Browse //
Bram Cohen - The Creation of BitTorrent, DRM, Advertising
San Francisco, April 2007 
HDV Version (2.09 GB)Ogg Version (18.2 MB)Subtitles
(you need a BitTorrent Client to download the Videos)
Creative Commons License
Creative Commons Attribution Share-Alike

Bram Cohen is the author of the BitTorrent protocol, the preferred method for the distribution of files using distributed networks. Here he outlines the background to BT's development and the engineering problems it was designed to resolve. Thereafter Bram discusses the reasons for the failure of pay-for-access models on the net, and explains why advertising has become the preferred form of monetization of online 'content'.

Play from beginning
00:00:00 Interview with Bram Cohen, developer of BitTorrent
00:00:05 Well there's is the universe which did happen and the universe which should have happened,
00:00:10 and pretty much everyone seems to agree that in the universe which should have happened,
00:00:21 BitTorrent was written by some venture capital backed company
00:00:33 which then got sued and kinda sort of acquired after getting sued into oblivion
00:00:42 and the CEO having a scare of almost having been
00:00:44 thrown in prison, or actually being thrown in prison.
00:00:48 Of course in the universe which actually happened,
00:00:53 - which doesn't really make any sense so it isn't really worth talking about.
00:00:57 In the universe which actually happened, BitTorrent was written by some guy
00:01:03 in his living room who was living off of credit cards and became explosively huge,
00:01:10 prior to taking any investment whatsoever.
00:01:13 And then the person who wrote it somehow wound up working with Hollywood.
00:01:22 And running a very legal business without any legally scary situations.
00:01:31 When you have television or radio and by that I mean television over the airwaves,
00:01:40 It's kinda like someone is screaming really really loudly and everyone else can listen.
00:01:46 And so anyone who wants to can tune in
00:01:49 Now this is not very efficient from an energy standpoint
00:01:53 and it involves some very very loud screaming.
00:01:57 It is highly effective at broadcasting signals to space aliens.
00:02:02 If there were any aliens hanging out around earth in the 1900s,
00:02:10 they'd have had a very very easy time watching television.
00:02:13 And getting a good look at our culture that way,
00:02:16 because that's where most of the signal was going, was out into space.
00:02:23 The internet doesn't work that way - on the internet
00:02:29 you have all these machines which are connected to the net
00:02:32 and you can send a message to anyone else on the net,
00:02:37 but it goes to just them - there's no broadcast concept there.
00:02:42 So there's this question of how do you make broadcasting work?
00:02:50 Unfortunately broadcasting can be rather difficult in that if you have something that's very popular
00:02:58 and very large and very high quality, it can become very expensive to distribute it.
00:03:06 It becomes expensive to be popular. With broadcast over the airwaves this doesn't happen,
00:03:12 You're screaming loud enough, everybody can hear you - not a problem.
00:03:17 On the internet it's a huge problem so BitTorrent was based
00:03:22 on this very fundamental calculation of well if you're sending out something,
00:03:25 and everyone wants the same thing, they can just send it to each other,
00:03:30 So there's this logistical problem of how to make that happen,
00:03:35 so I figured out how to make it happen.
00:03:37 Well the basic problem is a pretty simple one,
00:03:41 there's plenty of upload capacity out there not being used how do we use it?
00:03:45 So that's a trivial calculation on its own the problem is,
00:03:52 these are what you call low quality resources.
00:03:54 They're peers, they're untrusted, they are of unknown potential transfer rate
00:04:01 they're not of terribly good transfer rate to begin with and they're not very coordinated.
00:04:09 This isn't really so much of a problem of make something that works,
00:04:13 so much as make something that works reliably,
00:04:16 that can handle the fact that peers just sort of disappear and never come back again.
00:04:22 When I started working on it there was a bunch of people working on very much the same thing.
00:04:30 I decided on an approach that was actually much, much more ambitious
00:04:35 than a lot of the things that had been successful up until that time.
00:04:39 In that you when you're distributing something on the internet,
00:04:45 if you're doing it via HTTP, you kinda don't want everyone
00:04:50 to come and download the same thing at the same time.
00:04:52 You want nice small things that are distributed around when people download them.
00:04:57 and this is good for making it so you don't have too much load on the one central server,
00:05:03 and I went and did this calculation and figured, well no, I want to do the exact opposite thing.
00:05:10 I want everyone downloading the same thing at the same time.
00:05:13 Because if, and this at the time this was a pretty big if, if you can get a handle
00:05:18 on all the difficult logistical problems of making it actually happen,
00:05:23 then you can make it so that the initial place only has to upload one copy
00:05:30 of the whole thing and everything else will be distributed between peers
00:05:35 and you actually get maximum efficiency in the very situation you were trying to avoid
00:05:41 when you were doing everything via HTTP.
00:05:43 So in that sense I was being rather ambitious although other people were working on the same problem.
00:05:51 The difference was in terms of approach, that I came up with an architecture
00:05:57 which was designed first around reliability and efficiency
00:06:02 - in fact only reliability and efficiency- it's an utterly bizarre architecture
00:06:09 unless you consider it from the point of view of ok...
00:06:12 first thing, let's just assume that peers are unreliable
00:06:16 that we don't know what transfer rates are
00:06:19 that peers are untrusted and tend to go away
00:06:22 and then, out of what's left, how do we make something work
00:06:28 other people were trying the tree-based architectures, which proved not to work
00:06:34 but the reasons why are very much centered around reliability,
00:06:40 and are not obvious unless you've done some networking and know that just...
00:06:44 peers go away and never come back
00:06:46 DRM has a lot of political momentum right now
00:06:50 it's just like if you're putting content up online you have to have DRM and...
00:06:57 whether this is psychological, whether technicals are saying it,
00:07:03 whether lawyers are saying it, whether just collectively everybody feels
00:07:06 that somebody must be saying it, so you have to speak in one voice demanding it...
00:07:13 ... is a little unclear, and varies from place to place
00:07:17 When you go to the movie theatre you pay
00:07:19 when you get a DVD you pay, when you rent a DVD you pay
00:07:26 there are a few things going on
00:07:28 one of the big things is what people associate with their home experience is watching television
00:07:31 You turn on the television and then you watch,
00:07:34 and people are rather disinclined
00:07:38 in some ways
00:07:40 paying for something to watch at home
00:07:42 they want to just have ads and watch it
00:07:45 by analogy with television
00:07:48 another thing that happens is that people are leery
00:07:50 frequently leery of paying for anything online
00:07:54 just putting in a credit card number,
00:07:56 because credit cards are so fundamentally insecure
00:07:59 makes people very nervous, they don't want to do that,
00:08:02 and it's an annoying process entering in your credit card number.
00:08:05 Now the ridiculous insecurity of credit card numbers has a lot to do with
00:08:10 with the ridiculous way banks work in the United States
00:08:13 where things are big and bloated and poorly done technologically
00:08:18 and there's very little if any incentive to fix it.
00:08:22 I would say that people are a little used to now,
00:08:26 when they download videos from the net, not paying for it,
00:08:31 it's just what they've been doing,
00:08:33 the paying for it model just hasn't been there!
00:08:38 So people just habitually aren't very used to paying for things.
00:08:44 At a certain point when you get down to what's termed level of cost
00:08:49 the actual price being charged is de minimis,
00:08:54 whether that's a dollar, or ten cents or one cent.
00:08:59 it's a little hard to say.
00:09:03 But at some point the actual price paid ceases to be a concern.
00:09:09 and that's for extremely popular content
00:09:13 isn't very de minimis
00:09:15 when you multiply it over the number of people who are paying it
00:09:22 that's effectively what advertising online winds up being
00:09:27 the monetization on it is small.
00:09:33 Generally speaking a penny per impression,
00:09:36 but winds up adding up in the end.
00:09:39 The question there is what is the form of that monetization?
00:09:43 Is it via advertising, which people are implicitly paying for in some way,
00:09:48 Is it explicitly paying where there is this usability issue of making the payment,
00:09:57 and the concern about fraudulent charges happening when this payment is happening
00:10:02 concerns about incentivizing spam, bladibladibla...
00:10:08 So at a certain level
00:10:10 there's the cost of the distribution and there's the value gotten from the thing
00:10:14 and if the cost
00:10:16 is some very small fraction of the value gotten from the thing
00:10:20 people cease to pay any attention to it whatsoever!
00:10:26 And the question then becomes what is the form of the payment.
00:10:31 Advertising is certainly a compelling model
00:10:34 in that it's very very simple,
00:10:39 Whenever you have payment going through a distributor
00:10:43 there's the whole issue of making the payment happen
00:10:46 both in terms of authorizing charges and redistributing the money
00:10:50 and it has to somehow hook into the payment system,
00:10:53 and advertising is somewhat inefficient
00:10:57 in that it implicitly hooks into the payment system
00:11:00 via some complicated route of people watching the advertising
00:11:03 doing something eventually, somewhere out there,
00:11:06 But it's much much simpler in general,
00:11:09 a more straightforward way of doing things.