Contents Telektronikk - Telenor

144
Number of type
2 connections
Figure 3 Set of feasible states with a completely partitioned trunk
group, case with 2 traffic types
Ei ≈
Feasible state
⎧
⎨
⎩
1−α d i
1− D
A
· Erl
di·A
V · Erl
�
D
ζ �
D·A
V
�
A , ζ
, A2
V
The Lindberger and Labourdette & Hart
methods are meant for network planning
purposes where the calculations must be
iterated a great number of times in the
process of optimising a network. For the
purpose of dimensioning of the trunk
groups between a network node and the
adjacent network nodes the exact methods
can be used.
5 Service protection
methods
, α �= 1
�
, α =1
A complete sharing strategy will probably
not be an optimal dimensioning strategy,
though this depends on the traffic
mix and the required maximum connection
blocking probability for the different
traffic types. The reason for this is that a
complete sharing strategy will result in a
higher connection blocking probability
the higher the bandwidth demands are.
An effect of this will be that requests for
high bandwidth services will be rejected
in heavy traffic periods due to lack of
resources. Under normal conditions the
connection blocking probability may be
satisfactory, but then at the expense of a
low network utilisation.
boundary given by
linear CAC
Number of type
1 connections
For trunk groups consisting of more than
one ATM link, this may partly be remedied
by using routing methods when
hunting for free capacity. Two such routing
methods are
- the concentration method, and
- the separation method.
The concentration method is a sequential
hunting with homing. Since we then
always start the hunting with the same
link, this link will always be heavily
loaded and the last link in the sequence
will be less loaded (depending on the
offered traffic).
When using the separation method we
still do sequential hunting with homing,
but the traffic types are divided in two
classes (the types with the highest bit
rates is one class) with opposite homing
positions.
The advantage of these methods is a
lower connection blocking probability
for traffic types with high capacity
demands. A disadvantage is a higher processor
load due to more retrials before
success in the hunting process.
Routing methods alone are not the best
way to compensate for the bad effects of
a complete sharing stategy and we have
to use service protection methods for best
network utilisation with given connection
blocking constraints. Methods for pro-
tecting certain services in the connection
establishment phase will normally give a
lower dimensioned capacity for carrying
the offered traffic streams with the constraints
given by these connection blocking
objectives. Such methods are described
below.
5.1 Completely partitioned
trunk groups
This is the straightforward service protection
method in which case the trunk
group is divided in K parts, one for each
traffic type. It is illustrated in Figure 3
for K = 2. The needed capacity is dimensioned
for each traffic type separately,
based on offered traffic and connection
blocking objective for this traffic type.
Standard dimensioning methods can be
used as if only this traffic type was offered
to the system. The total dimensioned
capacity will be the sum of the dimensioned
capacity for each traffic type.
This method may be used in combination
with one of the methods given below, i.e.
the traffic types may be divided in
classes with complete partitioning of the
trunk group between the classes and use
of one of the methods below inside a
class.
5.2 Traffic type limitation
method
This method is also called the partial
sharing method.
The total number of connections of traffic
type i is limited by
ni ≤ n max
i
for i = 1,...,K, where
K�
i=1
≤ D
di
The method is illustrated in Figure 4 for
a case with 2 traffic types,
n max
1
di · n max
i
>D.
= D
,n
d1
max
2
< D
.
d2
For this method we still get a product
form solution for the state probabilities
and we may also apply a modified version
of Roberts and Kaufmanns recursion
algorithm for the overall occupancy distribution: