Hacker News Reader

Comment by JoeAltmaier | Hacker News Reader

"Tried to be p2p".

Most of the time, our routers/APs defeated that. Skype would 'fall back' to using a server, which I believe it did 90% of the time.

View on HN

I'm not convinced. It would use every trick in the book to punch holes; stun, upnp, sip, it's own protocol on top of udp, it would even send data over port 80 or 443. And if that failed Skype would use a third Skype client to open a port, or even route the calls through the third party.

I always thought this was the major reason why Skype was so successful. It would penetrate almost any workplace firewall and "just worked" for most people. Shady, yes, but it worked.

And all those tricks either worked for you (your endpoint), or they didn't. No middle ground.

I worked 8 years in this environment (collaboration tools) and it was our holy grail to get p2p connections. We tried all those tricks and more. It worked a fraction of the time (at that time, 8 years ago).

I guess most holes got patched. I worked with the same things around 2007.

Are you sure? I don't know the specifics on how Skype used to work, but I'd assume that they were just using intermediary server to establish connection via NAT hole punching, and then proceed to communicate peer to peer.

That's what doesn't work (NAT hole) depending on the endpoint router/firewall. More often than we think.

It all depends on the filter used in the NAT logic in the router. They can depend on any of IP/PORT/Protocol/MAC of the source, and indeed that hadn't been commercially standardize for a long time.

The trouble comes in 'handing off' the discovered address from the TURN server to the peer. The endpoint may or may not accept it, coming from a 'different guy' (the other endpoint now, instead of the TURN server).

I remember the old (pre-MS) skype being quite good at punching holes in firewall by using a number of tricks.

Maybe it would not beat a determined sysadmin, but in many cases it was more than enough

It was P2P sometimes, like if you were running in a university or research institution.

When skype got installed in there, it took over the university network and quickly became master node for the whole geographic area.

For historic reasons, these organizations used to have many computers on public IP (they got large IP block when internet was created very long ago) and had amazing network connectivity.

Where did you get this "90% of the time" statistic? "I believe"?

Worked in the industry - wrote collaboration tools including media nodes, media endpoints, proxies and servers. Tried all the tricks in the book. Started out by running snoop tools on Skype to see what they tried, went from there.

Lots of endpoints in the world flatly didn't allow p2p. Some whole nations disallowed entire protocols (Ireland, UDP). Some routers dropped every other UDP packet (ATT). Some reordered UDP (swapped packets in pairs). It was a total mess.

Anyway we ended up with something ok, and a high-performance media node to handle most of the traffic which wasn't p2p.

Reply on news.ycombinator.com