Hacker News Reader

Comment by delinka | Hacker News Reader

That's just spin to make the architecture sound like a bad idea. Yes, p2p anything uses more of your upload bandwidth than any other means of communicating on the 'net. IMO, that's not "bad" because it keeps the tubes all decentralized and more capable of routing around damage. Like it's supposed to do.

View on HN

Nope, it's not about that. Skype can, if you're not running behind a NAT, make you a supernode, and in that case you will be routing calls between NAT'ed hosts "on your dime". Basically, it may be that you contribute your bandwidth to calls that have nothing to do with you. Presumably this isn't the case in the "cloud based" service, where the some server would act as a gateway between NAT'ed hosts.

They could just easily run a whole bunch of supernodes themselves in the cloud, and turn off the supernode functionality for client ip addressses. I don't see why they make it out to be a big dealbreaker.

Of course they could, but then it wouldn't be p2p anymore, right?

I think it could probably still be considered p2p. Skype only falls back to using a supernode if two clients cannot form a connection on their own: http://saikat.guha.cc/pub/iptps06-skype/

If you read my original comment, that's precisely what I said. Supernodes are used to link NAT'ed hosts. If the supernodes were run by Skype on "the cloud" they wouldn't really be peers anymore, but act as servers. That's what I said in my 2nd post.

Does Skype actually route voice traffic through a supernode? I thought it was just routing TCP/UDP requests to open up ports behind a NAT'd router. As well, supernodes may also be handling some authentication traffic. But I'm not 100% familiar with Skype's protocol.

Yes, supernodes are used to negotiate the communication trough NATs, the method's called UDP hole punching: http://www.google.com/search?num=100&hl=en&q=skype+n...

Reply on news.ycombinator.com