1 00:00:00,159 --> 00:00:05,471 [MUSIC]. 2 00:00:05,471 --> 00:00:09,551 Good day viewers, in this segment I'll talk about peer-to-peer content 3 00:00:09,551 --> 00:00:13,852 distribution systems using BitTorrent as an example. 4 00:00:13,852 --> 00:00:17,872 So, peer-to-peer content delivery is interesting to us, because it provides an 5 00:00:17,872 --> 00:00:22,195 alternative to client server content delivery using CDNs. 6 00:00:22,195 --> 00:00:26,500 And, and it's an alternative that runs without dedicated infrastructure. 7 00:00:26,500 --> 00:00:30,130 That's what makes it interesting. We're going to look at these systems just 8 00:00:30,130 --> 00:00:34,534 to understand a little bit about how they work using BitTorrent as an example. 9 00:00:34,534 --> 00:00:37,584 And I'm really just going to give you just a quick overview of these systems, 10 00:00:37,584 --> 00:00:43,331 so that you see what they look like. So, here's a little bit of context for 11 00:00:43,331 --> 00:00:47,684 where peer-to-peer systems came from. We've seen delivery, where traditional 12 00:00:47,684 --> 00:00:50,811 client-server methods, which was scaled up over time to become 13 00:00:50,811 --> 00:00:55,570 content-distribution networks. These systems are efficient, and they're 14 00:00:55,570 --> 00:00:58,874 able to scale up delivery for very popular content, and they work quite 15 00:00:58,874 --> 00:01:02,305 effectively. They are also quite reliable and that's 16 00:01:02,305 --> 00:01:06,410 because there are managed infrastructure. People pay for content deliveries so you 17 00:01:06,410 --> 00:01:09,880 can hire people to carefully administer all of these systems. 18 00:01:09,880 --> 00:01:12,825 And run them well and make sure they provide good service. 19 00:01:12,825 --> 00:01:18,386 But traditional models, the CDN systems have some disadvantages too and the two 20 00:01:18,386 --> 00:01:23,117 real disadvantages that come to mind first, the need for a dedicated 21 00:01:23,117 --> 00:01:28,386 infrastructure. It has to be provisioned and carefully 22 00:01:28,386 --> 00:01:30,790 managed. All of that costs money, so we're paying 23 00:01:30,790 --> 00:01:34,860 maybe quite a lot, maybe more than we need to for this content delivery. 24 00:01:34,860 --> 00:01:39,490 And second, there is an element of centralized control law oversight. 25 00:01:39,490 --> 00:01:43,648 Whoever has set up the CDN is really in control, and able to make decisions about 26 00:01:43,648 --> 00:01:48,729 what goes on, on that CDN. Now that may present problems for some 27 00:01:48,729 --> 00:01:52,702 people for many reasons. You might not be interested, for 28 00:01:52,702 --> 00:01:55,462 instance, in some large internet companies such as Google and Facebook 29 00:01:55,462 --> 00:01:59,063 delivering all of your content. And having a very good idea of what's 30 00:01:59,063 --> 00:02:01,654 going on. So if you care about privacy, 31 00:02:01,654 --> 00:02:05,408 peer-to-peer might be more interesting. Other people have different examples of 32 00:02:05,408 --> 00:02:07,520 course such as downloading illegal content. 33 00:02:09,630 --> 00:02:13,790 So for peer-to-peer, the real goal of this technology is to provide delivery 34 00:02:13,790 --> 00:02:19,215 without the dedicated infrastructure. All centralized control we've seen in 35 00:02:19,215 --> 00:02:25,060 these other systems while still retaining the advantages that we got from the CDM. 36 00:02:25,060 --> 00:02:27,902 So, we still want to efficient but deliver content at large scale when 37 00:02:27,902 --> 00:02:31,976 there's very popular content. Quite reliable in providing a system 38 00:02:31,976 --> 00:02:35,040 which is really very usable all of the time. 39 00:02:35,040 --> 00:02:37,790 It provides good performance and always up and so forth. 40 00:02:39,450 --> 00:02:42,313 This all sounds great. The key question is, just how on earth 41 00:02:42,313 --> 00:02:45,642 are we going to make this work? And all we have is a pool of clients who 42 00:02:45,642 --> 00:02:48,762 one would exchange it for measuring among themselves with no dedicated 43 00:02:48,762 --> 00:02:52,670 infrastructure to help them. Well, the key idea here is that we're 44 00:02:52,670 --> 00:02:55,910 going to have to have the participants help themselves. 45 00:02:55,910 --> 00:02:58,610 The participants who are called peers in the system, that's why it's P2P 46 00:02:58,610 --> 00:03:03,560 technology. Peer-to-peer took off very shortly after 47 00:03:03,560 --> 00:03:07,848 some of the early systems such as Napster demonstrated a peer-to-peer system in 48 00:03:07,848 --> 00:03:12,202 1999. This was really used for stealing music 49 00:03:12,202 --> 00:03:17,619 or downloading illegally copied music. it's since become defunct because it was 50 00:03:17,619 --> 00:03:19,959 shut down. There was a flurry of activity on 51 00:03:19,959 --> 00:03:23,296 different peer-to-peer systems, including BitTorrent, was one of the fairly early 52 00:03:23,296 --> 00:03:26,658 ones. It was developed around 2001 and it 53 00:03:26,658 --> 00:03:30,725 quickly became very popular. It's probably the dominant peer-to-peer 54 00:03:30,725 --> 00:03:34,276 system today, and it's just a substantial fraction of internet content, so it's 55 00:03:34,276 --> 00:03:40,837 really, see a really wide spread use. Okay, so let's just talk about some 56 00:03:40,837 --> 00:03:44,520 peer-to-peer in general, and then I'll move on to bitTorrent. 57 00:03:44,520 --> 00:03:47,824 So for peer-to-peer in general for delivering content, there are a few 58 00:03:47,824 --> 00:03:51,160 challenges that we're going to have to deal with. 59 00:03:51,160 --> 00:03:54,890 Just because of this very different setting in client-servers. 60 00:03:54,890 --> 00:03:58,040 The real problem we have here is, there are no servers on which we can rely for 61 00:03:58,040 --> 00:04:01,309 anything. Uh-oh, so this means that all other 62 00:04:01,309 --> 00:04:05,731 communication to deliver the content is going to be between the participants or 63 00:04:05,731 --> 00:04:09,288 peers. This is why it's called a peer-to-peer 64 00:04:09,288 --> 00:04:12,970 system, that's the communication pattern, not client-server. 65 00:04:12,970 --> 00:04:16,876 So all of these peers are going to have to organize themselves, somehow, into 66 00:04:16,876 --> 00:04:22,160 coherent architecture to accomplish the task of content delivery. 67 00:04:22,160 --> 00:04:25,100 Well, you might be able to work out how to do this if you just have a group of 68 00:04:25,100 --> 00:04:29,165 friends that are small scale. But we would like these peer P systems to 69 00:04:29,165 --> 00:04:32,350 scale up to large scale, hundreds of thousands of clients working together to 70 00:04:32,350 --> 00:04:36,080 distribute content amongst themselves, for example. 71 00:04:36,080 --> 00:04:41,120 At this scale, there are a few problems. Then I'll just go through them here. 72 00:04:41,120 --> 00:04:45,790 The first problem is that off limited capabilities of the participants. 73 00:04:45,790 --> 00:04:49,054 So, all of the participants, the peers are just you know, they're clients that 74 00:04:49,054 --> 00:04:54,160 want to download the content, too. We don't have the luxury of provisioning 75 00:04:54,160 --> 00:04:58,645 highly powerful high end servers that can distribute content to many clients or in 76 00:04:58,645 --> 00:05:03,690 clients. Everyone's just a client in some sense. 77 00:05:03,690 --> 00:05:06,825 Well, they may range in their capabilities, we don't have the luxury 78 00:05:06,825 --> 00:05:10,490 of, of deciding exactly who's going to be how powerful. 79 00:05:10,490 --> 00:05:14,186 So, the question then is, how can one peer, which might not be a particularly 80 00:05:14,186 --> 00:05:17,714 high-end peer, be able to deliver its content to all of the other peers in the 81 00:05:17,714 --> 00:05:22,030 system if they want to download that content? 82 00:05:23,110 --> 00:05:25,769 That's one issue. Another issue is that of participation 83 00:05:25,769 --> 00:05:29,713 incentives and peers are really helping one another, at least that's what we want 84 00:05:29,713 --> 00:05:32,970 to happen. But if as a peer all you really want to 85 00:05:32,970 --> 00:05:35,970 do is download the content, you know, you want to get a copy of this file we 86 00:05:35,970 --> 00:05:40,434 haven't really yet worked out. Why it is that you're going to bother 87 00:05:40,434 --> 00:05:43,634 sending copies of that file to everyone else to help them out since that's not 88 00:05:43,634 --> 00:05:49,076 what you're really trying to do. And the third issue is that of 89 00:05:49,076 --> 00:05:52,505 decentralization. With the managed server infrastructure, 90 00:05:52,505 --> 00:05:56,040 you know who to contact to find out where the file is. 91 00:05:56,040 --> 00:05:59,410 With a peer-to-peer system, the set of peers is changing over time. 92 00:05:59,410 --> 00:06:02,308 It's just whoever wants to be involved in the download and spreading of this 93 00:06:02,308 --> 00:06:06,205 content right now. So, it's not clear in this system how you 94 00:06:06,205 --> 00:06:11,680 would go about finding out who to contact to get a copy of the content. 95 00:06:11,680 --> 00:06:15,190 So, we would like systems that work well when they're very decentralized. 96 00:06:15,190 --> 00:06:19,690 No one's in charge and all the members of this system constantly changing. 97 00:06:19,690 --> 00:06:23,810 So, these are actually three substantial challenges to deal with. 98 00:06:23,810 --> 00:06:26,490 Well, there are a few, there are, there's a rough lean approach which we could use 99 00:06:26,490 --> 00:06:30,260 to deal with each one. [COUGH] So, to overcome the limited 100 00:06:30,260 --> 00:06:35,012 capabilities of an individual peer, we can still enable one node to distribute 101 00:06:35,012 --> 00:06:42,070 its content to all of the other peers by simply making a distribution tree. 102 00:06:42,070 --> 00:06:46,670 This is like multicasting, a really sort of like what a CDN is doing over time. 103 00:06:46,670 --> 00:06:50,310 So, if you just imagine this source node here has content and everyone wants to 104 00:06:50,310 --> 00:06:53,651 get. Well, we could get to all nodes by having 105 00:06:53,651 --> 00:06:57,698 the source send only two copies to these two nodes. 106 00:06:57,698 --> 00:07:02,120 And these two nodes each would only send two copies, and they're actually, I've 107 00:07:02,120 --> 00:07:08,420 reached all the nodes in the network. Let me just clean that up, here it is. 108 00:07:08,420 --> 00:07:11,368 You can see by following the distribution pattern no node did a lot of work, yet we 109 00:07:11,368 --> 00:07:15,853 got the file to everyone in the system. Of course, with this tree, I have a small 110 00:07:15,853 --> 00:07:19,818 picture here but we could scale up to a, to a very large number of nodes using 111 00:07:19,818 --> 00:07:25,282 this technique. So, and, and actually we wouldn't have to 112 00:07:25,282 --> 00:07:28,780 go through very many levels to get to a very large system because of the way we 113 00:07:28,780 --> 00:07:34,530 are growing exponentially. So, this kind of transfer is typically 114 00:07:34,530 --> 00:07:40,470 done not instantaneously, but with replicas over time. 115 00:07:40,470 --> 00:07:43,470 So, each of the other nodes that you sending to, like here a replica of the 116 00:07:43,470 --> 00:07:46,970 file, it started off here. A replica is being stored here. 117 00:07:46,970 --> 00:07:50,960 And sometime a little later it'll be transferred to these other two nodes. 118 00:07:50,960 --> 00:07:55,548 Now, this approach means that in some sense the capacity of a peer-to-peer 119 00:07:55,548 --> 00:08:01,209 network is self scaling with its size. If you have more nodes in the system or 120 00:08:01,209 --> 00:08:04,749 you, you have more capacity to distribute the content to all of the nodes 121 00:08:04,749 --> 00:08:09,198 themselves. So it has, it certainly has the potential 122 00:08:09,198 --> 00:08:12,870 to be able to work and deliver content to all of the nodes. 123 00:08:14,800 --> 00:08:18,643 In terms of participation incentives I'll note that the peers here play two 124 00:08:18,643 --> 00:08:23,540 different roles. The pink arrows here like these arrows. 125 00:08:24,850 --> 00:08:28,130 downloads to this particular node that's shaded in gray. 126 00:08:28,130 --> 00:08:31,820 Let's just call this one the peer that we're interested in looking at. 127 00:08:31,820 --> 00:08:35,140 So, this is really the node helping itself. 128 00:08:35,140 --> 00:08:37,635 It's downloading the content that it wanted all of the time. 129 00:08:37,635 --> 00:08:41,055 At the same time, this, because it participates in a peer to peer system, 130 00:08:41,055 --> 00:08:45,080 this node is doing some other work. These are the blue arrows. 131 00:08:45,080 --> 00:08:47,600 Oh, I forgot a pink arrow there doesn't matter. 132 00:08:47,600 --> 00:08:50,966 These blue arrows from the point of view of that same node, were uploads to other 133 00:08:50,966 --> 00:08:54,806 nodes. These uploads don’t help it directly at 134 00:08:54,806 --> 00:08:57,530 all. They’re really doing other nodes a favor. 135 00:08:57,530 --> 00:08:59,905 So, this is the dual roles and why it’s called a peer. 136 00:08:59,905 --> 00:09:03,628 It’s performing, if you like, the roles of both the client and the server. 137 00:09:03,628 --> 00:09:08,668 To encourage participation in a network, we can try and couple the two roles 138 00:09:08,668 --> 00:09:12,890 together. Essentially, we can try and create a 139 00:09:12,890 --> 00:09:18,250 system where nodes are working together so that one of them is sort of saying. 140 00:09:18,250 --> 00:09:22,010 Well, I’ll upload content to you if you upload it to me. 141 00:09:22,010 --> 00:09:25,482 If they both do that, then we can create a system that works that way and will 142 00:09:25,482 --> 00:09:30,420 encourage a bit of cooperation and they'll all satisfy they're goals. 143 00:09:30,420 --> 00:09:33,620 Of course, we just have to come up with a protocol to do that, and that's part of 144 00:09:33,620 --> 00:09:36,626 the trick. And finally, in terms of 145 00:09:36,626 --> 00:09:42,800 de-centralization, nodes even though the set of nodes are changing. 146 00:09:42,800 --> 00:09:46,481 The, the peers in the system need to be able to learn where to get the content. 147 00:09:46,481 --> 00:09:51,449 To solve this problem, we're going to use something called DHTs, distributed hash 148 00:09:51,449 --> 00:09:55,590 tables. Some of you might have heard of DHTs, 149 00:09:55,590 --> 00:10:00,224 maybe not. DHTs started as, an academic line of 150 00:10:00,224 --> 00:10:03,424 work. In the academic community about 2001, 151 00:10:03,424 --> 00:10:08,410 with work on systems called Corn and Ken, you might have heard of. 152 00:10:08,410 --> 00:10:12,025 they turn out to be all of the rage in the networking research community. 153 00:10:12,025 --> 00:10:15,160 There was a huge amount of work on them, there still is. 154 00:10:15,160 --> 00:10:18,480 They're some of the most highly sighted network papers of all time. 155 00:10:18,480 --> 00:10:22,190 Now, I don't really have time to tell you how these new algorithms work if you like 156 00:10:22,190 --> 00:10:25,830 I'm sure you can read about them a little further. 157 00:10:25,830 --> 00:10:29,862 Actually, in the text there's a section on the distributed hash tables in the 158 00:10:29,862 --> 00:10:34,180 same section. As peer-to-peer systems which will 159 00:10:34,180 --> 00:10:37,624 explain how one of them cord works. But what I'm going to tell you now is 160 00:10:37,624 --> 00:10:44,440 simply what a DHT does. So, a DHT is essentially an index. 161 00:10:44,440 --> 00:10:47,430 You can look this up for their name to find out where their peers are. 162 00:10:47,430 --> 00:10:50,390 That's what we want to do. Now it's a distributed index, so the 163 00:10:50,390 --> 00:10:54,704 index is spread across all of the peers. And it's fully decentralized and 164 00:10:54,704 --> 00:10:57,620 efficient. These are the properties we want. 165 00:10:57,620 --> 00:11:00,810 So, what does all of this mean? Well well, I guess I'm repeating this a 166 00:11:00,810 --> 00:11:04,096 little bit, but let me just say, the DHT's an index that's spread across all 167 00:11:04,096 --> 00:11:08,500 of the peers. So, it's distributed. 168 00:11:08,500 --> 00:11:13,659 The index, the provides a mapping from say, the name of a file or something you 169 00:11:13,659 --> 00:11:18,279 want to get to the set of peers who currently have pieces of it that you can 170 00:11:18,279 --> 00:11:24,183 contact. And any peer who's in the system is by 171 00:11:24,183 --> 00:11:29,650 using the DHT algorithms able to look at this index and find out who the peers are 172 00:11:29,650 --> 00:11:37,730 to contact to get a copy of this file. So, an index is a very simple thing. 173 00:11:37,730 --> 00:11:42,190 it turns out to be difficult to do in a way which is fully decentralized. 174 00:11:42,190 --> 00:11:45,860 Even when the peers are changing it is still very efficient and fast to look up. 175 00:11:45,860 --> 00:11:48,280 That's the magic of DHTs, but I won't go into it. 176 00:11:50,050 --> 00:11:54,640 Okay, so let's move on to a BitTorrent to see an example of a peer to peer system. 177 00:11:54,640 --> 00:11:57,392 So, BitTorrent is the main peer-to-peer system that's in use in the internet 178 00:11:57,392 --> 00:12:01,670 today. It was developed starting in 2001 by 179 00:12:01,670 --> 00:12:05,070 Graham Cohen. Here he is shown on the right. 180 00:12:05,070 --> 00:12:07,994 finally, we found someone younger than myself to show you a picture of, that's 181 00:12:07,994 --> 00:12:13,620 great. BitTorrent after it's introduction in 182 00:12:13,620 --> 00:12:19,567 2001 grew very rapidly. And it sends a very large amount of 183 00:12:19,567 --> 00:12:23,386 traffic because it's used to transfer very large files, movies for instance 184 00:12:23,386 --> 00:12:28,970 copies of movies are moved around. As a result today, it constitutes, a 185 00:12:28,970 --> 00:12:32,508 BitTorrent traffic constitutes a significant fraction of all internet 186 00:12:32,508 --> 00:12:37,830 activity, so this is one of the main kinds of traffic on the internet. 187 00:12:37,830 --> 00:12:41,300 And it's used for much legal as well as illegal content. 188 00:12:41,300 --> 00:12:45,020 maybe I should actually say illegal and legal content. 189 00:12:45,020 --> 00:12:48,852 Depending on what you think is mostly transferred with BitTorrent. 190 00:12:48,852 --> 00:12:52,842 So just to switch to some of the, a little closer to the bit torrent 191 00:12:52,842 --> 00:12:56,716 terminology. Data is delivered using these things 192 00:12:56,716 --> 00:12:59,298 called torrents. A torrent you might imagine is just a big 193 00:12:59,298 --> 00:13:02,350 gush of data that's going to give you the file you want. 194 00:13:02,350 --> 00:13:06,839 The BitTorrent system transfers files in pieces, so it's going to break up a file 195 00:13:06,839 --> 00:13:11,320 into pieces. Because by using many pieces in parallel, 196 00:13:11,320 --> 00:13:16,110 this parallelism is what's going to give us a high level of performance. 197 00:13:16,110 --> 00:13:19,386 And lets you get a fast transfer rate even though the individual peers might 198 00:13:19,386 --> 00:13:24,056 not be very fast themselves. It's notable for its treatment of 199 00:13:24,056 --> 00:13:27,402 incentives as part of the protocol. What this really means is that the 200 00:13:27,402 --> 00:13:30,489 protocol does actually consider incentives and try and do something about 201 00:13:30,489 --> 00:13:32,850 it. We'll see that in just a minute. 202 00:13:32,850 --> 00:13:37,830 And in terms of decentralization. Some of the early versions weren't really 203 00:13:37,830 --> 00:13:40,734 decentralized, they used an index which was stored on just another server that 204 00:13:40,734 --> 00:13:45,702 you had to find and know about. But the most recent versions have now 205 00:13:45,702 --> 00:13:50,162 switched to using decentralized indexing technology using a DHT. 206 00:13:51,420 --> 00:13:53,920 Okay, so here's the architecture. Oh, sorry. 207 00:13:53,920 --> 00:13:56,296 Actually, before we get to the architecture, here's the steps to 208 00:13:56,296 --> 00:13:58,996 download a torrent. And then, I'll show you a picture and 209 00:13:58,996 --> 00:14:02,480 we'll go through these steps again. This is what you do. 210 00:14:02,480 --> 00:14:04,832 Step one, start with a torrent description, you're going to know what 211 00:14:04,832 --> 00:14:08,294 you want to download. Maybe someone, your friend, mailed you a 212 00:14:08,294 --> 00:14:12,710 link, or a description of a torrent and says check this out, it's great. 213 00:14:12,710 --> 00:14:16,740 Okay, once we've got this description, step two is to find out the peers you can 214 00:14:16,740 --> 00:14:20,710 contact to get the torrent, download the file. 215 00:14:20,710 --> 00:14:24,678 You can do this here by contacting this special tracker to join the set of live 216 00:14:24,678 --> 00:14:30,550 peers and get the list of other peers that are currently downloading the file. 217 00:14:30,550 --> 00:14:33,730 This list will contain at least a seed peer. 218 00:14:33,730 --> 00:14:36,978 A special node which is initialized through the copy of the content to 219 00:14:36,978 --> 00:14:41,345 bootstrap this whole process. This is the older, centralized way. 220 00:14:41,345 --> 00:14:46,305 Alternatively, you can use this decentralized, fully decentralized DHT 221 00:14:46,305 --> 00:14:49,040 index. Look it up, for the, under the name of 222 00:14:49,040 --> 00:14:53,470 the term and find a set of peers who you can contact to get a copy of this file. 223 00:14:54,850 --> 00:14:58,690 Now you know who to contact the peers, and what you begin to do is start trading 224 00:14:58,690 --> 00:15:02,446 pieces with them. These peers everyone's trying to download 225 00:15:02,446 --> 00:15:05,206 the same file at the same time, so everyone trades or swaps the different 226 00:15:05,206 --> 00:15:08,694 blocks they have. As you contact different peers, they may 227 00:15:08,694 --> 00:15:12,502 have some different blocks and you may have some blocks they want. 228 00:15:12,502 --> 00:15:15,635 So, you begin to trade them until you get the complete file. 229 00:15:15,635 --> 00:15:20,499 And as you do so, the rule, the incentive rule is that you're going to favor peers 230 00:15:20,499 --> 00:15:26,325 that upload to you rapidly. And choke peers that don't upload to you 231 00:15:26,325 --> 00:15:31,655 rapidly, by slowing your upload to them. So, somehow you're providing a little bit 232 00:15:31,655 --> 00:15:34,813 of feedback. People who were doing, who were helping 233 00:15:34,813 --> 00:15:40,170 the system by downloading Tu, well you'll also provide them good service. 234 00:15:40,170 --> 00:15:43,810 People who aren't really helping you going to try and cut off service to them. 235 00:15:43,810 --> 00:15:46,393 That way if they're not helping anyone in the system no one's going to give them 236 00:15:46,393 --> 00:15:49,252 good service. I talked about that a, a little bit 237 00:15:49,252 --> 00:15:54,395 through the steps, but I think they'll become clear when we just look at them. 238 00:15:54,395 --> 00:15:57,650 A little more talk about how they solve the challenges. 239 00:15:57,650 --> 00:16:02,029 So, here is a peer-to-peer system. You can see, I've circled one peer we're 240 00:16:02,029 --> 00:16:05,820 going to look at, but all of these other nodes are peers, too. 241 00:16:05,820 --> 00:16:09,975 Peer, peer, peer. So, all of these peers are trying to 242 00:16:09,975 --> 00:16:15,060 download the file, receive the torrent that is, at the same time. 243 00:16:15,060 --> 00:16:17,290 So that they all have the same goal in mind. 244 00:16:17,290 --> 00:16:21,370 One of them here, the seed peer, already has a, a special copy of the content. 245 00:16:21,370 --> 00:16:23,920 That's to get the system going. Someone has to have the content after 246 00:16:23,920 --> 00:16:27,165 all. Now, in the initial versions of 247 00:16:27,165 --> 00:16:31,455 BitTorrent, the way it works is once you, you have a, you have a, a description of 248 00:16:31,455 --> 00:16:35,556 the torrent. It's metafile that just says what it is, 249 00:16:35,556 --> 00:16:40,157 what it is you're trying to download. In this, there is information about who 250 00:16:40,157 --> 00:16:43,904 the tracker node is. If you then contact the tracker node, 251 00:16:43,904 --> 00:16:47,670 this line here, you can get the list of live peers. 252 00:16:47,670 --> 00:16:51,280 So that will have like, one, two, three, all of these nodes. 253 00:16:53,190 --> 00:16:55,360 It will tell you who they are so, you know, who the peers are. 254 00:16:55,360 --> 00:16:59,290 And then, you proceed with the BotTorrent Protocol. 255 00:16:59,290 --> 00:17:03,448 You contact a randomly chosen set of peers, like maybe these peers, to trade 256 00:17:03,448 --> 00:17:07,330 chunks with them. You'll ask what bits of pieces of the 257 00:17:07,330 --> 00:17:12,546 file they have and then you will try to download those pieces from them. 258 00:17:12,546 --> 00:17:15,770 And they will also be asking you what pieces you have. 259 00:17:15,770 --> 00:17:18,808 And as you're talking to multiple different peers you'll get multiple 260 00:17:18,808 --> 00:17:21,727 different pieces. And you'll be able to trade them with 261 00:17:21,727 --> 00:17:25,690 other pieces and this way everyone will be helping everyone. 262 00:17:25,690 --> 00:17:28,772 And some peers eventually will leave this system when they have the new file, but 263 00:17:28,772 --> 00:17:32,452 other peers will arrive. Especially if many peers are trying to 264 00:17:32,452 --> 00:17:35,386 download at the same time. You know, there'll be a life set of 265 00:17:35,386 --> 00:17:40,742 people available to help everyone else. So, let's just revisit those challenges 266 00:17:40,742 --> 00:17:43,780 and see how it all works. How we've met them. 267 00:17:43,780 --> 00:17:47,608 In terms of dealing with the limited capabilities of an individual peer, we've 268 00:17:47,608 --> 00:17:51,146 dealt with that by dividing the file into pieces that can be downloaded in 269 00:17:51,146 --> 00:17:55,180 parallel. This peer here that we're looking at is 270 00:17:55,180 --> 00:17:59,420 downloading from three different other peers concurrently. 271 00:17:59,420 --> 00:18:02,178 That means if these other three peers are very wimpy. 272 00:18:02,178 --> 00:18:05,402 Well, we're able to get a download from all of them so the rate at which we can 273 00:18:05,402 --> 00:18:09,567 download is the sum of these. It can be quite fast how infrastructure 274 00:18:09,567 --> 00:18:13,730 can be providing us a fairly good rate. Because of this parallelism even though 275 00:18:13,730 --> 00:18:16,750 the capabilities of each individual peer is limited. 276 00:18:16,750 --> 00:18:22,780 In terms of participation in centers, I talked about this technique of favoring 277 00:18:22,780 --> 00:18:28,430 the peers, which are uploading to you. You upload to them. 278 00:18:28,430 --> 00:18:32,470 On the other hand, we choke peers that aren't uploading to us. 279 00:18:32,470 --> 00:18:37,842 So this peer here, this unfortunate peer, is shown as choked by all the other three 280 00:18:37,842 --> 00:18:42,350 peers it's talking to. That's the stops here. 281 00:18:42,350 --> 00:18:46,060 This is the choking. And this other peer must be choking, 282 00:18:46,060 --> 00:18:49,819 because the rotten peer here with the x's. 283 00:18:49,819 --> 00:18:55,784 I'll say rotten, it's not really helping out, or unhelpful might be better. 284 00:18:55,784 --> 00:19:01,960 this unhelpful peer is not providing downloads to all of the other nodes. 285 00:19:01,960 --> 00:19:05,070 And so, they cut it off and they're favoring other P's. 286 00:19:05,070 --> 00:19:09,162 This is going to provide an incentive for everyone who has any content to trade, 287 00:19:09,162 --> 00:19:12,750 actually trade it. If you don't have anything because you're 288 00:19:12,750 --> 00:19:15,978 just beginning other nodes will give you basic level of service. 289 00:19:15,978 --> 00:19:18,582 But your download will take a very long time, unless you're willing to help 290 00:19:18,582 --> 00:19:22,848 everyone out. Okay, and finally in terms of 291 00:19:22,848 --> 00:19:25,630 decentralization. But when we switch to the fully 292 00:19:25,630 --> 00:19:30,630 decentralized model, the DHT index, is spread over all of the peers. 293 00:19:30,630 --> 00:19:34,284 So, a small piece of it you can see here is implemented, as part of every peer, 294 00:19:34,284 --> 00:19:39,862 together they're implementing. This DHT this overall index using the DHT 295 00:19:39,862 --> 00:19:45,251 technology, so any peer that confined another peer can join the system. 296 00:19:45,251 --> 00:19:49,399 And look up this index and will end up carry things different parts of this 297 00:19:49,399 --> 00:19:53,151 index. So, it's fully decentralized and we made 298 00:19:53,151 --> 00:19:57,231 our goal of knowing how to get around without having a well known central 299 00:19:57,231 --> 00:20:02,743 server that provides information. Okay, so that's a P2P systems and 300 00:20:02,743 --> 00:20:07,377 BitTorrent is an example. P2P systems are interesting to us as an 301 00:20:07,377 --> 00:20:12,600 alternative to CDN style client server distribution content. 302 00:20:12,600 --> 00:20:16,032 They they certainly have some potential advantages if they can do away with 303 00:20:16,032 --> 00:20:19,790 dedicated infostructure and decentralization. 304 00:20:19,790 --> 00:20:22,280 They also have some challenges to meet to do that. 305 00:20:22,280 --> 00:20:25,064 It's probably not as easy with peer-to-peer technology as it is with a 306 00:20:25,064 --> 00:20:27,950 managed, carefully-provisioned infrastructure. 307 00:20:29,160 --> 00:20:32,010 So if we can make it work, we're in good shape. 308 00:20:32,010 --> 00:20:34,634 but you know, there are still some problems to solve to make them work very 309 00:20:34,634 --> 00:20:39,252 effectively. At this stage, today, CDN technologies 310 00:20:39,252 --> 00:20:44,214 are probably more mature. Nonetheless, P2P technologies are coming 311 00:20:44,214 --> 00:20:49,715 along, and DHT technologies, too. And over time, we're finding that they're 312 00:20:49,715 --> 00:20:54,020 more widely used in systems. And these are the technologies that are 313 00:20:54,020 --> 00:20:57,350 adopted in different kinds of content distribution systems. 314 00:20:57,350 --> 00:21:01,839 Actually, P2P and DHT style technologies are used as part of the Skype system and 315 00:21:01,839 --> 00:21:07,244 its part of Amazon's system Amazon Cloud Computing Services. 316 00:21:07,244 --> 00:21:11,398 So, I think in the future what you can expect is more and more hybrid systems 317 00:21:11,398 --> 00:21:16,393 which may be used on peer-to-peer style techniques along. 318 00:21:16,393 --> 00:21:21,016 With a carefully producing infrastructure for example rather than having solely 319 00:21:21,016 --> 00:21:24,390 owned declines to it themselves, okay?