Академический Документы
Профессиональный Документы
Культура Документы
�������������������������
Volume 53 No. 1 2009, 133-151
V. Thangaraj1
�, S. Vanitha2
1,2Ramanujan Institute for Advanced Study in Mathematics
University of Madras
1e-mail: thangarajv@unom.ac.in
2e-mail: vanithasudhakar05@gmail.com
1. Introduction
In the study of queueing systems, determination of transient solution is very
much essential to analyze the behavior of the system. Transient analysis is
very useful for all queueing models to obtain optimal solutions which pave
way to control the system. Among several methods, continued fraction is one
of the techniques that is used to obtain transient solution. Even in the case
of an simple M/M/1 queue, analytical approach to obtain transient behavior
�Correspondence author
V. Thangaraj, S. Vanitha
is very difficult. In this regard, we have obtained the transient solution for
an M/M/1 feedback queue which is subject to catastrophes by employing an
effective continued fraction technique which is very simple.
The rest of the paper is organized as follows: In the next section, we des
cribe the mathematical model andpresent thedetailed analysis of the transient
solution using continued fraction methodology. Section 3 provides correspondi
ng steady state results and certain performance measures are given in Section
4. Section 5 investigates the busy period of our model. Finally, we give some
numerical examples to discuss the system performance measures.
ON THE ANALYSIS OF M/M/1 FEEDBACKQUEUE WITH...
#.
n
.
t p
# 2
(2 #�q(t -u))
-(#+�q+)(t-u)
In
�q (t -u)
where
-(#+�q+)t
P0(t)= n+1
(�q)n e
(#�q) 2
t
n=0
t p
1
.
-(#+�q+)u
+ n+1
(�q)n e du.
(#�q) 2
n=0
t. Let Pn(t)= P(X(t)= n), n =0,1,... be the state probabilities that there
V. Thangaraj, S. Vanitha
are n customers in the system at time t. Based on the above assumptions and
bytheChapman-Kolmogorovforwarddifferential-difference equations, the state
probabilities Pn(t),n =0,1,... forthequeueing system underinvestigation are
P0'
(t)= -#P0(t)+�qP1(t)+(1-P0(t)), (2.1)
P
'
(t)= #Pn-1(t)-(#+ �q+ )Pn(t)+�qPn+1(t),n =1,2,...,(2.2)
P0(0)=1. (2.3)
In the sequel, for any function f(�), letf#(z)denote its Laplace transform.
By taking Laplace transform, the above system of equations is transformed to
the following system of simultaneous equations.
##
(z + #+ )P0
(z) =1+ + �qP 1
(z), (2.4)
###
nn+1(z),
1+
# z
P0
(z)= , (2.6)
�qP1
(z)
(z + #+ )-
#
P0
(z)
Pn(z) #
= ,n =1,2,... . (2.7)
Pn-1(z) �qP
n+1(z)
(z + #+ �q+)-
P(z)
1+
# z
P0
(z)= . (2.8)
#�q
(z + #+)-
#�q
(z + #+ �q+ )-
(z + #+ �q+ )-���
1+
# z
P0
(z)= , (2.9)
(z + #+ )-#(z)
ON THE ANALYSIS OF M/M/1 FEEDBACKQUEUE WITH...
where
#�q
#(z) = (2.10)
#�q
(z + #+ �q+)-
(z + #+ �q+ )-���
#�q
= .
(z + #+ �q+)-#(z)
It is clear that #(z)satisfies the quadratic equation
! � !2
-4#�q
#(z),
(z)= ,
2
where ! = z +#+�q+. It is seen that
(z)is a uniquereal root within[0,1)
and 0 # z< 1. Hence we consider only
(z)for further discussion.
Substituting
(z)for #(z)in(2.9), weget after some algebraic calculations
1+
# z
P0
(z)=
".
.
(2.11)
! - !2
-4#�q
(z + #+ )-
! - !2
-4#�q
1+
z 2#�q
=
".
.
(2.12)
! - !2
-4#�q
1-�q
2#�q
p!n+1
! - !2
-4#�q
=(�q)n
2#�q
n=0
p!n+1
! - !2
-4#�q
+(�q)n . (2.13)
z 2#�q
n=0
Oninversion,(2.13) yieldsthefollowing explicit expressionfor P0(t).
P0(t)= n+1
(�q)n e
n=0
(#�q) 2
t
t p
1
.
-(#+�q+)u
+ n+1
(�q)n e du, (2.14)
u
(#�q) 2
n=0
0
V. Thangaraj, S. Vanitha
where In(�)is the modified Bessel function of the first kind of order n.
Pn(z)
Pn-1(z)
= . (2.15)
#�q
(z + #+ �q+ )-
#�q
(z + #+ �q+ )-
(z + #+ �q+ )-���
P (z) #
=
#.
#
,for n =1,2,..., (2.16)
Pn-1(z) ! + !2
-4#�q
2
.
n
! - !2
-4#�q
#
P (z)= P0
# (z), for n =1,2,... (2.17)
2�q
where ! = z + #+ �q+ .
#.
n
.
p
# (2 #�q(t -u))
-(#+�q+)(t-u)
In
�q (t -u)
".
#k
P0(t)!
(#+ )k+1
2
k=0
as t !1. (2.19)
ON THE ANALYSIS OF M/M/1 FEEDBACKQUEUE WITH...
Proof. We have,
1+
# z
P0
(z)= , (2.20)
#�q
(z + #+ )- #.
.
! - !2
-4#�q
! -
where ! = z + #+ �q+ .
z!0
2#�q
(#+ )- .
(#+ �q+ )+ (#+ �q+ )2
-4#�q
Multiplying numerator and denominator of the right hand side of above equat
ion by
".
#k
P0
(z)# ,
z (#+ )k+1
2
k=0
as z ! 0. (2.22)
(#-�q)2�q
m(t)! +
h.
Proof. Let us consider equations (2.1) and (2.2) with initial condition
(2.3).
Let P(z,t)= Pn(t)z. From the above assumptions, it is clear that the
n=0
@P(z,t) �q 1
@t z z
V. Thangaraj, S. Vanitha
Figure 1: The graph between #0
and #.
G1: � =5, =3,q =0.5 G2: � =5, =3,q =0.6
G3: � =8, =3,q =0.5 G4: � =5, =4,q =0.5
.
@P(z,t)
The mean system size is m(t)= nPn(t)= |z=1.
@z
n=1
Solving the above differential equation for m(t) with m(0) = nPn(t) =0,
n=1
we obtain,
# -t)-
�q -(t-u)
If m #(z)is the Laplace transform of m(t), then from above equation we get
# �q �q
m # (z)= - + P0
(z). (2.27)
z(z + ) z(z + ) z +
If > 0,thenfrom(2.27) and(2.11) we obtainthe required result(2.23).
ON THE ANALYSIS OF M/M/1 FEEDBACKQUEUE WITH...
Theorem 4. For > 0, the steady state distribution {#n;n # 0}of the
M/M/1 feedback queue with catastrophe corresponds to
#0
= (1-#), (3.1)
#n = (1-#)#n n =1,2,... (3.2)
where
(#+ �q+ )- #2
+ �q2
+ 2
+2# +2�q -2#�q
# = . (3.3)
2�q
V. Thangaraj, S. Vanitha
Figure 3: The graph between E(X)and .
G1: # =4,� =5,q =0.8 G2: # =4,� =5,q =0.6
G3: # =3,� =5,q =0.8 G4: # =4,� =6,q =0.8
Proof.
1+
# z
P0
(z)=
".
#
, (3.4)
! - !2
-4#�q
(z + #+)-
2
where ! = z + #+ �q+ .
Now multiplying(3.4) by z on both sides and taking the limit as z ! 0 we
get,
lim zP 0
(z)=#0
=
".
#
.
z!0
( + #)-
2
After some algebraic manipulation, the above expression reduces to
#0
= ,
2�q
ON THE ANALYSIS OF M/M/1 FEEDBACKQUEUE WITH...
which simplifies to
(#+ �q+ )- #2
+ �q2
+2
+2# +2�q -2#�q
#0
=1- .
2�q
.
!n
! - !2
-4#�q
##
z!0
2�q z!0
#n = lim zP (z)
z!0
".
#n
(#+ �q+ )- #2
+ �q2
+ 2
+2# +2�q -2#�q
=#0
.
2�q
#(z)= . (4.1)
1-#z
The mean, the variance of the system size namely E(Q),V ar(Q)and the mean
of the queue size E(Qq)are obtained as
E(Q)= , (4.2)
1-#
#
Var(Q)= , (4.3)
(1-#)2
#2
and E(Qq)= , (4.4)
(1-#)
V. Thangaraj, S. Vanitha
where # is given by equation (3.3).
Now E(Qq)= .
(n -1)#n = #2
1-#.
n=1
n=1
andP(Server is idle) = #0
=(1-#), (4.6)
where # is asgivenin(3.3).
U = �q #, (4.7)
Proof. The system throughput, U, is the rate at which customers exit the
queue whenever there are one or more customers in the system. With the exit
rate �q, we obtain
U =[1-#0]�q = �q #, (4.8)
Remark. The number of customers that were ejected per unit time is
#
n#n = E(X)= . (4.9)
1-#
n=1
Therefore, the fraction of customers who entered the queue and were ejected
when the server failed due to catastrophe is
E(X) #
= . (4.10)
##(1-#)
U �q
= #. (4.11)
##
ON THE ANALYSIS OF M/M/1 FEEDBACKQUEUE WITH...
We have {N#(0)=1}and the duration of the busy period is the first t for
which {N#(t)=0}. Let qn(t)= P{N#(t)= n|N#(0)=1},n =0,1,... be the
zero-avoiding state probabilities. Then q1(0)=1,qn(0)=0,n =1,2,....
q1'
(t)= -(#+ �q+ )q1(t)+�q q2(t), (5.1)
q
'
(t)= #qn-1(t)-(#+ �q+ )qn(t)+�q qn+1(t),n =2,3,.... (5.2)
#n/2
#n p
-(#+ �q+ )t
�q #t
j=1
##1/2
# 1
-(#+�q+)t
�q #t
V. Thangaraj, S. Vanitha
The Laplace transform of T is given by
(z + #+ �q+ )- (z + #+ �q+ )2
-4#�q
= .
2#
E(T)= - b # (z)|
z =0
dz
=
.
, (5.4)
2# (#+ �q+)2
-4#�q
d 2�q
z =0 3
dz 2
[(#+ �q+ )2
-4#�q]
##2
8�q#2
-!1
(#+ �q+ )2
-!1
V(T)= 3
, (5.6)
4#2
[(#+�q+ )2
-4#�q]2
where !1
=(#+ �q+ )2
-4#�q.
6. Numerical Illustrations
In the Sections 3 and 4, we have obtained explicit expressions for #n,n =
0,1,��� , the steady state probabilities of the system size, E(X), the mean
number of customersin the system, U, the system throughput. In this section,
we present some numerical illustrations to show the effect of the catastrophe
parameter and the arrival rate # on #0
E(X)and U.
Acknowledgements
References