[brite-users] Re: Huge node sets and Dijkstra's SP algorithm

Vladimir Blagojevic vladimir@cs.yorku.ca
Tue Feb 25 14:01:02 2003

Hey Anukool,

> More precisely, RT_BORDER stands for a border router in an AS.  That is,
> it is a router in AS A which has neighbors that fall in other ASes.
> Also, E_AS stands for an inter-AS edge, i.e. an edge that connects two
> ASes, or equivalently, two border routers belonging to two different ASes.

For topdown approach , it seems that I can not generate router level that
is singly connected graph on AS level - simple path between each two ASes
using only inter-AS edges.

> If you decide to do this, you may want to use the
> BottomUpHierModel.   This model does the following:  first, it
> generates a single-level router topology.  Second, it groups routers
> into AS boundaries  until a specified threshold (size) is reached.

Yes but as things stand right now , it seems to me that bottomup doesn't
differentiate between RT_NODEs and RT_BORDER nodes. Does this imply that
each node is directly connected to router level graph?

> You could also easily extend the TopDownHierModel so that it generates a
> router level topology for each AS of the size you desire.   I think that
> right now TopDownHierModel generates fixed size router topologies for each
> AS (clearly not realistic, but it was intended to be a dummy model that
> others would extend).

I think I have to look into this option :(

> I would also suggest you take a peek at Tangmunarunkit et al's CCR
> paper: Does Size Determine AS Degree.  In this paper they found that they
> number of routers corresponding to an AS (its size) follows a heavy tailed
> distribution.   Another suggestion would be to use rocketfuel
> traces for your router level topologies.

What is the difference between rocketfuel and skitter?