) )
R ( z ) = R ( x, z )dx = P0

0 1 a ( z )

z ( a ( z ) a0 ) + F ( z ) ( ) ( a ( z ) a0 ) ( pz + q ) ( a ( z ) ) z
*
1 (23)

z
( pz + q ) *
( a ( z ) ) + z *
( a ( z ) )
F ( z )

where Solving Equation (25) we get

F ( z ) = T * ( 1 ) +

1
(
( a ( z ) a0 ) 1 T * ( 1 ) ) { x
P ( x, z ) = P ( 0, z ) exp 1 x ( u ) du . After
0 }
some algebraic manipulations, we have
1
P0 = 24) P ( x, z ) = 1 T ( x ) e 1 x P ( 0, z ) (31)
q T * ( 1 ) + q ( )(1 + 11 )
Proof. Multiplying Equations (14)-(19) by z n Similarly, according to Equations (26) and (27)
and summing over n, n = 0,1, 2," , we obtain we get
Q ( x, z ) = 1 B ( x ) e (
1 a ( z ) ) x
the following equations: Q ( 0, z ) (32)
P ( x, z ) R ( x, z ) = 1 D ( x ) e (
1 a ( z ) ) x
R ( 0, z )
= ( 1 + ( x ) ) P ( x, z ) (25) (33)
x
Q ( x, z )
We can obtain the following formulae with
= ( a ( z ) + ( x ) ) Q ( x, z ) (26) the help of Equations (31)-(33):
x

R ( x, z ) 0 P ( x, z ) ( x ) dx = P ( 0, z ) T ( 1 )
*
(34)
= ( a ( z ) + ( x ) ) R ( x, z ) (27)
x
0 Q ( x, z ) ( x ) dx = Q ( 0, z ) ( a( z) )
*

P ( 0, z ) = ( pz + q ) Q ( x, z ) ( x ) dx
0 (35)

0 R ( x, z ) ( x ) dx = R ( 0, z ) ( a( z) )
*
(36)
+ R ( x, z ) ( x ) dx 1 P0 (28)
0

1 and
Q ( 0, z ) = { P0 ( a ( z ) a0 ) + ( a ( z ) a0 )
z 1
P( z) = P ( 0, z ) 1 T * ( 1 ) (37)

P ( x, z )dx + P ( x, z ) ( x )dx} 1
0 0
1 1 * ( a ( z ) ) (38)
(29) Q ( z ) = Q ( 0, z )
a ( z )
R ( 0, z ) = P0 ( a ( z ) a0 ) + ( a ( z ) a0 )
1 1 * ( a ( z ) )
R ( z ) = R ( 0, z )
P ( x, z )dx + P ( x, z ) ( x ) dx
0 0
a ( z )

(30) (39)
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
314 J Syst Sci Syst Eng

Combining Equations (29), (30), (34) and (37) and on simplification we have
P ( 0, z ) *
Q ( 0, z ) =
P0
z
( a ( z ) a0 ) +
P
*
(
T ( 1 ) + ( a ( z ) a0 ) 1 T ( 1 ) ) (40)
0 1
P ( 0, z ) *
R ( 0, z ) = P0 ( a ( z ) a0 ) +
P0

T ( 1 ) + ( a ( z ) a0 ) 1 T ( 1 )
1
*
( ) (41)

Combining Equations (28),(35),(36),(40) and (41), we obtain the following result:
( pz + q ) ( a ( z ) a ) * ( a ( z ) ) + z ( a ( z ) a ) * ( a ( z ) ) z
P ( 0, z ) = P0 0 0 1
(42)

z
( pz + q ) *
( a ( z ) ) + z *
( a ( z ) )
F ( z )

Substituting Equations (40), (41) and (42) into (37)-(39), we finally get (21)-(23).
At this point, the only unknown is P0 which can be determined using the normalizing condition,
P0 + P (1) + Q (1) + R (1) = 1. Thus, setting z = 1 in Equations (21)-(23), applying LHospitals rule
whenever it is necessary, and using the normalizing condition we can get the following result after
rearrangement
q T * ( 1 ) + q ( a )(1 + 11 )
P0 =1

1 (
a 1 T * ( 1 ) ( a11 + p ) (1 + a11 )
1 1
)
which yields Equation (24).
Corollary 1 In the steady state, the probability generating function of the number of customers in the
system under steady state is K ( z ) = P0 + P ( z ) + zQ ( z ) + R ( z ) , which has the following expression:

q (1 z ) * ( a ( z ) ) ( ) ( a ( z ) a0 ) * ( a ( z ) ) F ( z ) 1 T * ( 1 ) [1 a( z )]
K ( z ) = P0

(43)
1 a ( z) z ( pz + q ) * ( a ( z ) ) + z * ( a ( z ) ) F ( z )

The mean number of customers in the system Ls under steady state conditions is obtained by
differentiating (43) with respect to z and evaluating at z = 1 ,
Ls = K ' (1)
1
( ) a1 a12 + 21 + 2 (1 + a11 ) 1 +
1

(1 T *
( 1 ) ) + 21 1
1
(
a 1 T * ( 1 )

)
= 1

2 ( ) q (1 + a11 ) + q 1 1 a1 1 T * ( 1 )

( )

a1 a1 ( 2 + 2 ) + 2 ( p1 + 1 ) +
1
2

(1 T *
)
(1 ) [ ( a11 + p) + (1 + a11 )]

+

*
( )
2 1 a1 1 T (1 ) ( a11 + p ) (1 + a11 )
1
Proof. Based on the results of P0 , P ( z ) , Q ( z ) and R ( z ) obtained in Therorem 2, it is readily to
derive the expression of K ( z ) .
Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng 315

Remark 1 If q = = = = 1 , then Equation (43) becomes

K (z) =
(1 z ) * ( C ( z ) ) {1 c + (1 T ( ))}
1 1
*

(44)
(
* ( C ( z ) ) T * ( ) + C ( z ) 1 T * ( ) z
)
which is the probability generating function of (1 z ) * ( z ) 1 ( + )
K (z) =
the number of customers in the system similar z ( + ) * ( z )( z + )
to our model but without admission control, (46)
feedback, and starting failures. Further, if we
assume that the arrival process is not batch The above Equation (46) coincides with
arrival process, i.e., c1 = 1, ck = 0, k = 2,3," , the Equation (2.13) in Choi et al. (1993).
then K ( z ) becomes When the service time has an exponential
distribution with parameter l, Equation (46)
(1 z ) * ( z ) 1 T * ( ) agrees with Proposition 1 in Fayolle (1986).
K (z) = In the sequel we will discuss some
(
z * ( z ) T * ( ) + z 1 T * ( )
) performance measures for the system in steady
(45) state. Let U be the steady state probability
The above result agrees with that the server is busy, I, the steady state
Gomez-Corral (1999) without feedback and probability that the server is idle during the
starting failures. retrial time, R , the steady state probability
Remark 2 Further, if the retrial time that the server is under repair and D , the
distribution is exponential with parameter v, steady state probability that the server is down
or idle. From Theorem 2, we obtain:
then T * ( ) = . In this case (45) yields
+
a11
U = Q (1) = ,
q

(1 T *
( 1 ) ) a 1 1 + a11 +

a q
1 1
I = P (1) = ,
q T * ( 1 ) + q ( )(1 + 11 )

1 ( ) (1 a0 ) ( a 1 1 q ) + a1T * ( 1 ) + ( ) a1
R = R (1) = ,
q T * ( 1 ) + q ( )(1 + 11 )

a11
D = P0 + P (1) + R (1) = 1 .
q
Define H ( z ) = P0 + P ( z ) + Q ( z ) + R ( z ) . Then H ( z ) represents the probability generating
function for the number of customers in the orbit. Using Equations (21)-(24) and simplifying, we get
the following expression:
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
316 J Syst Sci Syst Eng

*
T ( 1 ) z ( pz + q ) ( a ( z ) ) z ( a ( z ) )
* *

H ( z ) = P0 1

z ( pz + q ) ( a ( z ) ) + z ( a ( z ) ) F ( z )

* *

( a ( z ) a0 ) (1 * ( a ( z ) ) ) + z (1 * ( a ( z ) ) ) + F ( z)( z )
+ (47)
(1 a( z ) ) { z ( pz + q ) * ( a ( z ) ) + z * ( a ( z ) ) F ( z )}
where
( z ) ( ) (1 a ( z ) ) z * ( a ( z ) ) ( pz + q ) * ( a ( z ) ) 1 * ( a ( z ) )

( ) ( a ( z ) a0 ) * ( a ( z ) ) + 1 + 1 * ( a ( z ) ) ( )( pz + q )

1
( a ( z ) a0 ) * ( a ( z ) ) z.

Hence, the mean number of customers in the orbit is given by
Lq = H ' (1)
1
( ) 2 (1 + a11 ) qa1

( )
( )
1 T * ( 1 ) p a11 + a1q ( 21 + a12 ) 2 a1 p 1 1 1 a1 1 T * ( 1 )

=


( )
2 ( ) q (1 + a11 ) + q 1 1 a1 1 T * ( 1 )


2a1a2
1
( )
1 T * (1 ) [ ( a11 + p) + (1 + a11 ) ] + 2 a12 ( 2 + 2 ) + 2 a1 ( p11 + 1 )
+

( )
2 1 a1 1 T * (1 ) ( a11 + p ) (1 + a11 )
1

Let Ws be the average time a customer system busy period of our model is obtained in
spends in the system under steady state. Due to direct way by the theory of regenerative
Littles formula, we have processes which leads to the limiting
Ls probabilities
Ws = .
a1 {
P0 = lim P ( C ( t ) , N ( t ) ) = ( 0, 0 )
t
}
An interesting performance measure in the
as follows:
retrial context is the mean of the system busy
E (T00 )
period. The system busy period LB is defined P0 = ,
E (T00 ) + E ( LB )
as the period that starts at an epoch when an
arriving customer finds an empty system and where T00 is the amount of time in a
ends at the next departure epoch at which the regenetative cycle during which the system is
system is empty. The mean length of the in the state ( 0, 0 ) . It is clear that
Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng 317

1 1
E (T00 ) = 1 / 1 . So that E ( LB ) = 1 , which arrives at
P0

1


(
a11 + a11 + q ( ) 11 + 1 a1 q 1 T * (1 ) )
E ( LB ) =

1

( )
1 ( a11 + p ) (1 + a11 ) 1 a1 1 T * (1 )

5. Numerical Illustrations and three curves which in decreasing order

In this section, we present some numerical correspond to = 20,30, 60, respectively.
examples to study the effect of the parameters
on the main system performance. To this end,
we consider that the service time distribution,
the retrial time distribution and the repair time
distribution are all exponentially distributed with
Parameters , and , respectively. We also
assume that the batch size follows a geometric
1
distribution with mean (1 ) such that
ck = (1 ) , k 1, [ 0,1) .
k 1

By (1) we have

a1 =
1 Figure 1 System empty probability P0 vs.
and
(1 )(1 ) In Figure 2, the variation of P0 , the
a0 = , 1 a0 = .
1 (1 ) 1 (1 ) probability of no customers in the system, is
We next concentrate our attention on the idle studied for increasing values of the feedback
probability of the server. Since we deal with a probability p . We assume that the parametric
retrial queue with starting failures, feedback and values of the system are ( , , , , , , ) =
Bernoulli admission control mechanism, we will ( 0.3, 20, 0.5, 0.6, 0.8, 20, 0.5) . We have presented
put the emphasis on the influence of the three curves in decreasing order which
feedback probability p, the probability of correspond to = 20,30, 60. It is observed that
the customers getting succeful commencement the larger retrial rate , the larger the
of service for new and retrial customers and probability of no customers in the system P0 .
admission control probability . Finally, to understand the influence of the
Figure 1 shows the effect of , the admission control probability on the
probability of the customers getting successful probability P0 , we fix the parameters of the
commencement of service, on P0 . We now take system as ( , , , , p, , ) = (0.3,30,0.5,0.6,
0.4,0.8,20,0.5) and = 5,10,15, respectively.
( , , , p, , , ) = ( 0.3, 20, 0.6, 0.4, 0.8, 20, 0.5)
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
318
We observe that the probability P0 is
increasing when the service rate increases. That
is, when the admission control probability is
fixed, the larger the service rate, the larger the
probability P0 .
Figure 2 System empty probability P0 vs. p
Figure 3 System empty probability P0 vs.
6. Conclusions and Future Work
A batch arrival retrial queue with starting
failures, feedback and admission control has
been investigated in this paper. It is assumed that
the primary customers arrive in bulk according
to a Poisson process and are subject to
admission control and starting failures. The
J Syst Sci Syst Eng
retrial time is governed by an arbitraty
distribution and that the customer at the head of
the orbit queue is allowed for access to the
server. The necessary and sufficient condition
for the system to be stable is obtained. Some
performance measures, such as mean system
size, mean orbit size, the probability of the
server being under repair, the server utilization,
etc. are obtained. The impact of the parameters
on the performance measures are illustrated.
It should be noted that the admission policy
used in this paper is just a primary attempt to
describe the existing control mechanism for
admission of customers in retrial queues. Further
work should be continued to study more
practical admission policies other than the
simple Bernoulli type. Another assumption that
retrial times are only started when the server
becomes free may be relaxed by considering
different retrial policies arising from practical
problems.
Acknowledgments
The authors thank the anonymous referees
for their constructive suggestions and comments
on the earlier version of this paper.
References
[1] Artalejo, J.R. (1997). Analysis of an M/G/1
queue with constant repeated attempts and
server vacations. Computers and Operations
Research, 24: 493-504
[2] Artalejo, J.R. (1999). A classified
bibliography of research on retrial queues:
progress in 1990-1999. Top, 7 (2): 187-211
[3] Artalejo, J.R. (1999). Accessible
bibliography on retrial queues. Mathematical
and Computer Modelling, 30: 1-6
Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
J Syst Sci Syst Eng
[4] Artalejo, J.R. & Atencia, I. (2004). On the
single server retrial queue with batch arrivals.
Sankhya, 66: 140-158
[5] Artalejo, J.R., Atencia, I. & Moreno, P.
(2005). A discrete-time Geo[ ] /G/1 retrial
X
queue with control of admission. Applied
Mathematical Modelling, 29: 1100-1120
[6] Artalejo, J.R. & Gomez-Corral, A. (2008).
Retrial Queueing Systems: A Computational
Approach. Springer, Berlin
[7] Bianchi, G. (2000). Performance analysis of
the IEEE 802.11 distributed coordination
function. IEEE Journal on Selected Areas in
Communications, 18: 535-547
[8] Choi, B.D., Shin, Y.W. & Ahn, W.C. (1992).
Retrial queues with collision arising from
unslotted CSMA/CD protocol. Queueing
Systems, 11: 335-356
[9] Choi, B.D., Rhee, K.H. & Park, K.K. (1993).
The M/G/1 retrial queue with retrial rate
control policy. Probability in Engineering
and Informational Sciences, 7: 26-46
[10] Choi, B.D. & Kulkarni, V.G. (1992).
Feedback retrial queueing systems. In: Bhat,
U.N., Basawa, I.V. (eds.), Stochastic Model
Related Fields, pp. 93-105. Oxford
University Press, New York
[11] Choi, B.D., Kim, Y.C. & Lee, Y.W. (1998).
The M/M/C retrial queue with geometric
loss and feedback. Computers and
Mathematics with Applications, 36: 41-52
[12] Cooper, R.B. (1981). Introduction to
Queueing Theory. North-Holland, New York
[13] Falin, G.I. (1990). A survey of retrial
queues. Queueing Systems, 7: 127-167
[14] Falin, G.I. & Templeton, J.G.C. (1997).
Retrial Queues. Chapman & Hall, London
[15] Farahmand, K. (1990). Single line queue
319
with repeated demands. Queueing Systems, 6:
223-228
[16] Fayolle, G. (1986). A simple telephone
exchange with delayed feedbacks. In: Boxma,
O.J., Cohen, J.W. & Tijms, M.C. (eds.),
Teletraffic Analysis and Computer
Performance Evaluation, pp. 245-253.
North-Holland, Amsterdam
[17] Gomez-Corral, A. (1999). Stochastic
analysis of a single server retrial queue with
general retrial times. Naval Resrarch
Logistics, 46: 561-581
[18] Ke, J.-C. & Chang, F.-M. (2009). Modified
vacation policy for M/G/1 retrial queue
with balking and feedback. Computers and
Industrial Engineering, 57: 433-443
[19] Krishna Kumar, B., Pavai Madheswari, S.
& Vijayakumar, A. (2002). The M/G/1
retrial queue with feedback and starting
failures. Applied Mathematical Modelling,
26: 1057-1075
[20] Kulkarni, V.G. & Choi, B.D. (1990).
Retrial queues with server subject to
breakdowns and repairs. Queueing Systems,
7: 191-208
[21] Kulkarni, V.G. & Liang, H.M. (1997).
Retrial queues revisted. In: Dshalalow, J.H.
(eds.), Frontiers in Queueing, pp. 19-34.
CRC Press, Boca Raton
[22] Li, H. & Zhao, Y.Q. (2005). A retrial
queue with a constant retrial rate server
downs and impatient customers. Stochastic
Models, 21: 531-550
[23] Gomez-Corral, M. & Artalejo, J.R. (1995).
Analysis of an M/G/1 queue with two
types of impatient units. Advances in
Applied Probability, 27: 840-861
[24] Sennott, L.I., Humblet, P.A. & Tweedi,
 Wang and Zhou: A Batch Arrival Retrial Queue with Starting Failures, Feedback and Admission Control
320
R.L. (1983). Mean drifts and the
non-ergodicity of Markovchains. Operations
Research, 31: 783-789
[25] Wang, J., Cao, J. & Li, Q. (2001).
Reliability analysis of the retrial queue with
server breakdowns and repairs. Queueing
Systems, 38: 363-380
[26] Wang, J. (2006). Reliability analysis of
M/G/1 queues with general retrial times
and server breakdowns. Progress in Natural
Science, 5: 464-473
[27] Wang, J. & Zhao, Q. (2007). Discrete-time
retrial queue with general retrial times and
starting failures. Mathematical and Computer
Modelling, 45: 853-863
[28] Wang, J. & Li, J. (2009). A single server
retrial queue with general retrial times and
two-phase service. Journal of Systems
Science and Complexity, 22: 291-302
[29] Yang, T. & Li, H. (1995). On the
steady-state queue size distribution of the
discrete-time Geo/G/1 queue with repeated
customers. Queueing Systems, 21: 199-215
J Syst Sci Syst Eng
[30] Yang, T. & Templeton, J.G.C. (1987). A
survey on retrial queues. Queueing Systems,
2: 201-233
Jinting Wang is a full professor at Beijing
Jiaotong University. He received his Ph.D.
Degree in Operations Research and Cybernetics
from Chinese Academy of Sciences (CAS) in
2000. His research interests include issues
related to queueing theory, reliability and
applied probability. He is author or co-author of
several publications in Queueing Systems:
Theory and Applications (QUESTA), Journal of
Multivariate Analysis, European Journal of
Operational Research, etc.
Peng-Feng Zhou is a Master student in the
Department of Mathematics, School of Science
at Beijing Jiaotong University. He received his
B.Sc. in Mathematics from Beijing Jiaotong
University. His research area includes retrial
queues and reliability.