Applied Mathematical Modelling 31 (2007) 880894

www.elsevier.com/locate/apm

Vacation policies for machine repair problem

with two type spares
Jau-Chuan Ke
a

a,*

, Kuo-Hsiung Wang

Department of Statistics, National Taichung Institute of Technology, No. 129, Sec. 3, Sanmin Road, Taichung 404, Taiwan, ROC
b
Department of Applied Mathematics, National Chung-Hsing University, Taichung 402, Taiwan, ROC
Received 1 October 2003; received in revised form 1 November 2005; accepted 20 February 2006
Available online 12 June 2006

Abstract
This paper studies the machine repair problem consisting of M operating machines with two types of spare machines
(S = S1 + S2), and R servers (repairmen) who leave for a vacation of random length when there are no failed machines
queuing up for repair in the repair facility. At the end of the vacation the servers return and operate two vacation policies.
First, the servers take vacations repeatedly until they nd the repair facility has at least one waiting failed machine in the
queue. Second, the servers do not take a vacation again and remain idle until the rst arriving failed machine arrives, which
starts a busy period in the repair facility. For both policies, the servers have two service rates for repair-slow and fast. The
matrix geometric theory is used to nd the steady-state probabilities of the number of failed machines in the system as well
as the performance measures. Some special cases are given. A direct search algorithm is used to simultaneously determine
the optimal values of the number of two types of spares and the number of servers while maintaining a minimum specied
level of system availability.
 2006 Elsevier Inc. All rights reserved.
Keywords: Machine repair problem; Multiple vacations; Single vacation; Spare

1. Introduction
We consider a machine repair problem where a group of M operating machines is under the supervision of
one or more repairmen (servers) in the repair facility. If at any time a machine fails, it is sent to the repair
facility for repair. This incidence may lead to loss of production because the failed machine must stay in
the repair facility for some time. To avoid any loss of production, the plant always keeps some spare machines,
say S (S < M), so that a spare machine can immediately act as a substitute when an operating machine fails.

Corresponding author. Tel.: +886 422196638; fax: +886 422196331.

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

0307-904X/$ - see front matter  2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2006.02.009

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

881

Meanwhile, the repairmen (servers) leave for a vacation of random length whenever there are no failed
machines queuing up for repair in the repair facility. If the repairmen (servers) nd at the repair facility
any failed machines in the queue when they return from a vacation, they immediately repair the failed
machines until there are no failed machines on the waiting list. A real-world example of this vacation model
can be realized in the manufacturing/production-assembly system where the servers in their idle time may be
assigned to perform some extra operations such as additional work, preventive maintenance. On the other
hand, machine repair related problems represent a group of very important problems used to analyze timesharing computer systems, multi-programmed computer systems and multi-access communication channels
(see Medhi [1]).
In a classical machine repair model, it is assumed that the servers remain idle until the failed machines present. Numerous machine repair problems have been proposed, for example, no-spare M/M/R model by Feller
[2], cold-standby M/M/R model by Toft and Boothroyd [3], and warm-standby M/M/R model by Sivazlian
and Wang [4]. Wang [5] further investigated the M/M/R machine repair problem with mixed-standby spare.
Recently, Wang and Lee [6] generalized Sivazlian and Wang [4] model to the M/M/R machine repair problem
with cold-standby and multiple modes of failure. Cost (prot) models in the machine repair problem have
been investigated by several authors including Ashcroft [7], Elsayed [8], Hilliard [9], Gross et al. [10,11], Sivazlian and Wang [4], Wang [5], and Wang and Lee [6]. As for a vacation model, comprehensive and excellent
surveys, including some applications, were examined by Doshi [12] and Takagi [13]. Gupta [14] rst analysed
the M/M/1 machine repair problem with warm spares and server vacations. He gave an algorithm to compute
the steady-state probability distribution of the number of failed machines in the system, but no cost analysis
was provided. Comparable work on multi-server queueing models or multi-server repair problem models with
multi-server vacations is rarely found in the literature. This motivates us to develop the M/M/R machine
repair problem with two types of spares where the servers are characterized by two vacation policies: multiple
vacations vs. single vacation.
The purpose of this paper is threefold. The rst is to present a matrix geometric method for developing
steady-state solutions for M/M/R machine repair problems under two vacation policies with two types of
spares. These solutions are used to obtain the various system performance measures, such as the expected
number of failed machines, the expected number of idle and busy servers, machine availability, operative utilization, etc. The second is to develop a cost model for these two policies to determine the joint optimum number of two types of spares and servers in order to minimize the steady-state expected cost per unit time, while
maintaining a specied level of system availability. The third is to perform a sensitivity analysis to investigate
the eect on the joint optimum number of spares and servers if the system parameters take on other specic
values. A comparison of Policy 1 with Policy 2 on the various system performance measures under the optimal
operating conditions will also be provided.
2. Problem statement
We consider a machine repair model with L identical and independent machines that are subject to breakdowns, and R servers who are responsible for maintaining these machines. As many as M machines can be
operated simultaneously in parallel; the rest of the S = L  M machines are available as spares. Two types
of spare machines are considered, S1 units of type 1 and S2 units of type 2, where S1 + S2 = S. If all spares
are being used and a breakdown occurs, we say that the system becomes short. In this case there are less than
M operating machines.
2.1. Assumptions
When each server nds at the repair facility no failed machines queuing up for repair, he leaves the repair
facility for a vacation with random length. Upon returning from the rst vacation in an idle period, each server inspects the repair facility and decides whether to take another vacation, to remain idle at the repair facility, or to resume repairing the waiting failed machines. One of the following two vacation policies are
considered;

882

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

Policy 1 (multiple vacations): If there is not an empty queue in the repair facility, the server starts to repair the
waiting failed machines. Otherwise, if there is an empty queue in the repair facility, the server leaves
immediately for another vacation. This pattern continues until he returns from a vacation to nd at
least one failed machine waiting in the queue.
Policy 2 (single vacation): If there is an empty queue in the repair facility, he waits idly for the rst failed
machine to arrive, upon which he starts repairing the failed machines. On the other hand, if there
is not an empty queue, he immediately starts to repair the waiting failed machines.
Furthermore, our analysis is based on the following assumptions.
1. Operating machines are subject to breakdowns according to an independent Poisson distribution with rate
k. Spare machines of type 1 and type 2 are subject to failure according to independent Poisson processes
with rates a1 and a2, respectively. Type 2 spares have a higher failure rate than type 1 units, where
0 6 a1 6 a2 6 k.
2. When an operating machine breaks down, it will immediately be backed up by an available spare. When a
type 2 spare machine moves into an operating state, its failure characteristics will be that of an operating
machine, and when a type 1 spare machine moves into a type 2 spare state, its failure characteristics will be
that of a type 2 spare.
3. Whenever a machine fails, it is immediately sent to the repair facility where repair work is provided in the
order of breakdowns. The repair (service) times of failed machines provided by R servers are assumed to be
independent and identically exponential distributions with parameters l and l1. Each server repairs at the
slow rate l until there n (n P R) failed machines in the system, at which epoch he switches to his faster rate
l1, in order to reduce the level of failed machines in the queue.
4. Each server can repair only one failed machine at a time. The failed machine which on arriving at the repair
facility nds it busy or on a vacation must wait in the queue until a server is available.
5. When a failed operating machine or a failed spare machine is repaired, it is as good as a new one and goes
into standby state unless the system is short and the repaired machine is immediately sent back to an operating state.
6. All vacations of servers are independent and identically distributed as an exponential with rate v, and are
independent of the other processes.

2.2. Practical justication of our model

A number of practical problems arise which may be formulated as one in which the performance measures of a system are often improved by the provision of spare machines or parallel redundant machines
with facilities to repair the failed one. One particular problem where this model is applied is in the study
of a environmental engineering system. Consider a sanitary landll eld has a pool of M gas power generators and two types of spare generators, where type 1 has S1 spare generators and type 2 has S2 spare
generators. The generators are subject to breakdowns, and are maintained by a repair facility which has R
repairmen. The failure times of the operating generator are independent of each other and have a common
exponential distribution with mean 1/k. The failure time of a type 1 (type 2) generator has an exponential
distribution with mean 1/a1 (1/a2) with a1 6 a2 6 k. An operating generator that breaks down is replaced
immediately by a type 1 spare generator if one is available, and a type 2 spare generator is switched to
type 1 generator if any is available. The repair of a failed generator is switched to type 2 generator mode
unless the system is short in which case the repaired generator is immediately sent back to an operating
mode. A failed generator immediately enters a repair facility which is able to repair at most R generators.
The R workers of the sanitary landll eld are in charge of monitoring, maintaining and repairing these
generators at a slow rate l or at a fast rate l1. The repair time of a failed generator has an exponential
distribution. The R workers, who are not operating, are assigned temporary cleaning jobs at the sanitary
landll. The cleaning jobs would correspond to the vacation of workers. For eciency, the management
can analyze and model the power system of the sanitary landll eld using our models. These results

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

883

developed in this paper are used to the evaluating and predicting the performance measures of the
system.
3. Steady-state results
3.1. Policy 1: server with multiple vacations
For the M/M/R machine repair model with two types of spares and server multiple vacations we describe
the state of the system by the pairs {(i, n): i = 0, 1, . . . , R, and i 6 n 6 M + S1 + S2 = L}, where i and n are in
the subscript denoting the number of busy servers and the number of failed machines in the system,
respectively.
Let n represent the number of failed machines in the system. The mean failure rate kn and mean repair rate
ln for this system are given by
8
Mk S 2 a2 S 1  na1 ; n 0; 1; 2; . . . ; S 1 ;
>
>
>
< Mk S S  na ;
n S 1 1; S 1 2; . . . ; S 1 S 2 S;
2
1
2
kn
>
L  nk;
n S 1; S 2; . . . ; M S L;
>
>
:
0;
otherwise;
and
8
n i; i 1; i 2; . . . ; R  1;
>
< il;
ln

>
:

Rl1 ;
0;

and n R; R 1; R 2; . . . ; M S L;
otherwise;

where i is the number of the available servers when the failed machines occur.
In steady-state, the following notations are used:
P0,n  probability that there are n failed machines in the system when all the servers take vacations;
Pi,n  probability that there are n failed machines in the system when there are i servers working;
PR,n  probability that there are n failed machines in the system when all the servers are working,
where i = 1, 2, . . . , R, and i 6 n 6 L.
Referring to the state-transition-rate diagram for the M/M/R machine repair problem with two types of
spares and server multiple vacations shown in Fig. 1, we have the following steady-state equations:
3.1.1. Steady-state equations
(i) i = 0
k0 P 0;0 lP 1;1 ;
Rv kn P 0;n kn1 P 0;n1 ;

1
1 6 n 6 L  1;

RvP 0;L kL1 P 0;L1 :

(ii) 1 6 i 6 R  1
ki ilP i;i R  i 1vP i1;i ilP i;i1 i 1lP i1;i1 ;

i n 6 R  1;

ki il R  ivP i;n R  i 1vP i1;n kn1 P i;n1 ilP i;n1 ; i < n < R  1;
kR1 il R  ivP i;R1 R  i 1vP i1;R1 kR2 P i;R2 il1 P i;R ;
i < n R  1;

4
5
6

kn il1 R  ivP i;n R  i 1vP i1;n kn1 P i;n1 il1 P i;n1 ;

i < n; R 6 n 6 L  1;

il1 R  ivP i;L R  i 1vP i1;L kL1 P i;L1 :

884

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

0
R=3

vacation

0,0

0,1

2
0,2

vacation

0,3

1,1

2 2

R 2 =1

vacation

1,4

31

2,4

2 1

2 1

3
R 3=0

2,5

3,5

3,4

3,3

vacation

2,3

2 1

1,5

2,2

1,3

0,5

1,2

4
0,4

1
R 1 = 2

31

31

Fig. 1. State-transition rate diagram for the M/M/R machine-repair problem with two types of spares and multiple vacations (for R = 3
and L = 5).

(iii) i = R
kR Rl1 P R;R vP R1;R Rl1 P R;R1 ;

Rl1 P R;L vP R1;L kL1 P R;L1 ;

kn Rl1 P R;n vP R1;n kn1 P R;n1 Rl1 P R;n1 ;

10
R 1 6 n 6 L  1:

11

3.1.2. Matrix-geometric solutions

A discrete transforms method was developed by Thiruvengadam and Jaiswal [15] while solving the set of
dierential-dierence equations for the machine repair problem. Alternatively, we provide a matrix-geometric
method to analyze the resulting system of linear equations (1)(11) and further develop the closed-form
expressions for the steady-state probabilities Pi,n, 0 6 i 6 R and i 6 n 6 L. One nds that the transition rate
matrix Q of this Markov chain has the block-tridiagonal form:
1
0
A0 B 0
C
B
C
B C 1 A1 B1
C
B
C
B
C 2 A2 B 2
C
B
C
B
C
B
C
A
B
3
3
3
C
B
C:
B
QB
..
.. ..
C
C
B
.
.
.
C
B
C
B
C
B
C R2 AR2 BR2
C
B
B
C R1 AR1 BR1 C
A
@
CR

AR

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

885

The rate matrix Q of this state process is similar to the quasi birth and death type, and this class of Markov
process has been extensively investigated by Neuts [16]. The matrix Q is a square matrix of order
(L + 1)(R + 1)  R(R + 1)/2 and each entry of the matrix Q is listed in the following:
1
0
wi 0    0 0
B 0 0  0 0C
C
B
C
B
B 0 0  0 0C
C
B
; 1 6 i 6 R;
12
Ci B . .
.. .. C
..
B .. ..
. .C
.
C
B
C
B
@ 0 0  0 0A
0  0 0
1
0
0

0
0
C
B R  iv
0

0
0
C
B
C
B
C
B
0
R  iv   
0
0
C
B
Bi B
C; 0 6 i 6 R  1;
..
..
..
..
..
C
B
.
.
.
.
.
C
B
C
B
A
@
0
0
   R  iv
0
0
0

0
R  iv
1
0
y i;i
0
0
0

0
0
0
0
zi;i
Bx
0
0

0
0
0
0 C
C
B i1;i zi1;i y i1;i
C
B
C
B 0
x
z
y
0



0
0
0
0
i2;i
i2;i
i2;i
C
B
C
B
0
xi3;i zi3;i y i3;i   
0
0
0
0 C
B 0
C
B
B .
.. C
..
..
..
..
..
..
..
..
..
Ai B
. C
.
.
.
.
.
.
.
.
C; 0 6 i 6 R;
B
C
B
.
C
B
.. z
0
0
0
0
y L3;i
0
0 C
B 0
L3;i
C
B
B 0
0
0
0
0
   xL2;i zL2;i y L2;i
0 C
C
B
C
B
0
0
0
0

0
xL1;i zL1;i y L1;i A
@ 0
0



xL;i

13

14

zL;i

where Bi and Ci are (L + 1  i) (L  i) and (L + 1  i) (L + 2  i) matrices, respectively, Ai is the square

matrix of order (L + 1  i), and

il; 1 6 i 6 R  1;
wi
il1 ; i R;

il; i 1 6 j 6 R  1;
xj;i
il1 ; R 6 j 6 L;
y j;i kj ; i 6 j 6 L  1;
8
kj R  iv il; if i 1 6 j 6 R  1;
>
>
>
<
kj R  iv il1 ; if R 6 j 6 L;
zj;i
>
kj il;
if 0 6 i j 6 R  1;
>
>
:
kj il1 ;
if i j R:
Let P denote steady-state probability vector of Q. By partitioning the vector P as P = {P0, P1, P2, . . . , PR}
where Pi = {Pi,i, Pi,i+1, Pi,i+2, . . . , Pi,L}, for 0 6 i 6 R is a 1 (L + 1  i) vector, one nds that the steady-state
equations PQ = 0 are given by
P 0 A0 P 1 C 1 0;
P i1 Bi1 P i Ai P i1 C i1 0;

16i6R1

886

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

and
P R1 BR1 P R AR 0:
After substitutions, we obtain
P R P R1 BR1 A1
R ;

15

P i P i1 X i ; 1 6 i 6 R  1;
P 0 A0 X 1 C 1 0;

16
17

where
1

X i Bi1 Ai X i1 C i1

is a (L + 2  i) (L + 1  i) matrix with Ci, Bi and Ai given in (12)(14). Furthermore X R BR1 A1

R .
Eq. (17) determines P0 up to a multiplicative constant. The other Eqs. (15) and (16) determine
PR, PR1, PR2, . . . , P1, up to the same constant, which is uniquely determined using the following normalizing
equation
R
X

P i e 1;

i0

where e represents a column vector with each component equal to one. A computer program was developed to
solve Pi and Pi,n for 0 6 i 6 R, and i 6 n 6 L.
3.1.3. System performances
Our analysis is based on the following system performance measures of the M/M/R machine repair problem with two types of spares and server multiple vacations. Let
E[F]  the expected number of failed machines in the system,
E[Fq]  the expected number of failed machines in the queue,
E[O]  the expected number of operating machines in the system,
E[S]  the expected number of spare machines in the system acting as standbys,
E[S1]  the expected number of spare machines of type 1 in the system acting as standbys,
E[S2]  the expected number of spare machines of type 2 in the system acting as standbys,
E[B]  the expected number of busy servers in the system,
E[V]  the expected number of vacation servers in the system,
E[I]  the expected number of idle servers in the system,
M.A.  machine availability (the fraction of the total time that the machines are working),
O.U.  operative utilization (the fraction of busy servers).
The expressions for E[F], E[Fq], E[O], E[S1], E[S2], E[B], E[V] and E[I] are obtained as follows:
EF 

R X
L
X

EF q 

18

nP i;n ;

ni

i0

R X
L
X
n  iP i;n ;

19

ni

i0

EO M 

R
L
X
X

n  SP i;n ;

20

i0 nS1

ES

R X
S
X
S  nP i;n ;
i0

ni

21

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

S2
R X
X

ES 2 

S 2  nP i;n ;

22

ni

i0

ES 1  ES  ES 2 ;
R X
L
X
iP i;n ;
EB
i1

EV 

887

23
24

ni

R X
L
X
i0

R  iP i;n ;

25

ni

EI R  EB  EV :

26

Following Benson and Cox [17], the machine availability and the operative utilization are dened by
EF 
;
L
EB
O:U :
:
R

27

M:A: 1 

28

Let E[W] and E[Wq] denote the expected waiting time in the system and the expected waiting in the queue,
respectively. Using the Littles formula we obtain
EW  EF =ke ;
EW q  EF q =ke ;
where the eective arrival rate into the system
ke

R X
L
X
i0

kn P i;n :

ni

3.2. Policy 2: server with single vacation

For the M/M/R machine repair model with two types of spares and server single vacation we describe the
state of the system by the pairs {(i, n): i = 0, 1, . . . , R, n = 0, 1, 2, . . . , L}, where this i denotes the number of servers available for repair (busy or idle), and n, as before, denotes number of failed machines in the system.
In the analysis we use the same notations as those for the rst model. Referring to the state-transition-rate
diagram for the M/M/R machine repair problem with two types of spares and server single vacation shown in
Fig. 2, we obtain the following steady-state equations:
3.2.1. Steady-state equations
(i) i = 0
k0 RvP 0;0 lP 1;1 ;
kn RvP 0;n kn1 P 0;n1 ;

29
30

1 6 n 6 L  1;

RvP 0;L kL1 P 0;L1 :

31

(ii) 1 6 i 6 R  1
k0 R  ivP i;0 R  i 1vP i1;0 lP i1;1 ;

n 0;

32

kn R  iv nlP i;n R  i 1vP i1;n i 1lP i1;n1 kn1 P i;n1 ;

1 6 n < i 6 R  2;
ki R  iv ilP i;i R  i 1vP i1;i i 1lP i1;i1 ki1 P i;i1
ilP i;i1 ;

n i 6 R  2;

33
34

888

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

0
R=3

vacation

0,0

0,2

0,1

vacation

2 2

vacation

2,1

2,0

0
R 3=0

3,0

31

4
2,4

2,5

2 1

3,2

3,1

2 1

1,5

2,3

2,2

1,4

2 1

vacation

0
R 2 =1

1,3

0,5

1,2

4
0,4

1,1

1,0

3
0,3

0
R 1 = 2

3,5

3,4

3,3

31

31

Fig. 2. State-transition rate diagram for the M/M/R machine-repair problem with two types of spares and single vacation (for R = 3 and
L = 5).

kR1 v R  1lP R1;R1 2vP R2;R1 Rl1 P R;R kR2 P R1;R2

R  1l1 P R1;R ;

n i R  1;

35

kn R  iv ilP i;n R  i 1vP i1;n kn1 P i;n1 ilP i;n1 ;

i < n 6 R  2;

36

kR1 R  iv ilP i;R1 R  i 1vP i1;R1 kR2 P i;R2 il1 P i;R ;

i < n R  1;

37

kn R  iv il1 P i;n R  i 1vP i1;n kn1 P i;n1 il1 P i;n1 ;

i < R 6 n 6 L  1;

38

R  iv il1 P i;L R  i 1vP i1;L kL1 P i;L1 :

39

(iii) i = R
k0 P R;0 vP R1;0 ;
kn nlP R;n vP R1;n kn1 P R;n1 ;

40
1 6 n 6 R  1;

kn Rl1 P R;n vP R1;n kn1 P R;n1 Rl1 P R;n1 ;

Rl1 P R;L vP R1;L kL1 P R;L1 :

R 6 n 6 L  1;

41
42
43

We still analyze the model of Policy 2 employing matrix-geometric solution described in previous section.
Let P be the steady-state probabilities and P = {P0, P1, P2, . . . , PR}, where Pi is a 1 (L + 1) row vector and
Pi = {Pi,0, Pi,1, Pi,2, . . . , Pi,L}, for 0 6 i 6 R. Let Q denote the corresponding transition rate matrix, then Q can
be expressed as

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

A0

B0

BC
B 1
B
B
B
B
B
QB
B
B
B
B
B
B
@

A1
C2

889

1
B1
A2

B2

C3

A3

B3
..
.

..

..

C R2

AR2

BR2

C R1

AR1
CR

C
C
C
C
C
C
C
C:
C
C
C
C
C
C
BR1 A
AR

It should be noted the expressions for Pi, Ai, Bi, and Ci are dierent from those in Policy 1. The matrix Q is a
square matrix of order (L + 1) (R + 1). Each entry of the matrix Q is a square matrix of order L + 1 given as
follows:
Bi R  ivI;
0
0
0
B w1;i
0
B
B
B 0 w2;i
B
Ci B .
..
B ..
.
B
B
@ 0
0
0
z0;i
B x1;i
B
B
B 0
B
Ai B .
B ..
B
B
@ 0
0





..
.


0
0
..
.
0

0C
C
C
0C
C
;
.. C
.C
C
C
0A

0    wL;i
y 0;i 0   
z1;i y 1;i   

0
0
0

0
0

x2;i
..
.

z2;i
..
.


..
.

0
..
.

0
..
.

0
0

0
0




xL1;i
0

zL1;i
xL;i

where Ai, Bi, Ci and

8
jl;
>
<
wj;i jl1 ;
>
:
0;

1
0
0 C
C
C
0 C
C
;
.. C
. C
C
C
y L1;i A
zL;i

I are the square matrix of order (L + 1), and

if j 6 i; 1 6 i 6 R and 1 6 j 6 R  1;
if j 6 i; 1 6 i 6 R and R 6 j 6 L;
otherwise;

y j;i kj ; if 0 6 i 6 R and 0 6 j 6 L  1;

8
il; if j > i; 0 6 i 6 R and 1 6 j 6 R  1;
>
<
xj;i il1 ; if j > i; 0 6 i 6 R and R 6 j 6 L;
>
:
0;
if j 6 i; 0 6 i 6 R and 1 6 j 6 L;
8
k R  iv il; if j P i; 0 6 i 6 R and 0 6 j 6 R  1;
>
< j
zj;i kj R  iv il1 ; if j P i; 0 6 i 6 R and R 6 j 6 L;
>
:
kj R  iv jl; if 0 6 i 6 R and 0 6 j < i:
One nds that the steady-state Eqs. (29)(43) can be rewritten to PQ = 0 given by
P 0 A0 P 1 C 1 0;
R  i 1vP i1  P i Ai  P i1 C i1 0;

16i6R1

890

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

and
vP R1  P R AR 0:
Using the same procedure as described in Policy 1, the new expression for Pi derived above are used. A computer program was developed to solve Pi and Pi,n for 0 6 i 6 R, and 0 6 n 6 L.
3.2.2. System performances
The system performance measures of the M/M/R machine repair problem with two types of spares and
server single vacation is given by
R X
L
X

EF 

i0

EF q 

44

nP i;n ;

n0

L
X

nP 0;n

n0

R X
i1
X
i1

EO M 

R
L
X
X

nP i;n

n0

R X
L
X
i1

n  iP i;n ;

45

ni

n  SP i;n ;

46

i0 nS1

ES 2 

S2
R X
X
S 2  nP i;n ;
i0

R X
S
X
S  nP i;n ;

ES

i0

EV 
EI

50
51

n0

R X
i1
X
i1

49

ni

R X
L
X
R  iP i;n ;
i0

48

n0

ES 1  ES  ES 2 ;
R X
L
X
iP i;n ;
EB
i1

47

n0

iP i;n :

52

n0

4. Special cases
We present some existing results in the literature which are special cases of our model.
Case 1: The M/M/R machine repair problem with warm spares considered by Sivazlian and Wang [4] is a special case of our model with v = 0, a1 = a2 and l = l1.
Case 2: If R = 1, a1 = a2 and l = l1, our model becomes the M/M/1 machine repair problem with spare and
server multiple vacations (or single vacation) considered by Gupta [14].
Case 3: The corresponding result for the ordinary M/M/R/L queueing system (see Gross and Harris [18]) can
be found by setting v = 0, a1 = a2 = 0, l = l1 and S = L  1.
Case 4: The new result for the ordinary M/M/R/L queueing system with server multiple vacation (or single
vacation) can only be obtained by setting a1 = a2 = 0, l = l1 and S = L  1.

5. Cost analysis for both policies

We construct a total expected cost function per unit time for the M/M/R machine repair problem under
two vacation policies with two types of spares, and impose a constraint on the system availability in which

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

891

S1, S2 and R are decision variables. Our objective is to determine the optimum number of two type spares
(S1, S2), say S 1 ; S 2 , and the optimum number of servers R, say R*, simultaneously so that the cost is minimized and the system availability is maintained at a certain level.
First let
Ch  cost per unit time when one failed machine joins the system,
Ce  cost per unit time of a failed machine after all spares are exhausted (downtime cost),
C s1  cost per unit time when one machine is functioning as a type 1 spare (inventory cost),
C s2  cost per unit time when one machine is functioning as a type 2 spare (inventory cost),
Cb  cost per unit time when one server is busy,
Ci  cost per unit time when one server is idle,
Cf  cost per unit time of each server,
r  reward per unit time when one server is on vacation.
Using the denitions of these cost element listed above, the total expected cost function per unit time is
given by:
T cost S 1 ; S 2 ; R C h EF  C e M  EO C s1 ES 1  C s2 ES 2  C b EB C i EI RC f  rEV :

53

The cost parameters in (53) are assumed to be linear in the expected number of the indicated quantity.
The production system always requires minimum of M machines in operation. Following the concept of
Wang and Lee [6], the cost minimization problem can be illustrated mathematically as
Minimize

T cost S 1 ; S 2 ; R:

Subject to

A:V :policy1

S 1 ;S 2 ;R

R X
S
X
i0

A:V :policy2

or

54

R X
S
X
i0

P i;n P A;

55

P i;n P A;

56

ni

n0

where A.V.(policy 1) and A.V.(policy 2) are the steady-state probability that at least M machines are in operation
and function properly (system availability), and A is the minimum fraction of time all M machines are in operation and function properly.
The decision variables S1, S2 and R are required to be integer values and to be determined by a suitable
non-linear optimization algorithm. A direct search method may be used to obtain potentially useful results.
The optimization algorithm is a direct search approach over a grid whose boundaries for decision variables
are selected in order to guarantee that the global optimum is obtained in the interior region (see Hilliard
[9]). The direct search algorithm can be terminated after obtaining a global optimum. In applying the direct
search algorithm, it is necessary to have upper bounds for three decision variables. The upper bound used on
the decision variables S1, S2 and R is 2M.
We describe the direct search algorithm for obtaining the optimal values S 1 ; S 2 ; R . The specic steps in
the direct search algorithm are as follows:
Step 1. Find the optimal number of type 1 spares, S 1 , and the optimal number of type 2 spares, S 2 , for R
servers, i.e.,
min T cost S 1 ; S 2 ; R T cost S 1 ; S 1 ; R;
S 1 ;S 2

subject to the constraint is satised.

Step 2. Find the set of all minimum cost solutions for R = 1, 2, . . . , i.e.,
H fT cost S 1 ; S 2 ; R : R 1; 2; . . .g:

892

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

Step 3. Find the set of all minimum cost solutions for R = 1, 2, . . . , i.e.,
min H T cost S 1 ; S 2 ; R :
R

We provide an example to illustrate the direct search algorithm.

Example. Consider M = 10, the cost elements and availability as follows (for Policy 1)
C h 10; C e 125; C s1 60; C s2 35; C b 75; C i 40; C f 80; r 60; and A:V : 0:8:
Step 1. Find S 1 and S 1 for R servers necessary to satisfy the availability requirement, where
R = 1, 2, 3, 4, 5, 6, . . . (see Table 1).
Step 2. From Table 1, H = {$714.251, $724.662, $777.399, $834.817, $899.189, $946.947, . . .}.
Step 3. From step 2, the optimal solution T cost S 1 ; S 2 ; R $714:251 is achieved at S 1 0, S 2 8 and R* = 4.
5.1. Sensitivity analysis
We perform a sensitivity analysis for changes in the joint optimum values S 1 , S 2 and R* along with changes
in specic values of the system parameters k, a1, a2, v, l and l1.
For Policy 1 the minimum expected cost Tcost(S1,S2,R) and the values of various system performance measures A.V., E[F], E[Fq], E[S1], E[S2], E[B], E[O], E[V], E[I], M.A., and O.U. at the optimum values S 1 , S 2 and R*
are shown in Table 2 for dierent values of (k, a1, a2).
From Table 2, we observe that (i) T cost S 1 ; S 2 ; R increases as k or a2 increases; (ii) the optimum values of S2
and R, S 2 and R* increase as k increases. From the last ve columns of Table 2, S 1 ; S 2 ; R are not aected at
all when a1 varies from 0.0 to 0.05 or a2 varies from 0.05 to 0.6. Intuitively, this seems too insensitive to
changes in a1 and a2.

Table 1
The expected cost T cost S 1 ; S 2 ; R and the system availability A.V. (k = 0.6, a1 = 0.01, a2 = 0.05, v = 10, l = 2.5, l1 = 4.0)
R

S 1 ; S 2 ; R
T cost S 1 ; S 2 ; R
A.V.

(0, 8, 4)
714.251
0.8473

(0, 6, 5)
724.662
0.8143

(0, 6, 6)
777.399
0.8348

(0, 6, 7)
834.817
0.8210

(0, 7, 8)
899.189
0.8970

(0, 7, 9)
946.947
0.866

Table 2
System performance measures of the machine-repair problem with two types of spares and multiple vacations under optimal operating
conditions (v = 1.0, l = 2.5, l1 = 4.0)
(k, a1, a2)

(1.2, 0, 0.05)

(0.6, 0, 0.05)

(0.6, 0.025, 0.05)

(0.6, 0.05, 0.05)

(0.6, 0.05, 0.3)

(0.6, 0.05, 0.6)

S 1 ; S 2 ; R
T cost S 1 ; S 2 ; R

(0, 9, 5)
752.6414
0.8393
6.3327
3.1089
0.0000
3.0475
3.2235
9.6202
1.7765
0.0000
0.6667
0.6647

(0, 6, 3)
446.3984
0.8388
3.9513
2.3411
0.0000
2.4100
1.6102
9.6387
1.3898
0.0000
0.7530
0.5367

(0, 6, 3)
446.3984
0.8388
3.9513
2.3411
0.0000
2.4100
1.6102
9.6387
1.3898
0.0000
0.7530
0.5367

(0, 6, 3)
446.3984
0.8388
3.9513
2.3411
0.0000
2.4100
1.6102
9.6387
1.3898
0.0000
0.7530
0.5367

(0, 6, 3)
460.0558
0.8214
4.2256
2.5024
0.0000
2.1687
1.7231
9.6058
1.2769
0.0000
0.7359
0.5744

(0, 6, 3)
473.1885
0.8045
4.4796
2.6478
0.0000
1.9447
1.8318
9.5757
1.1682
0.0000
0.7200
0.6106

A.V.
E[F]
E[Fq]
E[S1]
E[S2]
E[B]
E[O]
E[V]
E[I]
M.A.
O.U.

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

893

Table 3
System performance measures of the machine-repair problem with two types of spares and multiple vacations under optimal operating
conditions (k = 1.0,a1 = 0.01,a2 = 0.05)
(v, l, l1)

(0.1, 3.0, 5.0)

(10, 3.0, 5.0)

(10, 1.0, 5.0)

(1.0, 3.0, 3.0)

(1.0, 3.0, 5.0)

(1.0, 3.0, 8.0)

S 1 ; S 2 ; R
T cost S 1 ; S 2 ; R

(0, 12, 8)
799.1006
0.8018
8.6267
6.2393
0.0000
4.0761
2.3875
9.2971
5.6125
0.0000
0.6079
0.2984

(0, 5, 3)
493.4302
0.8425
3.3946
1.2208
0.0000
1.9168
2.1739
9.6886
0.8261
0.0000
0.7737
0.7246

(0, 5, 3)
522.2357
0.8270
3.6936
1.2867
0.0000
1.6478
2.4070
9.6585
0.5930
0.0000
0.7538
0.8023

(0, 7, 7)
759.4405
0.8029
5.2140
1.9880
0.0000
2.2191
3.2260
9.5670
3.7740
0.0000
0.6933
0.4609

(0, 7, 4)
563.2083
0.8278
4.8827
2.6937
0.0000
2.5220
2.1890
9.5953
1.8110
0.0000
0.7128
0.5473

(0, 6, 3)
435.1257
0.8225
4.0934
2.6397
0.0000
2.3517
1.4537
9.5549
1.5463
0.0000
0.7442
0.4846

A.V.
E[F]
E[Fq]
E[S1]
E[S2]
E[B]
E[O]
E[V]
E[I]
M.A.
O.U.

Table 4
System performance measures of the machine-repair problem with two types of spares and single vacation under optimal operating
conditions (v = 1.0, l = 2.5, l1 = 4.0)
(k, a1, a2)

(1.2, 0, 0.05)

(0.6, 0, 0.05)

(0.6, 0.025, 0.05)

(0.6, 0.05, 0.05)

(0.6, 0.05, 0.3)

(0.6, 0.05, 0.6)

S 1 ; S 2 ; R
T cost S 1 ; S 2 ; R
A.V.
E[F]
E[Fq]
E[S1]
E[S2]
E[B]
E[O]
E[V]
E[I]
M.A.
O.U.

(0, 8,5)
740.7840
0.8065
5.9905
2.8252
0.0000
2.4642
3.1653
9.5452
1.7300
0.1046
0.6672
0.6331

(0, 5, 3)
440.4962
0.8030
3.5715
2.0050
0.0000
1.8648
1.5665
9.5637
1.2985
0.1350
0.7619
0.5222

(0, 5, 3)
440.4962
0.8030
3.5715
2.0050
0.0000
1.8648
1.5665
9.5637
1.2985
0.1350
0.7619
0.5222

(0, 5, 3)
440.4962
0.8030
3.5715
2.0050
0.0000
1.8648
1.5665
9.5637
1.2985
0.1350
0.7619
0.5222

(0, 6, 3)
465.4845
0.8403
4.3076
2.1369
0.0000
2.3084
1.7208
9.6539
1.1918
0.0875
0.7477
0.5736

(0, 6, 3)
477.1179
0.8198
4.3398
2.5016
0.0000
2.0451
1.8382
9.1651
1.1031
0.0587
0.7288
0.6127

Table 5
System performance measures of the machine-repair problem with two types of spares and single vacation under optimal operating
conditions (k = 1.0, a1 = 0.01, a2 = 0.05)
(v, l, l1)

(0.1, 3.0, 5.0)

(10, 3.0, 5.0)

(10, 1.0, 5.0)

(1.0, 3.0, 3.0)

(1.0, 3.0, 5.0)

(1.0, 3.0, 8.0)

S 1 ; S 2 ; R
T cost S 1 ; S 2 ; R
A.V.
E[F]
E[Fq]
E[S1]
E[S2]
E[B]
E[O]
E[V]
E[I]
M.A.
O.U.

(0, 12, 8)
799.2766
0.8074
8.5434
6.1481
0.0000
4.1342
2.3954
9.3224
5.5940
0.0107
0.6117
0.2994

(0, 5, 3)
524.7508
0.8616
3.1988
1.3266
0.0000
2.0735
1.8723
9.7276
0.3939
0.7338
0.7867
0.6241

(0, 5, 3)
539.9204
0.8417
3.5720
1.4337
0.0000
1.7395
2.1382
9.6886
0.3045
0.5573
0.7619
0.7128

(0, 9, 5)
767.4091
0.8201
6.2882
3.1195
0.0000
3.1438
3.1687
9.5680
1.7041
0.1272
0.6690
0.6337

(0, 7, 4)
568.8661
0.8467
4.6770
2.5021
0.0000
2.6759
2.1749
9.6471
1.7180
0.1072
0.7249
0.5437

(0, 6, 3)
438.7800
0.8420
3.9023
2.4417
0.0000
2.4852
1.4606
9.6125
1.4678
0.0717
0.7561
0.4869

894

J.-C. Ke, K.-H. Wang / Applied Mathematical Modelling 31 (2007) 880894

For Policy 1 the minimum expected cost Tcost(S1, S2, R) and the values of various system performance measures A.V., E[F], E[Fq], E[S1], E[S2], E[B], E[O], E[V], E[I], M.A., and O.U. at the optimum values S 1 , S 2 and R*
are shown in Table 3 for dierent values of (v, l, l1).
One sees from Table 3 that (i) T cost S 1 ; S 2 ; R increases as v or l or l1 decreases; (ii) the optimum values of
S2 and R, S 2 and R* decrease as v or l1 increases.
The reader is referred to Tables 4,5 to arrive at the same conclusions as listed above for Policy 1.
It appears from our numerical investigations that (i) the optimal number of the more expensive spare is
zero; and (ii) T cost S 1 ; S 2 ; R and the optimum values of (S 1 ; S 2 ; R, S 1 ; S 2 ; R are all the same even though
a1 varies from 0 as much as 0.05.
6. Conclusions
The matrix-geometric method provided in this paper works eciently for the M/M/R machine repair problem with spares and server vacations in which vacation time distributions of the servers follow the negative
exponential distribution. We show that the results are more general than existing results in the literature
(see Sivazlian and Wang [4], Gupta [14], and Gross and Harris [18]). For both models of Policy 1 and Policy
2, a total expected cost function subjected to a minimum specied level of system availability is developed to
determine the optimal number of spares and the number of servers simultaneously, so as to minimize the total
expected cost function.
Acknowledgement
This work was supported by National Science Council of Taiwan, ROC under the grant Contract NSC 932213-E-025-005.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]

J. Medhi, Stochastic Processes, Academic Press, San Diego, CA, 1991.

W. Feller, An Introduction to Probability Theory and Its Applications, Vol. I, John Wiley and Sons, New York, 1967.
F.J. Toft, H. Boothroyd, A queueing model for spare coal faces, Operat. Res. Quart. 10 (1959) 245251.
B.D. Sivazlian, K.-H. Wang, Economic analysis of the M/M/R machine repair problem with warm standbys, Microelectron Reliab.
29 (1989) 2535.
K.-H. Wang, Cost analysis of the M/M/R machine repair problem with mixed standby spares, Microelectron Reliab. 33 (1993) 1293
1301.
K.-H. Wang, H.-C. Lee, Cost analysis of the cold-standby M/M/R machine repair problem with multiple modes of failure,
Microelectron Reliab. 38 (1998) 435441.
H. Ashcroft, The productivity of several machines under the care of one operator, J.R. Stat. Soc. B 12 (1950) 145151.
E.A. Elsayed, An optimum repair policy for the machine Interference problem, J. Opl. Res. Soc. 32 (1981) 793801.
J.E. Hilliard, An approach to cost analysis of maintenance oat systems, AIIE Trans. 8 (1976) 128133.
D. Gross, H.D. Kahn, J.D. Marsh, Queueing models for spares provisioning, Nav. Res. Logist. Q. 24 (1977) 521536.
D. Gross, D.R. Miller, R.M. Soland, A closed queueing network model for multi-echelon repairable item provisioning, IIE Trans. 15
(1983) 344352.
B.T. Doshi, Queueing system with vacations-a survey, Queueing Systems 1 (1986) 2966.
H. TakagiQueueing analysis: A foundation of performance evaluation, vacation and priority systems, Part I, Vol. 1, North-Holland,
Amsterdam, 1991.
S.M. Gupta, Machine interference problem with warm spares, server vacations and exhaustive service, Performance Evaluation 29
(1997) 195211.
K. Thiruvengadam, N.K. Jaiswal, Applications of discrete transforms to a queueing process of servicing machines, Opsearch (India) 1
(1964) 87105.
M.F. Neuts, Matrix Geometric Solutions in Stochastic Models: an Algorithmic Approach, The Johns Hopkins University Press,
Baltimore, 1981.
F. Benson, D.R. Cox, The productivity of machines requiring attention at random interval, J.R. Statist. Soc. B. 13 (1951) 6582.
D. Gross, C.M. Harris, Fundamentals of Queueing Theory, 2nd ed., John Wiley and Sons, New York, 1985.