I took a stab at estimating what it would take to knock out the top 4% of ASNs on the Internet, thus meeting Tu's threshold for partitioning the net. Of course, to really determine this requires perfect knowledge, and at least some NDA information from the providers involved. My rough estimate is 60 different physical locations worldwide would need to be destroyed. But I suspect congestive collapse would occur after 20 different physical locations were destroyed. This assumes you went after the nexus(es?) of the Net. You could possibly isolate the requisite nexuses, without destroying them. But knowing those locations requires a level of knowledge about the physical overlap between providers I don't have. I think there is a substantial physical overlap between providers' right of ways, so I wouldn't be suprised if the same level of effort to destroy the connectivity of the top 4% of the providers damaged the top 25% (SWAG) since many of the 'tier 2,' and even 'tier 3' providers co-locate in the same physical locations as the top 4%. But I have no proof, only a educated guess.
From the studies I know about, disrupting national telephone service would require less effort. A logical attack on the net would require substitutionally less effort.
This also assumes a static network. Network operators could substantially increase the number of peering locations and ASN interconnectivity in response. Those shadow connections and locations I spoke about. Unlike telephone switches, backbone routers are very fungible. So its not inconceivable, some new connectivity between ASNs would rapidly appear in places where there wasn't any before. A lot of Tier 2 and Tier 3 providers are multi-homed, but don't currently provide transit between backbones. This is an administrative policy, not a physical restriction. It is possible for substantial additional inter-ASN connectivity to be brought up very quickly, assuming you had enough remaining communications to coordinate it. Is the shadow connectivity of tier 2 and tier 3 providers enough to invalidate the theory the Internet as a scale-free network? I don't know. On Wed, 26 July 2000, Dave Crocker wrote:
The proffered theory is that a small number of nexus points that form an essential core. Take out that small number and the remaining alternate paths won't be sufficient.
25 is probably a large number, never mind one or two orders of magnitude more.
Some effort at figuring out how many (and which) would have to be simultaneously removed strikes me as a worthy exercise, to make sure that the number is high (and the specific sights are well enough protected.)
Perhaps it's already been done and the results kept quiet. That's fine.
As long as the fragility is nowhere near as extreme as the article implies.