_____  

From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of David Barrett
Sent: Saturday, September 01, 2007 12:24 PM
To: 'theory and practice of decentralized computer networks'
Subject: Re: [p2p-hackers] Effective TCP and UDP NAT Traversal (no relaying)



There was recently (July '07) a big discussion on this topic on the BEHAVE
mailing list.  I posted some of my results here:

 

http://www1.ietf.org/mail-archive/web/behave/current/msg02496.html

 

Basically, I can get about direct 90% peer connectivity using UDP.  Rumor is
Alex can get something like 99% (Alex - care to share any detailed
numbers?).   

I do care, but, sorry, can't share. It is not 99% though, I can tell you
this much. At some point I made a comment on this list, that 97-98% should
be achievable. Since then we added TCP NAT traversal and a handful of UDP
traversal tricks. On other hand Hamachi's userbase profile has changed.
Specifically, the percentage of home users (behind dumb routers) declined in
favor of people behind proxies and other funny devices (e.g. load
balancers). We got more people connecting from far places (hop count and
latency-wise) as well. So the number works out to be about the same as it
was before. With just the home users I suspect we would've been in 95-97%
range, though I might be off.

The IETF is pushing a protocol named ICE, but nobody knows how well it works
(Adam - have you come up with any numbers yet?).

 

Overall, it's much harder than it looks, even in the "simple" cases.  The
basic algorithm is:

 

1) Figure out the IP address of each node's NAT

2) Share each node's pair of (LAN,NAT) IPs with the other node via some
central server

3) Try to connect over both the LAN and NAT addresses.

4) Apply a lot of voodoo tricks

5) Oftentimes it works

 

Everybody seems to use a variation on this algorithm, though Alex recently
made some suspicious comments suggesting otherwise.

The way you described it, my comment still stands. I cannot elaborate more,
but it is possible to deduce what Hamachi does differently.

 

Alex

_______________________________________________
p2p-hackers mailing list
[email protected]
http://lists.zooko.com/mailman/listinfo/p2p-hackers

Reply via email to