Contents Telektronikk - Telenor
Contents Telektronikk - Telenor
Contents Telektronikk - Telenor
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
much less if a basic bandwidth unit can<br />
be found.<br />
This recursion formula has been extended<br />
to Bernoulli and Pascal arrivals in [2],<br />
but in these cases we only get approximate<br />
values for the connection blocking<br />
probabilities.<br />
4.2.2 Convolution algorithm<br />
This algorithm for calculating the global<br />
state probabilities (1), is recursive in the<br />
number of sources. The algorithm was<br />
first published in [7]. In the paper it is<br />
shown that it is allowed to truncate the<br />
state space at K (K = the number of traffic<br />
types) and renormalise the steadystate<br />
probabilities. For Poisson traffic<br />
this method is not so fast as the recursion<br />
formula above. For Bernoulli and Pascal<br />
traffic this method gives exact values, not<br />
only for the time blocking probabilities,<br />
but also for the connection blocking<br />
probabilities.<br />
The algorithm goes in three steps. In the<br />
first step we calculate the state probabilities<br />
pi (ni ) (ni = 0, ..., n max<br />
i = D/di ) for<br />
each traffic type i as if this traffic type<br />
was alone in the system. For this we can<br />
use the product form expression (1) with<br />
K =1.<br />
In the next step we calculate the global<br />
state probabilities by successive convolutions:<br />
Q1 (n1 ⋅ d1 ) = p1 (n1 ) for n1 = 0,...,n max<br />
1<br />
d/di �<br />
Qi(d) = Qi−1(d − j · di) · pi(j)<br />
j=0<br />
for d = 0,...,D and i = 2,...,K<br />
The global state probabilities are now<br />
given by Q(d) = QK (d) for d = 0,...,D<br />
after normalising.<br />
The time blocking probability can then<br />
be calculated as above, but for Bernoulli<br />
and Pascal arrivals we need another step<br />
to calculate the connection blocking<br />
probabilities.<br />
In this third step we deconvolute Q(d) by<br />
finding Qi (d) such that<br />
d/di �<br />
Q(d) = Q i (d − j · di) · pi(j)<br />
j=0<br />
The connection blocking probability for<br />
traffic type i is now given by:<br />
Bi =<br />
�D d=D−di+1<br />
� D<br />
d=0<br />
�d/di j=0 λi(j) · Qi (d − j · di) · pi(j)<br />
�d/di j=0 λi(j) · Qi (d − j · di) · pi(j)<br />
Due to numerical problems it may be better<br />
to calculate Q i by convolutions of<br />
homogenous state probabilities p i .<br />
4.2.3 Lindberger method<br />
This is an approximate method.<br />
Let dmax = max{d1 ,...,dK } and n = D -<br />
dmax . We assume that dmax is much<br />
smaller than n. The blocking states are<br />
now approximated by [11]:<br />
�<br />
Q(n + k) ≈ Erl � n ′ , A<br />
z<br />
for k = 1,...,dmax , where<br />
K�<br />
A =<br />
is the total bandwidth demand of the<br />
offered traffic,<br />
z =<br />
is the peakedness factor (and a measure<br />
of the mean number of slots for a connection<br />
in the mixture (‘equivalent average<br />
call’), see [3]),<br />
n ′ =<br />
i=1<br />
Ai · di<br />
� K<br />
i=1 Ai · d 2 i<br />
A<br />
1 n + 2 1<br />
−<br />
z 2<br />
z<br />
and Erl(x,A’) is the Erlang formula (linear<br />
interpolation between integer values<br />
of x).<br />
The validity of this method for ATM<br />
should be tested and perhaps other values<br />
of n’ should be chosen depending on the<br />
traffic mix.<br />
Now the idea is to divide the traffic types<br />
into two classes and to dimension capacity<br />
using the model above for each of<br />
these classes separately. The first class is<br />
for low capacity traffic, i.e. suggested for<br />
connections with a highest peak rate /<br />
effective bandwidth in the order of<br />
0.5 Mb/s on a 155 Mb/s ATM link. For<br />
this class the objective is to control the<br />
average connection blocking probability<br />
·<br />
� � k<br />
z A<br />
D<br />
E =<br />
(only the low capacity traffic types are<br />
considered now), and the highest individual<br />
connection blocking probability in<br />
the mixture Emax = max{Ei |i = 1,...,K}.<br />
If dmin = min{d1 ,...,dK } approximate formulas<br />
for these are<br />
For the other class, i.e. suggested for<br />
connections with peak rate/effective<br />
bandwidth between 0.5 Mb/s and 5–10<br />
Mb/s, the requirement for connection<br />
blocking probability is not so strong. For<br />
this class it is suggested to use the same<br />
connection blocking requirement for all<br />
traffic types in the class (trunk reservation,<br />
see below). An approximation for<br />
this blocking probability is given by<br />
Traffic types with higher capacity<br />
requirements will have to be treated<br />
separately, perhaps on a prebooking<br />
basis.<br />
4.2.4 Labourdette & Hart method<br />
This is another approximate method.<br />
Let<br />
the unique positive real root of the polynomial<br />
equation<br />
and<br />
� K<br />
i=1 Ai · di · Ei<br />
A<br />
�<br />
D − z + dmin<br />
E ≈ Erl<br />
,<br />
z<br />
A<br />
�<br />
z<br />
Emax ≈ dmax<br />
z<br />
·<br />
� � (dmax−z)/2z<br />
D<br />
· E<br />
A<br />
�<br />
D − dmax + dmin<br />
E0 ≈ Erl<br />
,<br />
z<br />
A<br />
�<br />
z<br />
K�<br />
K�<br />
A = Ai · di, V =<br />
i=1<br />
K�<br />
di · Ai · z di = D<br />
i=1<br />
ζ = log � D<br />
A<br />
log(α)<br />
�<br />
i=1<br />
Ai · d 2 i , α,<br />
for α ≠1. The blocking probability for<br />
traffic type i is then by this approximation<br />
method given by [10]:<br />
143