Contents Telektronikk - Telenor

180
0.005
0.004
0.003
0.002
0.001
0
0 500
0.005
0.004
0.003
0.002
0.001
Turn around time
0
0 50 100 150 200 250 300 ms
1000
1500
0.65
0.35
2000
2500
73
ms
1999
ms
3000
215 ms
3500
73
ms
4000 ms
a) ATM network user working toward computer with turn around times
73
ms
Turn around time
and user think time
Data
transf.
b) Corresponding source type model
Figure 2.11 Example of a source type model
1
1 12
2048
Table 2.1 Parameters governing the next transition of the composite model
The time until next state change is negative
exponentially distributed with parameter
The next type k and state i combination to be affected
by a state change is given by the probability
The specific source among these which is changing
state is given by the probability
The next state j of this source is given by the probability
� �
k
�
k
i
m (k)
1
m (k)
i (t)
T (k)
i
i (t)/T (k)
i
�
i m(k) i (t)/T (k)
i
m (k)
i (t)
p (k)
ij
2048
(3)
(4)
(5)
(6)
1.52 Mbit/s. The data is transferred as
512 byte blocks, split into 12 cells, which
are sent back to back.
The think and turn around time of the
user is measured with results as shown in
the figure. Two modes of the user may be
identified, typing and thinking, represented
by the upper and lower branch in Figure
2.11.b respectively. The number of
states and the parameters of the user-submodel
may be determined from the measured
data for instance by a non-linear
fit.
2.4.4 Composition
Up till now we have dealt with the
modelling of a single source type. To
meet the requirements for a useful generator,
it is necessary to generate traffic
from a number of sources which may be
of different types. In the rest of this subsection,
it is outlined how this is done.
Say we have K different source types
with m (1) , ..., m (k) , ..., m (K) sources of the
different types. Of the k’th type, let m (k)
i
denote the number of sources in state i at
time t. Since the composition of Markov
processes also is a Markov process, the
set of {m (k)
i (t)}∀(k,i) constitutes the global
state of the dialogue and burst level
part of the model.
The transition matrix for the composition
of sources can be obtained by Kroenecker
addition of the transition matrixes of
the m (1) , ..., m (k) , ..., m (K) sources. The
matrix become extremely large. A simple
example for illustration is shown in Figure
2.12. The state space of three two
state speech sources and one three state
date source yields a 24 state global state
space. The largest burst level state space
which may be defined for the STG has
104315 states.
However, by the principle used for generation,
it is not necessary to expand this
state space. The state space is traversed
according to a set of rules given by the
source type definitions and the number of
sources of the different types. The selection
of the next global event, state change
and time when it occurs, is governed by
the probabilities shown in Table 2.1.
The discussion of the state space above
relates only to the burst level state space.
In addition, a burst level state may be
entered at an arbitrary time relative to the
cyclic pattern described in Section 2.4.2.
Hence, we must add these patterns together
with a correct phase in order to
produce the multiplexed output from the
composition of the m (1) , ..., m (k) , ..., m (K)