i

i
International Journal of Information and Management Sciences
Volume 19, Number 4, pp. 621-634, 2008

The M/M/1 Queue with Single Working Vacation

Naishuo Tian Xinqiu Zhao
Yanshan University Yanshan University P. R. China P. R. China
Kaiyu Wang
Yanshan University P. R. China
Abstract
In this paper, we study an M/M/1 queue with single working vacation. Using quasi birth and death process and matrix-
geometric solution method, we give the distributions for the number of customers and the virtual time in the system in steady
state. Furthermore, we obtain expected busy period and expected busy cycle. Finally, we get the stochastic decomposition
structures of stationary indices.

Keywords: Working Vacation , Matrix-Geometric Solution, Busy Cycle, Stochastic Decomposition.

1. Introduction
The vacation queues have been investigated extensively because of their applications in computer
systems, communication networks, production managing and so forth. In
a classical vacation queue, a server may completely stop service or do some additional work during a
vacation. Proposing of various vacation policies provides more flexibility
for optimal design and operation control of the system. The details can be seen in the
monographs of Takagi (1991) and Tian and Zhang (2006), the surveys of Doshi (1986) and Teghem
(1986).
Servi and Finn (2002) introduced a class of semi-vacation policies: the server works at a lower rate
rather than completely stoping service during a vacation. Such a vacation
is called a working vacation(WV). Part of servers keep the system operating in a lower
Received February 2007; Revised April 2007; Accepted November 2007. Supported by National Natural Science Foundation,
China (Grant No. 10671170).
“M19N45” — 2008/11/5 — 17:15 — page 621 — #1
i
i
i
i
i
i
622 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008

rate and the other parts break or accomplish else assistant work during a vacation. If service rate
degenerates into zero in a working vacation, then the working vacation queue becomes a classical
vacation queue model. Therefore, the working vacation queue is an extension of classical vacation queue.
Tian and Zhang (2006), Zhang and Tian (2003) studied multiserver queues with partial servers vacation
where vacation servers completely stop service.
There is quite a difference between working vacation queue and classical vacation queue. During a
vacation, customers in the former may finish service and depart the system, however, customers in the
latter can impossibly depart the system. On the other hand, during a vacation, the number of customers in
the former can increase or decrease, however, the number of customers in the latter can only increase.
Therefore, the working vacation models have more complicated modalities and the analysis of this kind of
models is more difficult than classical vacation queue.
Servi and Finn (2002) studied an M/M/1 queue with multiple working vacations, and obtained the
PGF of the number of customers in the system and the LST of waiting time distribution, and applied
results to perform analysis of gateway router in fiber communication networks. Later, based on [9], Liu,
Xu and Tian (2006) gave simple explicit expressions of distributions for the stationary queue length and
waiting time which have intuitionistic probability sense. Furthermore, the authors got stochastic
decomposition structures of stationary indices, derived the distributions of additional queue length and
additional delay,and obtained expected regular busy period and expected busy cycle. Kin, Choi and
Chae(2003), Wu and Takagi (2006) generalized the work of [9] to an M/G/1 queue with multiple working
vacations. Baba (2005) investigated a GI/M/1 queue with multiple working vacations.
In this paper, we study an M/M/1 queue with single working vacation. Single woring vacation policy
has practical application background in managing science. When the number of customers in the system
is relatively few, the system is in a lower rate operation state in order to economize operation cost and
energy consumption. Using quasi birth and death process and matrix geometric solution method, we give
the distributions for the number of customers and the virtual time in system in steady state. Furthermore,
we obtain expected busy period and expected busy cycle. Finally, we get the stochastic decomposition
structures of stationary indices.
The rest of this paper is organized as follows. In Section 2 we describe the quasi birth and death
process model of the system, and get the exact expression of rate matrix
“M19N45” — 2008/11/5 — 17:15 — page 622 — #2
i
i
i
i
i
i
The M/M/1 Queue with Single Working Vacation 623

which assures that analytic solutions exist in various systems. In Section 3 and Section
4 we discuss the number of customers in the system and the virtual time of a customer
respectively, and give their stochastic decomposition structures,and obtain the distributions of additional
queue length and additional delay. Finally, we derive the computing
formulae of expected regular busy period and expected busy cycle in Section 5.
2. Quasi Birth and Death Process Model
Introducing a single working vacation policy into a classical M/M/1 queue with
arrival rate λ and service rate μ
b

in a regular busy period[3]. The server begins a working

vacation of random length V at the instant when the queue becomes empty, and vacation
duration V follows an exponential distribution with parameter θ. During a working
vacation an arriving customer is served at a rate of μ
v

. When a vacation ends, if there

are customers in the queue, the server changes service rate from μ
v

to μ
b

, and a regular
busy period starts. Otherwise, the server enters idle period, and a new regular busy
period starts when a customer arrival occurs. Working vacation V is a operation period
in a lower rate. When the number of customers in the system is relatively few, we set a
lower rate operation period in order to economize operation cost together with serving
customers. Therefore, this single working vacation policy has practical significance in
optimal design of the system.
We assume that interarrival times, service times, and working vacation times are
mutually independent. In addition, the service order is first in first out(FIFO).
Let Q(t) be the number of customers in the system at time t and let
J(t) =
{
0, the system is in a working vacation period at time t,
1, the system is not in a working vacation period at time t.
then {Q(t),J(t)} is a Markov process with the state space
Ω=
{
} (k,j) : k ≥ 0, j = 0,1
.
where state (0,1) denotes that the system is in idle period; state (k,1), k ≥ 1 indicates
that the system is in regular busy period; state (k,0), k ≥ 0 indicates that the system is
in working vacation period and there are k customers in the queue. The state transition
diagram of system is as follows:
“M19N45” — 2008/11/5 — 17:15 — page 623 — #3
i
i
i
i
i
i
624 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008
00 10 20 30
01 11 21 31
Ă
v
vvb
Ă
b b Figure 1. State transition diagram of system.
Using the lexicographical sequence for the states, the infinitesimal generator can be written as
Q ̃
=

A
00

CB
10

AC
BABC
AC
...

.
where
A
00

...
...
[
−(λ + θ) θ
0 −λ
]
,B
10

[
μ
v

]
=
=
μ
b

,
B=
0
0[
μ
v

]
,C=
[
λ0
0λ
]
,
A=
0
0[
−(λ μ b
+θ+μ
v

]
.
The structure of
)θ
0 −(λ + μ
b

)Q ̃
indicates that {Q(t),J(t)} is a quasi birth and death process, see Neuts (1981) or
Latouche and Ramaswam (1999). To analyze this QBD process, it is necessary to solve for the minimal
non-negative solution of the matrix quadratic equation
R2B + RA + C = 0 (1)
and this solution is called the rate matrix and denoted by R. Obviously, we have
Theorem 1. If ρ = λ(μ
b

)−1 < 1, the matrix equation (1) has the minimal non- negative solution
R=

r
θr
“M19N45” — 2008/11/5 — 17:15 — page 624 — #4
i
(1 − r) 0 ρ
μ
b

. (2)
i
i
i
i
i
The M/M/1 Queue with Single Working Vacation 625

where
r=
1
)
and 0 <r< 1.
Proof. Because the matrices A, B, C of (1) are all upper triangular, we can assume
that R has the same structure as
R=
]
.
Substituting R2 and R into (1) yields the following set of equations:

[
r
11

r
12 0 r
22

μ
v

r2
11

− (λ + θ + μ
v

)r
11

+ λ = 0,
μ
b

r2
22

− (λ + μ
b

)r
22

+ λ = 0,
(3)
μ
b

r
12

(r
11

+r
22

)+θr
11

− (λ + μ
b

)r
12

= 0.
To obtain the minimal non-negative solution of (1), taking r
11

= r(the other root is

greater than 1) in the first equation and r
22

= 1)in the second

equation of (3). Using elementary method, we can prove that 0 <r< 1. Substituting r
and ρ into the last equation of (3), we get
r
12

= ρ(the other root is r

22

θr
.
Because r satisfies the following equation
μ
v

=
r
2

− (λ + θ + μ
v

)r + λ = 0,
divided both sides of this equation by r, we get
λ+θ+μ
v

λ
,
equivalently, we have
θ
(1 − r) =
. (4)
Theorem 2. The QBD process {Q(t),J(t)} is positive recurrent if and only if ρ < 1.
Proof. Based on the theorem 3.1.1 of Neuts(1981), the QBD process {L(t),J(t)}
is positive recurrent if and only if the spectral radius SP(R) of the rate matrix R is less
“M19N45” — 2008/11/5 — 17:15 — page 625 — #5
i
2μ
v

(
λ+θ+μ
v

1−r
−
+μ
v

μ
b

√
(1 − r)
(λ + θ + μ
v

=
λ
r
r
) 2 − 4λμ
v
i
i
i
i
i
626 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008

than 1, and set of equations (x

0

,x
1

,x
2

,x
3

)B[R] = 0 has positive solution, where

B[R] =

[
A
00 B
10

C
]
=

−(λ + θ) θ 0 − λ μ
v

λ0λ
0
RB + A
0−
λ
(5)
B[R] is an irreducible and aperiodic generator with finite state. Therefore, (x
0

)B[R] =
0 has positive solution (for example, the balance probability vector of B[R] is a positive
solution). Thus, process {L(t),J(t)} is positive recurrent if and only if
SP(R) = max(r, ρ) < 1.
Note that 0 <r< 1, the above relation means that ρ < 1.
3. Queue Length Distribution
If ρ < 1, let (Q, J) be the stationary limit of the QBD process {Q(t),J(t)}. Let
π
k

,x
1

,x
2

,x
3

= (π
k0

,π
k1

), k ≥ 0, π
kj

= P{Q = k,J = j}, k ≥ 0,j = 0,1.

Theorem 3. If ρ < 1, the joint probability distribution of (Q, J) is

π
k0

π
01

=π
00

rk, k ≥ 0, θ
,
π
k1

=
(6)
=π
00

where
π
00

[
θr
=
(1 − r)(1 − ρ)
.
Proof. With the matrix-geometric solution method (see [5] or [7]), we have
π
k

)Rk−1, k ≥ 1. (7)
and π
0

= (π
k0

,π
k1

)=(π
10

,π
11

and π
1

satisfy the set of equations

(π
00

,π
01

,π
10

,π
11

)B[R]=0.
“M19N45” — 2008/11/5 — 17:15 — page 626 — #6
i
λ
π
00

1−ρ+
μ
b

(1 − r)
λ
θ
(1 − r)(1 − ρ) +
∑
k−1 j=0
rjρk−1−j +

−μ
v

μ
b

0 0 −μ
b

μ
b

(1 − r)
θr
θr
μ
b
ρk−1

+
]
, k ≥ 1.
μ
b

r
θ
(1 − r)
λ
r

i
i
i
i
i
=0
Taking π
00

as a known constant, we get

π
0

= (π
00

,π
01

)=π
00

(1,
θ
can be deter-
mined by the normalization condition.
With (6), the probabilities of the server in various state are as follows, respectively
P {J = 0} =
∞∑
k=0

π
k0

=
1−ρ
.
Now, we give stochastic decomposition structure of the number of customers Q in
system in steady-state
Theorem 4. If ρ < 1 and μ
b

>μ
v

, the number of customers Q in system can be decomposed

into the sum of two independent random variables: Q = Q
0

+Q
d

, where
The M/M/1 Queue with Single Working Vacation 627

Substituting B[R] in (5) into the above relation, we obtain

−(λ + θ)π
00

+μ
v

π
10

+μ
b

π
11

=0
θπ
00

− λπ
01

=0
λπ
00

−
λ
)
.
Note that
Rk =

, k ≥ 1.
Substituting (π
10

rk
θr
,π
11

) and Rk−1 into (7), we obtain (6). Finally, π

00

,
P{the server is in idle period} = π
01

=
θ
(8)
P{the server is in regular busy period } =
∞∑
k=1

π
k1

=
θr
“M19N45” — 2008/11/5 — 17:15 — page 627 — #7
i
1−ρ+
1−ρ+
=0
λπ
01

+
(
λ
rjρk−1−j
0 ρk
λ
λ
θ
θ
(1 − r)(1 − ρ) +
(1 − r)(1 − ρ) +
1−ρ+
μ
b

(1 − r)
r
μ
b

), π
1

r
(1 − r)
λ
r
π
10

λ
(1 − r)(1 − ρ)
θ
−μ
v

(1 − r)(1 − ρ) +
k−1∑
j=0

+
)
π
10
= (π
10

θr
μ
b

μ
b

μ
b

(1 − r)
−μ
b

(1 − r)
(1 − r)
θr
θr
,π
11

π
11

)=π
00

+
+
μ
b

(1 − r)
μ
b

μ
b

θ
θ
θr
(1 − r)
(1 − r)
(
r,
μ
b

+
(1 − r)
μ
b

θ
θ
(1 − r)
i
i
i
i
i
628 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008

Q
0

is the number of customers of a classic M/M/1 queue in steady-state and follows a geometric
distribution with parameter 1 − ρ; Additional number of customers Q
d

has a
modified geometric distribution
P{Q
d

K∗
θ+λ
(9)
where
K
= k} =
]
−1

.
Proof. With (6), the probability generating function of Q can be written as
Q(z) =
∗

=
π
00

∞∑
∞∑
k=0
k=0

zkπ
k1
=π
00

zkπ
k0

+
[
θ
]
=
1−ρ
]
.
Note that
θ
.
Q(z) can be rewritten as
Q(z) =
1−ρ
]
. (10)
On the other hand, from (4), we have
θr
.
Substituting the above relation into (10), we have
Q(z) =
1−ρ
]
.
In the above expression, note that
(K∗ )
−1

=1−ρ+
θ
1 − ρz
λ
+
K∗
λ
1 − rz
ρ=
[
θ
λ
“M19N45” — 2008/11/5 — 17:15 — page 628 — #8
i
1
1 − ρz
(1 − r)(1 − ρz) +
μ
b

θ
(1 − r)
=1−r+
θ
=
λ+θ
(1 − r)
(1 − r)(1 − ρ)
+
,
1 − ρz
K
μ
b

λ
∗

(1 − r)
[
λ+θ
θr
1 − rz
1−r
(1 − r) + r
K∗
λ
λ
λ
μ
b

[
λ+θ
(1 − r) + r − ρ +
(1 − r)(1 − ρ) +
(1 K∗
(
1−
μ
v

− r), k = 0,
(1 − r)
1 − rz
(1 − ρz) = (1 − r)+(r − ρ)
(1 − r) +
=
λ
z
1 − rz
1−r
[
θ+λ
λ
(1 − r) + r
(
1−
1 − ρz
λ
=ρ−r
μ
b

(1 − ρz) +
1
(
r−ρ+
)
(1 − r)rk, k ≥ 1.
(1 − r) + r
μ
v

μ
b

+
)
.
μ
v

μ
b

μ
b

(
1−
μ
b

μ
b

θ
(1 − r)
θr
1 − ρz
μ
b

θr
μ
b

μ
v

μ
b

(
1−
z
(1 − r)
(1 − r)
θr
θr
)
z(1 − r)
+
1 − rz
μ
v

μ
b

μ
b

z(1 − r)
θ
)
z(1 − r)
1 − rz
z(1 − r)
)
1 − rz
1 − rz
+
μ
b

θ
(1 − r)z
i
i
i
i
i
The M/M/1 Queue with Single Working Vacation 629

The above equation indicates

Q
d

(z) = K∗
[
λ+θ
]
.
is a PGF. Expanding Q
d

(z) into power series of z, we get the distribution of additional

number of customers Q
d

.
With the stochastic decomposition structure in theorem 4, we can easily get means
E(Q
d

) = K∗
(
1−
μ
v

.
4. Virtual Time Distribution
Denote the virtual time of a customer in the system by W, we have the following
stochastic decomposition results
Theorem 5. If ρ < 1 and μ
b

>μ
v

, the virtual time W can be decomposed into the

sum of two independent random variables: W = W
0

is the virtual time of a customer

in a corresponding classic M/M/1 queue and has an exponential distribution
with parameter μ
b

+W
d

, where W
0

(1 − ρ); Additional delay W

d

has the LST

Wd∗

] (s) = K∗
[
(
θ
. (11)
where
σ=
λ
(1 − r).
Proof. The classical relationship between the PGF of Q and the LST of virtual
time W is
Q(z) = W ∗
)
.
From theorem 4, the PGF of the number of customers Q can be written as
Q(z) =
(
λ(1 − z)
1−ρ
]
. (12)
Taking z = 1 − s/λ in (12), note that
1−ρ
“M19N45” — 2008/11/5 — 17:15 — page 629 — #9
i
1 − ρz
1 − rz
1 − ρz
,
1−r
μ
b

)
r
∣ ∣ ∣
z=1−s/λ

∣ ∣ ∣
z=1−s/λ

K∗
1−r
λ
[
λ+θ
λ
+
, E(Q) =
λ
(1 − r) + r
μ
v

μ
b

=
=
(1 − r) + r
)
(1 − r) +
r
μ
b

λ
r
(1 − r) + s
μ
b

(1 λ
− ρ) + s
r
(1 − r)
(1 − ρ)
(
1−
1−ρ
(
1−
(
1−
ρ
μ
v

μ
b

=
)
z(1 − r)
μ
v

+ K∗
μ
b

μ
v

μ
b

σ+s
1 − rz
)
z(1 − r)
σ
)
σ
σ+s
1 − rz
(
1−
.
μ
v

μ
b

)
r
1−r
i
i
i
i
i
630 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008

Substituting the above results into (12), we have

W ∗ (s) =
μ
b

(1 − ρ)
]
=
μ
b

(1 − ρ)
]
.
Note that (4) and the proof of theorem 4, we give the equivalent expression of K∗
(K∗ )−1 =
(
θ
)
,
this verifies that
Wd∗

(s) =
[(
θ
]
.
is a LST.
The result of theorem 5 indicates that additional delay W
d

equals zero with proba- bility

δ
1
= K∗
(
θ
)
(1 − r),
and follows exponential distribution with parameter σ with probability δ
2

. It is
easy to obtain
E(W
d

)=K
∗

(
1−
μ
v

.
5. Busy Period Analysis
The duration in which a server works at a rate of μ
b

continuously is called a regular busy period,

denoted by B. A regular busy period can start at the instant when a working vacation or an idle period
finishes. A busy cycle can be composed of a working vacation V , idle period after this V finishes (if
exists) and subsequent regular busy period
B.
Note that there is just a working vacation V in a busy cycle C, and E(V ) = θ−1. Using the limiting
theorem of alternative renewal process(see theorem 3.4.4 in [8]) and the first expression in (8), we have
P{J = 0} =
E(V )
μ
b
“M19N45” — 2008/11/5 — 17:15 — page 630 — #10
i
μ
b

μ
b

)
r
(1 − ρ) + s
(1 − ρ) + s
λ(1 − r)
E(C)
=
=
λ
K∗
K∗
λ
1
1−ρ+
+
E(Q
d

[
λ+θ
[(
θ
λ
μ
v

μ
b

+
λ
λ
)
(1 − r) +
μ
v

μ
b

+
), E(W) =
λ
λ
θ
(1 − r) + r
+
μ
v

)
(1 − r) +
μ
b

(1 − r)(1 − ρ) +
μ
v

μ
b

)
(1 − r) +
(
1−
1−ρ
(
1−
(
1−
μ
v

μ
b

μ
b

(
1−
μ
v

(1 − ρ)
μ
b

)
σ
μ
v

μ
b

σ+s
μ
b

1
θ
)(
1−
μ
v

μ
b

(
1+
)
σ
+K
σ+s
λ
s
1−r
∗

)
σ
r2
(
1−
σ+s
)
= 1−δ
1

μ
v

μ
b

,
)
r
λ(1 − r)
i
i
i
i
i
The M/M/1 Queue with Single Working Vacation 631

thus, we have expected busy cycle

E(C) =
1
.
Substituting E(C) of (13) into the above expression, we can conclude the expected idle
period in a busy cycle
E(I) =
1
.
By (13), we have
E(B) =
)
. (15)
6. Numerical Examples
In the above analysis, we obtain the expected queue length and the expected virtual
time in the steady state. The model in this paper can be used to model many practical
problems. For example, the wireless mobile devices accessing the wireless networks with
different data rate can also be analyzed. Certainly, the difference of parameters, such
as the lower service rate and vacation rate in the system, also may influence the queue
length and virtual time in the model. So, we present numerical examples to explain such
influence and the validity of our model.
According to the expression for E(Q
d

1
and θ on the expected
number of the customers under two situations(see Figure 2 and Figure 3). Evidently,
along with the increase of the μ
v

), we show the effect of μ

v

, in other words, the server works faster and faster, the

number of the customers in steady state decreases. And, we also find that the vacation
rate θ has some effect on the queue length, namely, when μ
v

is fixed, if θ is smaller,
)
. (13)
Similarly, using alternative renewal theorem and the second equation in (8), we have
π
01

=
E(I)
(1 − r). (14)
Finally, with the third equation in (8), for regular busy period B, we have
E(B)
“M19N45” — 2008/11/5 — 17:15 — page 631 — #11
i
E(C)
E(C)
=
1−ρ+
=
θ
1−ρ+
+
λ
1
(1 − r) +
λ
θ
μ
b

(1 − r)(1 − ρ) +
λ
θ
(1 − ρ)
(1 − r)(1 − ρ) +
μ
b

θ
(
1+
λ
λ
θ
μ
b

(1 − r)(1 − ρ)
(
1+
(1 − ρ)
1−r
1
r2
1−r
r2
μ
b

)
(
1+
θ
(
1+
μ
b

θ
(
1+
1−r
r2
1−r
r2
1−r
r2
)
)
i
i
i
i
i
632 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008
1.4
1.2
0
0.2
0.4
0.6
0.8
1.0 0
8
=0.9
1.0
=0.2
6
0.8
0.6
4
0.4
vacation
0
0.2
0.4
0.6
0.8
10
v
Figure 2. The relation of E(Q
d
2 0.2 0
4
0
4
3
2 vacation service rate
te
v
1
2
rate vacation service rate
v
1
a
c
atio
n
ra
) with μ
v and θ (ρ = 0.9).
the queue length E(Q
d
) with μ
v
Figure 3. The relation of E(Q
d and θ (ρ = 0.2).
) is bigger. Meanwhile, when the vacation rate θ increases to the ceratin extent,
i.e., the mean vacation time decreases, the vacation rate θ has little effect
on the queue length. Furthermore, from the comparison of Figure 2 and Figure 3, the
traffic intension ρ also causes the change of the expected queue length E(Q
d

). If ρ is
larger, E(Q
d

) also is larger, and it is reasonable in practice. In Table 1 and Table 2, using the expression for
the expected virtual time that we
derive in the above section, and the same conditions of Figure 2 and Figure 3, we firstly
obtain some special numerical values of the expected virtual time. We can find that, along with the
increase of the μ
v

, the virtual time in steady state decreases. And, if θ is

larger, E(W) also decreases.
Table 1. Mean virtual time E(W) with ρ = 0.2.
(
μ
(
v
(
(
(
(
(( θ
1.8042 0.1
1.4094 1.324 1.2925 1.2776 1.2693 0.3
1.5778 1.3554 1.3013 1.2803 1.2699 1.2641
0.5
1.4357 1.3152 1.283 1.2699 1.2633 1.2595
0.7
1.3415 1.2844 1.268 1.261 1.2575 1.2554
0.9
“M19N45” — 2008/11/5 — 17:15 — page 632 — #12
i
1.2757 1.2602 1.2555 1.2534 1.2523 1.2517
0.5 1 1.5 2 2.5 3
3
i
i
i
i
i
From the numerical analysis above, we can find the influence of parameters on the performance
measures in the system. The results are suitable to practical situations.
With the expressions for the performance measures, such as the expected queue length
E(Q
d

) and the virtual time E(W), the other researchers can model many practical
problems and give the suggestions to improve the systems.
References
[1] Baba, Y., Analysis of a GI/M/1 queue with multiple working vacations, Oper.Res.Letters, Vol. 33,
pp. 201-209, 2005. [2] Doshi, B., Queueing systems with vacations-a survey, Queueing Syst., Vol. 1, pp. 29-66, 1986. [3]
Gross, D. and Harris, C., Fundamentals of Queueing Theory, second edition, John wiley & Sons,
New York, 1985. [4] Kim, J., Choi, D. and Chae, K., Analysis of queue-length distribution of the M/G/1 queue with
working vacations, International Conference on Statistics and Related Fields, Hawaii, 2003. [5] Latouche, G. and Ramaswami,
V., Introduction to Matrix Analysis Methods in Statistics Modeling,
ASA-SIAM Series on Applied Probability, 1999. [6] Liu, W., Xu, X. and Tian, N., Some results on the M/M/1 queue with
working vacations, Oper. Res.
Letters, Vol. 35, No. 5, pp. 595-600, 2007. [7] Neuts, M., Matrix-Geometric Solutions in Stochastic models, Johns Hopkins
University Press, Bal-
timore, 1981. [8] 1983. [9] Servi, L. D. and Finn, S. G., M/M/1 queue with working vacations(M/M/1/WV), Performance
Evaluation, Vol. 50, pp. 41-52, 2002. [10] Takagi, H., Queueing Analysis, Vol.1, Elsevier Science Publishers, Amsterdam,
1991. [11] Teghem, J., Control of the service process in a queueing system, Eur. J. Oper. Res., Vol. 23, pp.
141-168, 1986. [12] Tian N. and Zhang, G., Vacation Queueing Models-Theory and Applications, Springer-Verlag, New
York, 2006. [13] Tian N. and Zhang, G., A two threshold vacation policy in multiserver queueing systems, Eur. J.
Oper. Res., Vol. 168, No. 1, pp. 153-163, 2006.
(
μ
(
v
(
(
(
(
(( θ
11.463 0.1
10.55 10.285 10.173 10.115 10.082
0.3
 11.071 10.4 10.208 10.127 10.085 10.061 0.5
10.713 10.266 10.139 10.086 10.058 10.042
0.7
10.396 10.149 10.079 10.049 10.033 10.024
0.9
“M19N45” — 2008/11/5 — 17:15 — page 633 — #13
i
The M/M/1 Queue with Single Working Vacation 633

Table 2. Mean virtual time E(W) with ρ = 0.9.

10.121 10.046 10.025 10.015 10.011 10.008
0.5 1 1.5 2 2.5 3
i
i
i
i
i
634 International Journal of Information and Management Sciences, Vol. 19, No. 4, December, 2008
[14] Wu, D. and Takagi, H., M/G/1 queue with multiple working vacations, Performance Evaluation,
Performance Evaluation, Vol. 63, No. 7, pp. 654-681, 2006. [15] Zhang, Z. G. and Tian, N., Analysis on queueing systems
with synchronous vacations of partial
servers, Performance Evaluation, Vol. 52, No. 2, pp. 269-282, 2003.

Authors’ Information
Naishuo Tian is a professor in the Department of Mathematics at Yanshan University in China. His re- search interests are
stochastic process, queueing systems, and performance evaluation models in computer net- works and supply chain systems. He
has published many papers in several international journals including Queueing Systems, Stochastic Models, Performance
Evaluations, Operations Research Letters, EJOR, INFOR and several other journals in China.
Department of Mathematics, College of Sciences, Yanshan University, Qinhuangdao, P. R. China.
E-mail: tiannsh@ysu.edu.cn Tel: +86-335-8074703
Xinqiu Zhao is a associate professor in the School of Electric Engineering at Yanshan University in China. Her research interests
are optic-fiber sensor, automatic control, and performance evaluation models in compute networking .She has published many
papers in several journals including ACTA Metrologica Sinica, Laser & Infrared, Journal of Optoelectronics Laser and several
other journals in China.
School of Electric Engineering at Yanshan University, Qinhuangdao, P. R. China.
E-mail: zxq5460@ysu.edu.cn Tel: +86-0335-8387556
Kaiyu Wang is a teacher in the Department of Computer at Yanshan University in China. His research interests are web
information extaction and performance evaluation of web server clusters. He has published some papers in several journals in
China.
Department of Computer, College of Information science and engineering, Yanshan University, Qinhuang- dao, P. R. China.
E-mail: wangky@ysu.edu.cn Tel: +86-335-8057078
“M19N45” — 2008/11/5 — 17:15 — page 634 — #14
i
i
i
i