[p2p-hackers] UDP hole punching performance

[Matthew Kaufman]

Alex Pankratov:
> Do you make this comparison with two Skype clients residing 
> behind two different NAT devices ?

I don't remember those details, see below.
 
> AFAIK in this case Skype won't be doing hole punching and
> be relaying the traffic through a third node instead. It
> will also rate-limit the transfer to a max of 5KB/sec (?).

In both cases, I checked with tcpdump, and UDP was the only protocol in use
and the traffic was going directly between the two computers involved. My
experience is that the throughput scales linearly with latency in a way that
would be explained if they don't use a sliding window at all, or certainly
not a big enough one. Who knows... Might be rate-limited too - the point is
that there are UDP applications that are slower than doing a file transfer
over TCP (FTP, HTTP), and Skype is an example.

Matthew