"Although their research is quite interesting there are a few things which limit its practicality. They can only detect tunnels going through ssh servers which they control."
Yeah. In the end, P2P protocols will just end up looking like HTTP requests and responses, with encrypted data steganographically encoded in what looks like your online bank. (Maybe incoming HTTP requests to a home DSL line would look suspicious, so let's make it XMPP or something.)
You can't filter the Internet. You can't break encryption. These are the realities of the Internet, so the ISPs had better start turning on their dark fiber. Sparing a nuclear war, Internet use is never going to decrease. Stop fighting entropy and "roll with it."
Sorry to burst your delusional bubble, but you don't need to break encryption in this case. The term to describe this particular "attack" is traffic analysis, and it is a real bitch to get around. P2P protocols will not end up looking like HTTP requests and responses because the nature of the information flow is nothing like a standard HTTP exchange; you do not send a small packet, get back a large chunk, and then exchange small updates -- when P2P traffic flows it is large chunks and its bidirectional nature is a dead giveaway.
The application of a simple Bayesian recognizer to categorize traffic flows might be new to the academic community, but among people developing large-scale traffic management and security monitors this is really old stuff.
this story has absolutely nothing to do with ISP throttling. If people continue using subjects that are just trying to be attention grabbers, HN will end up like digg.
To sum up the actual story, researchers used Bayesian learning filters to identify the type of encrypted tunnels that are going through your ssh server within 90-99% accuracy. Has absolutely nothing to do with ISP throttling p2p.
and only I control my server, so I'm good to go.