It still depends on the network that you are using for analysis.

Assuming the paper uses the same dataset as the link Sean provided the

giant cluster they are analyzing is only 8.3% of IP nodes in their

sample. It takes the removal of 25% when only looking at that small

densely connected section, it says nothing about what will happen to the

other 91.7% of nodes. Considering that 55% of the remaining nodes are

trees, they will be saying "Houston we have a problem" well before 25%.

Whether or not it matters that they have a problem in an entirely

different question. I've probably kicked this dead horse enough already

"Sean" == <sgorman1@gmu.edu> writes:

> it says nothing about what will happen to the other 91.7% of

> nodes. Considering that 55% of the remaining nodes are

> trees, they will be saying "Houston we have a problem" well

> before 25%.

The supposition would be that the remaining nodes are evenly

distributed around the core so the percentage of nodes outside of the

core without connectivity should be roughly the same as the percentage

of nodes removed from the core. At least until the core goes

non-linear...

>> It would be interesting to see what outdegree looks like as a

>> function of rank -- in the paper they give only the maximum and

>> average (geo. mean) outdegrees. Is there also a critical point

>> 25% of the way through the ranking? Probably not or one would

>> expect they'd have mentioned it...

It turns out that this is buried in one of the graphs (fig. 6) and

does not appear to have any special properties 25% of the way through.

It does have an inflection point around the 1000th node or so (2.5%).

-w