[brite-users] Waxman concerns

David Manz davidmanz at uidaho.edu
Wed Jul 19 15:30:59 EDT 2006


  I am looking and using Brite for various network topologies, from
small LAN to an Internet representation. 

However the more I have looked into Brite the more confused I became.

Using the default settings HS: 1000, LS: 100, Node: 1000, Waxman model,
alpha: 0.15, beta 0.2, and M=2 I got some interesting results.

Using the java exe growth type inc gives 2000 links.
Using the java exe growth type all gives 999 links.

Using the C++ exe growth type inc gives 2000 links (same)
Using the C++ exe growth type all gives a varying number between 4000-6000.

What explains the discrepancy between the java and C++ version?

Furthermore, why when using the waxman model is there even a difference
between growth types?  Waxman unlike BA does not have any room for
growth, it simply connects nodes based on distance and weights, (isn't
that correct).  But, in effect, Brite does N*M for inc growth, and N-1
for all growth. 

Why isn't the number of links based upon topology and the alpha value?

Now, given the Waxman model:
P(u, v) = ae-d/(bL)  

a should be the value to affect the number of links. 
However given default simulator conditions the varying of a (alpha) has
NO affect on the number of links. And the simulator simply creates N*M
or N-1 depending on growth type selected. 

Which really leads us to the question of what is m?  It is defined in
the user guide as "Number of connections per new node"?  But it seems
that the waxman model should be deciding the number of connections per
node not m? M seems to have superseded the algorithm?

Sorry for the long email, I really hope that someone can shed some light
onto the waxman implementation in Brite. 

In Summary:
Q1: diff between java and C++?
Q2: why does growth type even affect waxman? 
Q3: what is M and why does alpha have no affect?

   David Manz

More information about the brite-users mailing list