Figure 4: Average Number of Connections in Steady State.
Figure 5: Average Number of Accepted Connections (Two Destinations).
Figure 6: Average Number of Accepted Connections (Five Destinations).
In a first series of experiments, we studied the effect of the
relaxation factors 
, i.e., the factors determining whether
resources should be relaxed more at the source or at the receivers, on
the performance of the algorithm. In these experiments, we keep the
arrival rate 
constant to be 100 times the service
rate. Figure 4 shows the average numbers of accepted
connections for a varying value if and increasing numbers of
destinations per connection. In the case of one of two destinations
per connection, all connections are accepted. As to be expected, the
acceptance rate decreases with increasing numbers of destinations per
connection, as the connections put more load on the system.
For connections with one or two destinations, the value of
has virtually no effect on the call blocking probability. This is
because uneven resource allocation cancel each other out in average
for these essentially unicast connections.
As the number of destinations per connection increases, the effect of
varying becomes more pronounced. The results shows that the
acceptance rate is higher for a relaxation factor larger than 1. In
this case, more resources are relaxed at the destinations, since there
are many more destinations than senders, the total amount of resources
used per connections is reduced with increasing value for the
relaxation factor .
We notice that very large (larger than 2.0) or very small (smaller
than 0.5) values for do not significantly influence the
acceptance rate. This can be explained because the relaxation factor
determining the calculation of the target delay trd at the node,
which gives only an indication on how much resources can be
relaxed. The actual overall amount of resources freed depends on the
current resource use at that node, and the remaining portion is
carried forward for relaxation on the next node downstream. According
to the results on these experiments, we identified a value for
of 2.0 to give good resluts, and will therefore use it in the
following experiments.