On Jan 29, 2011, at 12:12 AM, Alex Pankratov wrote: > > >> -----Original Message----- >> From: p2p-hackers-boun...@lists.zooko.com >> [mailto:p2p-hackers-boun...@lists.zooko.com] On Behalf Of >> Roberto Roverso >> Sent: January 24, 2011 1:04 AM >> To: p2p-hackers@lists.zooko.com >> Subject: [p2p-hackers] NAT traversal state of the art >> >> I'm the author of the paper mentioned in this thread. >> >> I read the patent application and, however the idea of NAT >> type discovery is present and common to our paper (as to >> others: the STUN RFC above all), the patent describes just >> few types (6) of NAT behavior. In our paper, we define as >> many as (27). > > And yet you mention in paper that only 50% of combinations > actually occur in practice :) >
You are totally correct, at the time we published our results we had only observed 50% of the possible combinations. As of now however, the stats show that most of the types are indeed present in our test network ( 24 over 27). > I arrived at my classification working backwards compared to > you. I was sitting and sifting through the NAT discovery logs, > trying to group results into a set of classes. After few > iterations in a course of several months, I had five classes, > two of which had two additional (boolean) parameters, and the > traversal logic was built around this set. I suspect that if > the impossible 50% is eliminated from the table and adjacent > entries are collapsed, it might end up being the same set of > classes. > I'm not sure that is the case, we did not start our NAT classification from scratch, a good deal of the types explained in the paper (21 over 27) have been observed and classified by others in the field: http://www.rfc-editor.org/rfc/rfc4787.txt http://tools.ietf.org/id/draft-takeda-symmetric-nat-traversal-00.txt What we did, for the discovery/classification at least, was to bring together bits and pieces of the state of art and integrate it with other types we had observed in our test network. The real contribution of the paper instead lies in demonstrating theoretically and practically which traversal technique should be used in each one of the combinations. >> On top of that, the paper outlines how to carry >> out hole punching in each specific combination of NAT types >> (NAT type A against NAT type B). Although many details are >> omitted in the document due to space constraints, the >> description is probably enough for people to start implement >> state of the art NAT traversal logic in their p2p >> applications. Results from our test network show that the >> connection establishment success rate is very high using this >> model (~ 90%). > > Can I ask how wide spread was your client base? Our client base is made of a couple of thousands hosts located Sweden at the moment, as we haven't deployed our products in other countries yet. It would be interesting to test our solution on a "planetary" scale. I wonder if people would be willing to participate in such experiments such as NAT types data collection and automated testing to improve our NAT traversal library when we will release it (the NAT discovery software will be out in the next days). > When I first > launched Hamachi it was getting over 97% of p2p connectivity, > which frankly was way way more than I ever hoped for. But > then the app started to see adoption outside of North America > and the rate started dropping. There was a clear effect of > (funny enough) Italy and few Asian countries on an average > traversal rate. I wonder if you have noticed anything similar. > > Alex > > _______________________________________________ > p2p-hackers mailing list > p2p-hackers@lists.zooko.com > http://lists.zooko.com/mailman/listinfo/p2p-hackers As you said, connectivity success rate might vary very much from country to country, we are quite confident that good connectivity can be achieved even when difficult-to-traverse NATs are present, since one third of our customers are indeed behind "corporate" NATs, which are known to be very cumbersome to deal with. Italy is a good example of that, most of people having good bandwidth are in the Fastweb network, which is a giant private network with a multi-level NAT infrastructure. (I'm a former customer of that ISP). Regards, Roberto
_______________________________________________ p2p-hackers mailing list p2p-hackers@lists.zooko.com http://lists.zooko.com/mailman/listinfo/p2p-hackers
