Post a Comment On: Quinthar

The weak link in BitTorrent from a piracy perspective has always been the torrent sites.  They're the last centralized holdout vulnerable to attack from copyright enforcers (though those attacks have so far been futile).  Regardless, that vulnerability seems to be on the demise with the latest release of Tribler, which includes totally decentralized tracking ability.

I've long said copyright's days are numbered, and tools like this just make that number smaller and smaller.  Sure, copyright will still be enforceable on major customers like movie studios, satellite radio services, and other entities with a large financial and physical presence -- large enough to be worth defending, and worth attacking.

But Joe Plumber will be given an increasingly free hand to ignore copyright with impunity.  Whether that's morally right or wrong isn't the issue.  It's simply true, and more true every day.


In other news, I'm particularly interested in learning more about Tribler's "Give-To-Get" algorithm.  (The website is slammed right now, so I'll have to check it out later.)  I'm hoping/assuming it takes a less paranoid stand than the standard "tit-for-tat" algorithm BitTorrent employs, recognizing that the universe doesn't reset at the end of each download.

In short, if we share data via tit-for-tat, I only give you data if you also immediately give me data.  If you don't have any data for me, or if you give it to me slowly, then I'll withhold my data from you.  In a sense, data is like currency.

This is a brilliant model that allows for the protocol to succeed in scarce network conditions with different implementations: it protects each user from wasting data on users who don't respond in kind.

But it also makes downloads go unnecessarily slow in an abundant network situations because you can only download (on average) as fast as you can upload.  And because uploading is generally constrained to about a quarter your download speed, that means you can only generally download about 25% as fast as you could otherwise.

Now, BitTorrent gets around this with "seeds", who volunteer data without asking anything in return.  With enough seeds, anybody can download at full speed.  But seeds undermine the whole notion of tit-for-tat.

Indeed, the easy availability of seeds suggests that the whole assumption of tit-for-tat -- the scarce network environment -- is wrong.  Somehow, lots of users are more than willing to give away their bandwidth for free, without any obligation to do so.

Now, there are clever design decisions that encourage this: most torrent clients automatically begin seeding once you finish your download, some tracker sites monitor "seeding ratios" (the ratio of data uploaded to data downloaded), etc.

But the point is: despite there being no technical requirement for people to seed, people still do so, in huge numbers, and don't care.

Which brings me back to the original point: if this is "true" about the universe, then tit-for-tat is non-optimal.  It's like wearing a stillsuit in a rainforest.

So the question is: what *is* optimal.  And the answer is: upload when it's *cheap*, not when it's expensive.  Let me explain that:

Tit-for-tat makes you upload at the same time you download.  But the act of uploading actually makes you download slower.  Even worse, because downloading fast requires uploading fast, then the faster you download the more download capacity you're spending on uploading.  The upshot is even were it not for the asymmetric upload/download bandwidth ratio, tit-for-tat makes you upload when bandwidth is the most expensive.  Tit-for-tat takes a scarce bandwidth environment and makes it *worse*.

The alternative is to wait until the download is done and the upload later, when your network is idle (such as when you are watching the thing you just downloaded).  This way when you download, you download as fast as possible without wasting time uploading.  And when you upload, do it in a way that minimizes its impact upon the user who is volunteering the bandwidth.

Doing this, however, requires trust else people will download without ever uploading.  That trust is very difficult to enforce against people's wills.  (Even tit-for-tat suffers from BitThief problems.)

But the very fact that seeders are in such large supply in a tit-for-tat model shows that users are generally willing to donate their bandwidth voluntarily.  As such, even though there's no way to force people to upload, people still do it anyway.  If you make it more convenient to just "do the right thing" than try to fight the system, then people will just go with it and everybody wins.

Hopefully Tribler does this.  Once the website comes back, we'll see.