However this gels with their explanation that they have only centralized the call setup servers for reliability and voice/video traffic for mobile devices where P2P is not very feasible. On laptops and PCs, where P2P remains feasible, traffic is still routed directly been peers.
Now I haven't run the same experiment in the past year, but I hope somebody will, and that too at larger scale. Given this forum is "hacker" news, there is a distinct lack of technical investigation to verify technical claims, and a disturbing propensity to take tech media for its word, especially when biases are being confirmed.
Few people care about the contents of VOIP calls. The relationship maps you can generate from the metadata is far more useful.
Also, "reliability" doesn't make sense as a reason to centralize skype. Being very generous, it is a workaround for the problem of NAT removing the average user's ability to self-publish on the internet.
>Also, "reliability" doesn't make sense as a reason to centralize skype.
I've written a P2P app or two. The NAT issues that you handwave are a huge problem. It's not just the connectivity of nodes, it's the symmetry in which nodes are reachable. NAT and general internet weirdness make this a much harder problem than it needs to be. I had to do a significant re-architecting when I ran into these issues during internal testing, and I had less than two dozen users!
The next problem is latency. By the time you wait to discover a suitable supernode and for your "hello" message to reach your target peer, you will have already connected and started communicating with your peer if you use a centralized "registry" server.
This is compounded by churn. Even if there's a single intermediate hop in your routing, your chances of a successful message delivery drop drastically when a peer can leave anytime. Think about how you use your laptop and how long you keep it running and how abruptly you close it. The move towards laptops away from desktops means average node uptime is reducing sharply, without even considering the move to mobile.
This is further compounded by the need for presence for the case of Skype. You don't want to find a peer is unreachable after you start a call. This implies the need for a global state. But without central server, this can only be achieved with DHT, where the first two problems are even worse. Note the existing DHTs are all used for long running "sessions" where the session is the availability of a torrent. User presence is a lot more ephemeral.
And there's the original problem for P2P apps, of course: bootstrapping. Peers don't come online knowing all their other peers on the Internet. There has to be a way for them to discover each other, which, without Internet-wide multicast, means a central server. If you're going to have to solve this problem, you might as well solve the others.
There is actually a thread on p2p-hackers mailing list about this exact issue. Many experienced P2P devs agreed that whenever you can get away with a centralized solution, you should go for it. In this context, partially centralizing Skype as they did makes complete sense.
There may have been evil reasons for Skype's change. Or maybe, MS realised that needlessly taking customer resources to run the Skype network isn't suitable behavior?
You could still glean info from monitoring the network communication (although i personally haven't done anything).
Also if the leaked docs are to be believed, the fact that the agencies boasted a 3-fold increase in intercept-able traffic post-acquisition, then it is reasonable to assume that wiretapping was non-trivial before.
I would have commented on your tinfoil hat, except for what Microsoft did to Skype post-acquisition:- Completely rewriting its protocol architecture from one which was P2P with end-to-end encryption and practically impossible to wiretap or monitor, to a centralized architecture (ostensibly for scalability reasons) which made it much more easier to wiretap or obtain metadata.
http://www.zdnet.com/article/skype-ditched-peer-to-peer-supe...
http://www.theguardian.com/world/2013/jul/11/microsoft-nsa-c...
Here's a choice quote from the above article.
"In July last year, nine months after Microsoft bought Skype, the NSA boasted that a new capability had tripled the amount of Skype video calls being collected through Prism;"