August 12, 2009 at 15:55Simplicity in design. Episode 3. Fighting Freeriding
Simplicity does not precede complexity, but follows it. / Alan J. Perlis /
This story began in the early 2000s, when everyone realized with the Gnutella example that freeriding is killing a peer-to-peer network - a mass of parasitic users who only download but don’t give out. A lot of articles were written about trust and traffic accounting in peer-to-peer networks. For example, The EigenTrust Algorithm for Reputation Management in P2P Networks, one of the most popular and cited articles in this area, where Garcia-Molina himself was a co-author, was presented at WWW'2003. As they say, everything is in class "A". The article implied that peers would jointly calculate a PageRank-like metric, passing each other vectors of size N, where N is the number of participants. I know the unfortunate who even today continue to deal with such metrics, with O (N) data per node and complex output rules. And in real life, BitTorrent “won”, in which some reputational element of course was - the tit-for-tat algorithm, “you to me - I to you”, but all this was much simpler than reputation-based systems based on DHT and other things like that. It's just that each feast tries to answer in return and send traffic (interesting in the details) to the three peers that downloaded it the most in the last 20 seconds and to another, randomly selected. Of course, it should be recognized thatsimple reputation systems are also in use. Pofixes, as I call them, or gags. Sharing ratio enforcement is on the same torrents.ru, tvtorrents.com and where only it is not. But from my point of view, this proves the following: traffic will be the sea, if only there is some, even the most symbolic reward. If there is no reward at all, then there will be traffic. But not the sea. Like on thepiratebay.com.
In what direction is this area developing now? Apparently, in the direction of simplification. The fact is that the tit-for-tat algorithm required dividing the file into small pieces and then exchanging them using the rarest-first algorithm (which was also written about a lot of funny articles). Plus more "ratings." All this is quite complicated. But what if we remove the root cause of the conflict - the motive that causes users to stop sitting? More specifically, the problem is that surfing the Internet with BitTorrent running becomes very inconvenient. This is due to the fact that all 50 connections opened by BitTorrent get their share of traffic on an equal footing with those timid 4 TCP connections that the browser opens. As a result, the browser slows down. And if you abandon TCP and use a more timid transport loading P2P for P2P, when is it free and quickly inferior to bursts of TCP traffic? Many read the news aboutsuch transport (uTP) in µTorrent. Its development at BitTorrent Inc is carried out by Stanislav Shalunov. If uTP works well, you may be able to redeem users in free traffic.
I’ll add from myself. Then it will become possible to score on both tit-for-tat and rarest-first, and finally, download files in order and watch a movie “by click”, as on YouTube. Perhaps this will require a couple more prefixes, but fundamentally the equation is as follows: users can not download movies 24 hours a day. And they can sit 24 hours a day. Unless it creates inconvenience.
This story began in the early 2000s, when everyone realized with the Gnutella example that freeriding is killing a peer-to-peer network - a mass of parasitic users who only download but don’t give out. A lot of articles were written about trust and traffic accounting in peer-to-peer networks. For example, The EigenTrust Algorithm for Reputation Management in P2P Networks, one of the most popular and cited articles in this area, where Garcia-Molina himself was a co-author, was presented at WWW'2003. As they say, everything is in class "A". The article implied that peers would jointly calculate a PageRank-like metric, passing each other vectors of size N, where N is the number of participants. I know the unfortunate who even today continue to deal with such metrics, with O (N) data per node and complex output rules. And in real life, BitTorrent “won”, in which some reputational element of course was - the tit-for-tat algorithm, “you to me - I to you”, but all this was much simpler than reputation-based systems based on DHT and other things like that. It's just that each feast tries to answer in return and send traffic (interesting in the details) to the three peers that downloaded it the most in the last 20 seconds and to another, randomly selected. Of course, it should be recognized thatsimple reputation systems are also in use. Pofixes, as I call them, or gags. Sharing ratio enforcement is on the same torrents.ru, tvtorrents.com and where only it is not. But from my point of view, this proves the following: traffic will be the sea, if only there is some, even the most symbolic reward. If there is no reward at all, then there will be traffic. But not the sea. Like on thepiratebay.com.
In what direction is this area developing now? Apparently, in the direction of simplification. The fact is that the tit-for-tat algorithm required dividing the file into small pieces and then exchanging them using the rarest-first algorithm (which was also written about a lot of funny articles). Plus more "ratings." All this is quite complicated. But what if we remove the root cause of the conflict - the motive that causes users to stop sitting? More specifically, the problem is that surfing the Internet with BitTorrent running becomes very inconvenient. This is due to the fact that all 50 connections opened by BitTorrent get their share of traffic on an equal footing with those timid 4 TCP connections that the browser opens. As a result, the browser slows down. And if you abandon TCP and use a more timid transport loading P2P for P2P, when is it free and quickly inferior to bursts of TCP traffic? Many read the news aboutsuch transport (uTP) in µTorrent. Its development at BitTorrent Inc is carried out by Stanislav Shalunov. If uTP works well, you may be able to redeem users in free traffic.
I’ll add from myself. Then it will become possible to score on both tit-for-tat and rarest-first, and finally, download files in order and watch a movie “by click”, as on YouTube. Perhaps this will require a couple more prefixes, but fundamentally the equation is as follows: users can not download movies 24 hours a day. And they can sit 24 hours a day. Unless it creates inconvenience.