Next: Translating stationary probabilities into
Up: Matrix analytic methods
Previous: Matrix analytic methods
Contents
We first discuss some key properties of the stationary probabilities
in a QBD process. Consider a simpler case of the birth-and-death
process having the generator matrix shown in (3.1).
Let be the stationary probability that the birth-and-death
process is in level for . Then, it is easy to see
that
, where
, for . It turns out that
the stationary probabilities in a QBD process have a similar property.
Let
be the stationary probability vector in a QBD
process having generator matrix shown in (3.2).
Here, the -th element of vector
denotes the
stationary probability that the QBD process is in phase of level
, i.e. state (). It turns out that there exists a matrix
such that
for each .
Specifically, the stationary probability vector in the QBD process is
given recursively by
|
(3.3) |
where
is given recursively via:
|
(3.4) |
When the QBD process has an infinite number of levels, the state space
of the QBD process needs to be truncated, so that
matrices can be calculated from a certain large
enough integer to recursively via (3.4).
The truncation level needs to be chosen carefully such that
the stationary probability that the QBD process is above level
is negligible [28].
When the QBD process repeats after a certain level, , (i.e.,
,
, and
for all
),
is the same for all , and
(for ) is given by the
minimal nonnegative solution to the following matrix quadratic
equation3.3:
|
(3.5) |
Once
is obtained,
for
,
is given recursively via (3.4).
Various approaches for calculating have been proposed, and
in Figure 3.12 we show an algorithm for calculating ,
's, and other relevant matrices.
Figure:
Algorithms for calculating ,
's,
and other relevant matrices [111].
The input matrices (, , and )
for part (b) are given by the output of part (a).
Input: , , , |
Output: , , |
; |
; |
; |
; |
repeat |
; |
; |
; |
; |
; |
; |
; |
until
|
; |
; |
|
Input: , , |
Output:
's,
's,
's |
|
|
|
for to 1 |
|
|
|
end |
|
|
Once and
's are obtained,
the stationary probability vector
can be calculated recursively from
via (3.3) for . Thus, all that remains is to calculate .
Vector is given by a positive solution of
normalized by the equation
where and are vectors with an appropriate number of
elements of 0 and 1, respectively.
Next: Translating stationary probabilities into
Up: Matrix analytic methods
Previous: Matrix analytic methods
Contents
Takayuki Osogami
2005-07-19