1

Copyright 2002, Sanjay K. Bose 1

Analysis
of
A Finite Capacity, Single Server Queue
(M/G/1/K)
Copyright 2002, Sanjay K. Bose 2
(K-1)Waiting
Positions
Jobs leaving
without service
Arrivals Departures
Server
A Finite Capacity, Single Server, M/G/1/K Queue
Probability of Blocking
P
B
= P{arrival finds queue full}

P
B

(
1
-
P
B
)
(1-P
B
)
2
Copyright 2002, Sanjay K. Bose 3
We use the same notation as that used earlier for the M/G/1
queue in Section 3.2.
The system state (i.e. the number in the system) at the
imbedded points corresponding to the time instants just after a
job completion will form a Markov Chain
n
i
= Number left behind in the system by the i
th
departure
Imbedded Points Departure Instants of Jobs after
completing service
The max function in (1) will lead to loss of jobs which are
denied entry into the queue when the system is full
) 1 ( ,......, 1 } 1 , 1 min{
0 } 1 , min{
1
1 1
+

+
+ +
K n for K a n
n for K a n
i i i
i i i
(1)
Copyright 2002, Sanjay K. Bose 4
Considering the Markov Chain of states at the imbedded points
(corresponding to the departure instants), we will have at
equilibrium -
State Probability at equilibrium:
p
d,k
= P{system in state k} k=0,1,.....,(K-1)
} | {
1 ,
j n k n P p
i i jk d

+
State Transition Probability at equilibrium
0 j, k (K-1)
Note that the system state at the departure instant can only
be between 0 and (K-1)
3
Copyright 2002, Sanjay K. Bose 5

k
= P{k arrivals occurring in a service time}
dt t b e
k
t
t
t
k
k
) (
!
) (
0

(3)
Note that
k
may also be found as the coefficient of z
k
in the
expansion of L
B
( -z)
The state transition probabilities p
d,jk
for this Markov Chain (at
the departure instants) may then be found in terms of
k
as
given in the next slide
Copyright 2002, Sanjay K. Bose 6

'

1
0 ,
1
2 0
K m
m
k
k d
K k
K k
p

'

+
j K m
m
j k
jk d
K k
K k j
p
1
2 1
1
,

j=0
1 j (K-1)
(4)
State Transition Probabilities for the Departure Instants
4
Copyright 2002, Sanjay K. Bose 7

1
0
, , ,
1 ,........ 1 , 0
K
j
jk d j d k d
K k p p p
K Balance Equations
(5)
Normalization Condition

1
0
,
1
K
k
k d
p (6)
As usual, we can solve for the equilibrium departure state probabilities
{p
d,j
} j=0,1,.,(K-1) using any (K-1) equations from (5) along with
the normalization condition of (6).
Copyright 2002, Sanjay K. Bose 8
Alternatively, we can solve for {p
d,j
} j=0,1,.,(K-1) using the
following -

+
1
0
,
1
1
1 , 0 , ,
1
2 , ,......... 1 , 0
K
k
k d
k
j
j k j d k d k d
p
K k p p p
(7)
See notes for another solution approach
We now need to use {p
d,k
} k=0,1,.,K-1 to find the equilibrium
state probabilities {p
k
} k=0,1,.,K at an arbitrary time instant.
We would also like to find the probability P
B
that an arrival
finds the system full and is blocked, i.e. leaves without service.
5
Copyright 2002, Sanjay K. Bose 9
We summarize our equilibrium state probability definitions as
the following
{p
d,k
} k=0,.,K-1 State probabilities at departure instants
{p
a,k
} k=0,.,K State probabilities at arrival instants
regardless of whether the job joins the
queue or is blocked
{p
ac,k
} k=0,.,K-1 State probabilities at an arrival instant
when the job actually does join the
queue
Note that the departure instant implies the instant just after
a departure and the arrival instant implies the instant just
before the actual arrival.
Copyright 2002, Sanjay K. Bose 10
PASTA K k p p
k a k
,........, 1 , 0
,
(10)
1 ,....... 1 , 0
, ,
K k p p
k ac k d
Kleinrocks
Result
(11)
and

1
0
, ,
1
K
k
k a K a B
p p P
Arrival is not blocked
and finds k in system
Therefore
1 ,......., 1 , 0 ) 1 ( ) 1 (
, , ,
K k p P p P p p
k d B k ac B k a k
(12)
6
Copyright 2002, Sanjay K. Bose 11
Average arrival rate of jobs actually entering the system =
c
) 1 (
B c
P
Offered Traffic to the queue =
X
Actual Throughput of the Queue =
c
= (1-P
B
)
) 1 ( 1 1
0 B c
P p
probability of finding the system empty
Copyright 2002, Sanjay K. Bose 12
But from (12), for k=0, we have 0 , 0
) 1 (
d B
p P p
Therefore
0 ,
) 1 ( ) 1 ( 1
d B B
p P P
+

0 ,
1
1
d
B
p
P (14)
and
1 ,....... 1 , 0
1
,
0 ,

+
K k p
p
p
k d
d
k

(15)
Note that (15) implies that at equilibrium, for a given value of k in
the range k=0,.,(K-1), the state probabilities at an arbitrary
instant p
k
and the state probabilities at the departure instant p
d,k
are
strictly proportional.
7
Copyright 2002, Sanjay K. Bose 13
Summarizing
Find the state probabilities {p
d,k
} k=0,.,(K-1) at the
departure instants using either (5) & (6) or (7)
Find the state probabilities {p
k
} k=0,.,(K-1) at an
arbitrary instant using (15). Note that p
K
=P
B
Find the blocking probability P
B
using (14). This will also
be the same as the probability p
K
of observing the system to
be in state K at an arbitrary time instant
Copyright 2002, Sanjay K. Bose 14
Performance Results

,
_

+
+
+

K
k d
K
k
k d
d
k
p
K kp
p
kp N
0 0 ,
1
0
,
0 ,
) (
1
1
) (
1

) (
) 1 (
0 ,

+

d
B c
p
P
X p
K
kp X W W
d
K
k
k d q
+ +

) 1 (
1
0 ,
1
0
,


8
Copyright 2002, Sanjay K. Bose 15
An Alternate Analytical Approach for the M/G/1/K Queue
Consider the mean of the time interval between successive imbedded
points (i.e. departure instants).
X +

1
queue empty at the previous
departure instant
probability = p
d,0
X
queue non-empty at the
previous departure instant
probability =1- p
d,0

+
,
_

,
_

0 ,
0 ,
0 , 0 ,
0 ,
0
) 1 (
1
1
d
d
d d
d
p
p
p X p X
p
p
Therefore
Same as
in (15)
Copyright 2002, Sanjay K. Bose 16
To find p
k
, for k=1,.,(K-1), consider when an arbitrarily chosen
time instant falls within a service duration where x is the amount of
service already provided.
1
1
]
1

+
1
1
]
1

k
j
x
j k
j d c
x
k
d c k
dx
X
x B
e
j k
x
p
dx
X
x B
e
k
x
p p
1
0
,
0
1
0 ,
) ( 1
)! (
) (
) ( 1
)! 1 (
) (

(28)
when previous
departure leaves
system empty
when previous
departure leaves
system non-empty
X
x B ) ( 1
pdf of elapsed service time
c

P{chosen time instant will

fall within a service time}
9
Copyright 2002, Sanjay K. Bose 17
As before, let

0
1
0
)] ( 1 [
)! 1 (
) (
) (
!
) (
dx x B e
k
x
dx x b e
j
x
A
x
k
k j
x
j
k j
j k

k=1,2,......,

X A
k
k
1
where
Using the expression for A
k
, we can obtain
(29)
1
1
]
1

+
k
j
j k j d k d
c
k
A p A p p
1
1 , 0 ,

k=1,2,......,
(31)
(30)
Copyright 2002, Sanjay K. Bose 18
To simplify the expression for p
k
further, we use the result

+
+
k
j
j k j d k d k d
A p A p p
1
1 , 0 , ,
Prove using
recursion
Substituting this, we get the same result as obtained earlier in
(12) and (15) -
k d B k d
c
k
p P p p
, ,
) 1 (

k=0,1,,(K-1)
Note that we still need to find p
K
, the probability of finding
the system full at an arbitrary instant to complete the
analysis.
This is done in the following slides.
10
Copyright 2002, Sanjay K. Bose 19
For k=K, we need to take into account the fact that arrivals coming
when the system is full are blocked and denied entry into the
system.

+
1
1
]
1

1
1
0
,
1
0
0 ,
) ( 1
!
) (
) ( 1
!
) (
K
j j K k
x
k
j d c
K k
x
k
d c K
dx
X
x B
e
k
x
p
dx
X
x B
e
k
x
p p

(33)
Using (29),
this gives
1
1
]
1

1
1 1
, 0 ,
K
j j K k
k j d
K k
k d
c
K
A p A p p

(34)
Copyright 2002, Sanjay K. Bose 20
To simplify
(34) further,
we need the
result
1
0 ,
1
1 1
, 0 ,
+
1
1
]
1

d
K
j j K k
k j d
K k
k d
p A p A p (35)
shown by summing p
d,k
over
k=1,,K-1 and using (30)
Applying (35) to (34), and using p
K
=P
B
and
c
=(1-P
B
) we get our
earlier result
+

0 ,
1
1
d
B K
p
P p