1 00:00:05,930 --> 00:00:07,360 So 2 00:00:07,360 --> 00:00:09,130 if I can have your attention 3 00:00:20,860 --> 00:00:22,960 This morning when I talked about CSAIL 4 00:00:22,960 --> 00:00:26,930 I¡­I said a couple things about it One of the-- one-- one of the things 5 00:00:26,900 --> 00:00:30,560 I said was that they're four groupings within the lab 6 00:00:30,560 --> 00:00:32,930 and one of those four groupings is theory 7 00:00:32,930 --> 00:00:36,700 We have a very large theory group with a large number of faculty 8 00:00:36,700 --> 00:00:42,600 And in some ways theory is the hardest of the four groups to get research funding for 9 00:00:42,600 --> 00:00:45,660 In most of the sort of work we do 10 00:00:45,660 --> 00:00:48,460 we put in proposals and we say we're going to work on this problem 11 00:00:48,460 --> 00:00:49,400 and we're going to solve it 12 00:00:49,400 --> 00:00:51,430 and then we're going to work on this problem and we're going to solve it 13 00:00:51,430 --> 00:00:55,560 And we have a sort of linear chain about how we're going to go about doing something 14 00:00:56,060 --> 00:00:58,230 But in theory the theory guys sort of say 15 00:00:58,230 --> 00:01:02,200 ¡°Well we're going to think about this area and--and we might prove some theorems 16 00:01:02,200 --> 00:01:03,300 but we might not¡± 17 00:01:03,300 --> 00:01:05,700 because if they knew what they were going to do 18 00:01:05,700 --> 00:01:08,330 exactly they would have already done the work 19 00:01:08,330 --> 00:01:13,200 So it's very hard for them to say exactly what they going to do in the future 20 00:01:13,200 --> 00:01:15,630 and it's it's very much exploratory based research 21 00:01:15,630 --> 00:01:18,560 Even if it's you know on in the cryptography area 22 00:01:18,560 --> 00:01:23,630 or in the high dimensional geometry area which has some applications 23 00:01:23,630 --> 00:01:27,600 Exactly what they're going to do is is hard to to say ahead of time 24 00:01:27,600 --> 00:01:28,730 On the other hand 25 00:01:28,730 --> 00:01:32,100 out of the theory group has come a lot of spin off companies 26 00:01:32,100 --> 00:01:33,700 some of which have done rather well 27 00:01:33,700 --> 00:01:37,600 RSA security is one that you you've probably heard of from Rivest 28 00:01:37,600 --> 00:01:41,060 and his colleagues with the original encryption algorithms 29 00:01:41,060 --> 00:01:46,100 More recent ones you'll see some of them at the booth this afternoon 30 00:01:46,100 --> 00:01:46,930 in this room 31 00:01:46,930 --> 00:01:50,000 This is by the way is the room where we're going to have the spin off companies 32 00:01:50,000 --> 00:01:51,260 Peppercoin 33 00:01:51,260 --> 00:01:53,500 And another one of those companies is Akamai 34 00:01:53,500 --> 00:01:56,830 And Akamai was founded by Professor Tom Leighton 35 00:01:56,830 --> 00:01:59,500 And this brings me to the second thing that I mentioned this morning 36 00:01:59,500 --> 00:02:04,260 that CSAIL is an interdisciplinary lab with faculty from many different departments 37 00:02:04,260 --> 00:02:07,560 Tom is in the mathematics department here at MIT 38 00:02:07,560 --> 00:02:11,600 He received his PhD in 1981 from the mathematics department 39 00:02:11,600 --> 00:02:14,130 He's a faculty member in the mathematics department 40 00:02:14,130 --> 00:02:18,200 And in the theory group a large number of about half the faculty 41 00:02:18,200 --> 00:02:19,900 are actually from the math department 42 00:02:19,900 --> 00:02:21,500 and half are from the computer science department 43 00:02:21,500 --> 00:02:24,460 So it really is a a group that spans departments 44 00:02:24,460 --> 00:02:28,460 and even schools-- engineering and science here at MIT 45 00:02:28,460 --> 00:02:31,960 So Tom has been on the faculty since 1981 46 00:02:31,960 --> 00:02:34,000 He a few years back 47 00:02:34,000 --> 00:02:35,960 he started Akamai 48 00:02:35,960 --> 00:02:39,460 and many of you probably know a lot of the Akamai story 49 00:02:39,460 --> 00:02:42,830 but we're going to hear it first hand from Tom today 50 00:02:42,830 --> 00:02:45,960 Tom welcome Thank you 51 00:02:50,260 --> 00:02:51,630 Thanks very much 52 00:02:51,630 --> 00:02:54,560 If you have questions during the talk 53 00:02:54,560 --> 00:02:56,830 stand up or raise your hand or just shout out 54 00:02:56,830 --> 00:02:58,460 I'd be happy to try and address those 55 00:02:58,460 --> 00:03:02,100 and then maybe we'll have time at the end for for some final questions 56 00:03:02,800 --> 00:03:07,200 I'm going to talk about the history of Akamai focusing on the early days 57 00:03:07,200 --> 00:03:12,360 How we started doing academic research at the then lab for computer science 58 00:03:12,360 --> 00:03:15,030 what the problem area was and how we got going 59 00:03:15,030 --> 00:03:16,530 and then how we took it 60 00:03:16,530 --> 00:03:20,130 you know what caused us to take it from the lab into a company 61 00:03:20,130 --> 00:03:23,160 What was the m- what were the mechanisms there that were successful 62 00:03:23,160 --> 00:03:25,300 And then I'll just say just a little bit about 63 00:03:25,300 --> 00:03:27,100 you know the crazy times the bubble 64 00:03:27,100 --> 00:03:28,360 the bubble bursting and then 65 00:03:28,360 --> 00:03:30,760 you know going forward from there 66 00:03:31,660 --> 00:03:34,360 So we're going to start with the academic research 67 00:03:34,830 --> 00:03:37,430 None of the people who were involved in Akamai at the beginning 68 00:03:37,430 --> 00:03:41,300 ever had any idea they were going to start a company of any kind 69 00:03:41,300 --> 00:03:43,730 This was purely a research project 70 00:03:43,730 --> 00:03:47,730 that was motivated by some of the problems with the web 71 00:03:47,730 --> 00:03:51,660 The web at a high level is an amazing structure 72 00:03:51,660 --> 00:03:54,030 It's-- it's revolutionized the way we communicate 73 00:03:54,030 --> 00:03:56,760 and do business and it's very simple 74 00:03:56,760 --> 00:04:00,660 You have people that want to get access to content 75 00:04:00,660 --> 00:04:03,830 the end users at home or in the business or at a university 76 00:04:03,830 --> 00:04:07,530 and people that have content that those people want to get access to 77 00:04:07,530 --> 00:04:09,860 and they get the access through the internet 78 00:04:10,960 --> 00:04:14,660 Now if you look under the covers things get a little bit more complicated 79 00:04:14,660 --> 00:04:16,630 In particular 80 00:04:16,630 --> 00:04:20,560 the path that the bits take from an origin server 81 00:04:20,560 --> 00:04:24,930 or a website to the end user can be long and tortured through the internet 82 00:04:24,930 --> 00:04:27,730 You will typically go through many ISPs 83 00:04:27,730 --> 00:04:31,330 or network providers on the path from the origin to the destination 84 00:04:31,330 --> 00:04:32,630 And sometimes 85 00:04:32,630 --> 00:04:36,330 things can go wrong during that path 86 00:04:36,337 --> 00:04:39,360 And one of the main problems is at the origin infrastructure 87 00:04:39,365 --> 00:04:41,730 that you have a bottleneck right at the very start 88 00:04:41,734 --> 00:04:44,100 which we'll talk a little bit about 89 00:04:44,100 --> 00:04:46,100 And also at the peering points 90 00:04:46,100 --> 00:04:48,430 Either public peering points or private peering points 91 00:04:48,430 --> 00:04:50,860 where the bits go from one network to another 92 00:04:50,860 --> 00:04:52,200 you can get into trouble 93 00:04:52,200 --> 00:04:54,630 They can be lost or misdirected or stolen and 94 00:04:54,630 --> 00:04:56,930 and not arrive at the end destination 95 00:04:56,930 --> 00:04:59,460 And it was the foresight with these problems 96 00:04:59,460 --> 00:05:03,130 that led Tim Burners Lee to walk down the hall 97 00:05:03,130 --> 00:05:08,500 at LCS and come talk to my group which is the algorithms group at LCS 98 00:05:08,500 --> 00:05:10,860 And that's one of the great things about LCS is it's-- 99 00:05:10,860 --> 00:05:12,600 it's a multidisciplinary environment 100 00:05:12,600 --> 00:05:15,100 Today CSAIL is a multidisciplinary environment 101 00:05:15,100 --> 00:05:18,800 You have theoreticians from the math department of all places 102 00:05:18,800 --> 00:05:21,230 sitting more or less side by side with 103 00:05:21,230 --> 00:05:23,260 you know leading practitioners and people who 104 00:05:23,260 --> 00:05:25,400 you know have made major advances like Tim 105 00:05:25,400 --> 00:05:27,000 in founding the web 106 00:05:27,000 --> 00:05:29,160 And Tim knew the web was going to get into trouble 107 00:05:29,160 --> 00:05:30,760 even back in '95 108 00:05:30,760 --> 00:05:33,300 particularly with origin infrastructure problems 109 00:05:33,300 --> 00:05:36,300 that if you have a lot of people that want to get to a particular website 110 00:05:36,300 --> 00:05:38,460 at one time that was going to be a problem 111 00:05:38,460 --> 00:05:40,660 And he said ¡°Hey you guys work on algorithms 112 00:05:40,660 --> 00:05:43,960 and networks for very big problems and you do this theory stuff 113 00:05:43,960 --> 00:05:47,360 Maybe there's something you guys can think of here that would-- would be helpful¡± 114 00:05:47,360 --> 00:05:49,300 And that happened back in '95 115 00:05:49,300 --> 00:05:51,130 and-- and that's what got us started 116 00:05:51,500 --> 00:05:53,930 And let me say a little bit about those two problems 117 00:05:55,270 --> 00:05:56,060 the first being the centralization bottleneck 118 00:05:56,060 --> 00:05:58,400 which is still an issue today 119 00:05:58,400 --> 00:06:00,300 The basic idea is simple: 120 00:06:00,300 --> 00:06:03,900 a centralized website creates an inherent bottleneck 121 00:06:03,900 --> 00:06:06,230 and a great target for attackers 122 00:06:06,230 --> 00:06:09,630 And we've seen this with many key websites over the last several years 123 00:06:09,630 --> 00:06:15,060 from the White House to Microsoft to Sco that it's-- it's a nice target 124 00:06:15,060 --> 00:06:17,230 It's inherently not scalable 125 00:06:17,230 --> 00:06:20,560 As usage of the net grows and the number of bits you sink 126 00:06:20,560 --> 00:06:22,300 through broadband connection grows 127 00:06:22,300 --> 00:06:27,500 it puts a lot of burden on a single origin infrastructure 128 00:06:27,500 --> 00:06:28,500 You know 129 00:06:28,500 --> 00:06:31,060 the-- the web is a-- is a good news bad news story 130 00:06:31,060 --> 00:06:35,500 The good news is it's unprecedented in its ability to support communication 131 00:06:35,500 --> 00:06:39,560 Anybody can put up a web page without a license from the FCC 132 00:06:39,560 --> 00:06:41,660 or without having to buy a radio station 133 00:06:41,660 --> 00:06:44,900 and hundreds of millions of people can access that information 134 00:06:44,900 --> 00:06:46,360 It's really remarkable 135 00:06:46,360 --> 00:06:48,430 The bad news is is that hundreds of millions 136 00:06:48,430 --> 00:06:50,560 of people might actually try to access that information 137 00:06:50,560 --> 00:06:52,630 and they might try to do it at once 138 00:06:52,630 --> 00:06:57,360 And the first famous example of this was the Victoria's Secret fashion show 139 00:06:57,360 --> 00:07:00,100 that was advertised in the ¡®99 Super Bowl 140 00:07:00,100 --> 00:07:02,330 And probably a lot of you remember that 141 00:07:04,000 --> 00:07:06,330 they had these models walking down the runway in lingerie in the Super Bowl ad 142 00:07:06,330 --> 00:07:08,760 and they say if you want to see the the whole show 143 00:07:08,760 --> 00:07:11,560 come Tuesday night at nine o'clock to our website 144 00:07:11,560 --> 00:07:15,500 You know at Tuesday night at nine o'clock and ten seconds that site was toast 145 00:07:15,500 --> 00:07:18,600 along with most the internet in the surrounding area 146 00:07:18,600 --> 00:07:20,900 for the Victoria's Secret website 147 00:07:20,900 --> 00:07:23,130 And that's when you know people really started catching on 148 00:07:23,130 --> 00:07:25,960 There were several more examples where there was hot content 149 00:07:25,960 --> 00:07:30,630 and you couldn't support all the traffic of the people who wanted to get there 150 00:07:30,630 --> 00:07:32,800 And if you do try to support it 151 00:07:32,800 --> 00:07:35,760 you sort of have to guess how big that crowd's going to be 152 00:07:35,760 --> 00:07:38,200 You know all the e-commerce providers have to guess 153 00:07:38,200 --> 00:07:40,360 this Christmas how many people are coming 154 00:07:40,360 --> 00:07:42,360 how big do I build my infrastructure 155 00:07:42,360 --> 00:07:44,530 You know for the tax guys it's on April fifteenth 156 00:07:44,530 --> 00:07:47,230 How big to build the infrastructure for the one day? 157 00:07:47,230 --> 00:07:50,630 For the folks at you know online commerce for flowers or for mother's day 158 00:07:50,630 --> 00:07:54,460 there's big dates where you put all this provisioning in 159 00:07:54,460 --> 00:07:56,730 and if you guessed right 160 00:07:56,730 --> 00:07:58,730 you overspent for the rest of the year 161 00:07:58,730 --> 00:08:01,560 If you guessed wrong you overspent and you die 162 00:08:01,560 --> 00:08:02,760 And it's-- it's a problem 163 00:08:02,760 --> 00:08:06,560 And this is leading to the whole notion of on demand computing that you see IBM 164 00:08:06,560 --> 00:08:10,600 and a lot of the other big companies now putting a major marketing investment in 165 00:08:12,160 --> 00:08:14,200 What's less well understood are the problems 166 00:08:14,200 --> 00:08:16,360 in the inherent infrastructure of the internet 167 00:08:16,360 --> 00:08:19,230 So after Tim came to us and we started looking into the problem 168 00:08:19,230 --> 00:08:21,730 we realized there's even deeper more fundamental issues 169 00:08:21,730 --> 00:08:25,360 then just the centralized notion of distributing content 170 00:08:25,360 --> 00:08:27,900 And the first is that you know the 171 00:08:27,560 --> 00:08:31,460 people have a notion that the internet has a lot of unlit fiber 172 00:08:31,460 --> 00:08:33,630 There's a ton of capacity 173 00:08:33,630 --> 00:08:35,960 And to some extent that's true 174 00:08:36,300 --> 00:08:37,400 you know there's big trunks from New York to Los Angeles 175 00:08:37,400 --> 00:08:40,630 some of which aren't even lit today 176 00:08:41,300 --> 00:08:41,500 So there's capacity there 177 00:08:41,500 --> 00:08:44,900 But the problem is not in the in the trunk fiber 178 00:08:45,230 --> 00:08:47,530 The problem is in the the routers that connect these big pipes 179 00:08:47,530 --> 00:08:51,330 at the peering relationships between the various ISPs 180 00:08:51,330 --> 00:08:55,900 So you can get all the bits you want from LA to New York on MCI 181 00:08:55,900 --> 00:08:59,600 You try to hand off those bits from MCI to Sprint in New York 182 00:08:59,600 --> 00:09:03,200 or to a lesser known ISP that's handing off to an end user 183 00:09:03,200 --> 00:09:05,160 and now you got a problem 184 00:09:05,160 --> 00:09:07,460 And this goes back-not to a technology problem 185 00:09:07,460 --> 00:09:09,060 we can build big routers 186 00:09:09,060 --> 00:09:11,430 there's companies that are really good at doing that- 187 00:09:11,430 --> 00:09:14,200 but it comes to a economic business problem 188 00:09:14,200 --> 00:09:15,960 The networks have not figured out 189 00:09:15,960 --> 00:09:18,600 how to charge each other for taking each other's traffic 190 00:09:18,600 --> 00:09:20,530 and and give it pr- providing transit for 191 00:09:20,530 --> 00:09:23,700 that traffic across the country or across the world 192 00:09:23,700 --> 00:09:26,330 And so the best defense is to limit 193 00:09:26,330 --> 00:09:29,000 how much of the other guy's traffic you put on your backbone 194 00:09:29,000 --> 00:09:31,030 And you do that at the peering point 195 00:09:31,030 --> 00:09:32,600 And that-- if you make that small 196 00:09:32,600 --> 00:09:36,160 that limits it right there they can't get anymore traffic through 197 00:09:36,160 --> 00:09:39,260 This causes congestion the internet 198 00:09:39,260 --> 00:09:42,100 And the congestion's not typically so much in the big backbone pipes 199 00:09:42,100 --> 00:09:44,000 it's at these peering relationships 200 00:09:44,000 --> 00:09:45,260 And when the router gets swamped 201 00:09:45,260 --> 00:09:48,300 it drops the pipe-- drops the packets on the floor 202 00:09:48,300 --> 00:09:49,230 Now 203 00:09:49,230 --> 00:09:51,060 that's compounded by the fact 204 00:09:51,060 --> 00:09:56,330 that the algorithms that route the packets ignore congestion 205 00:09:56,330 --> 00:09:58,230 The algorithm decides the path 206 00:09:58,230 --> 00:10:01,600 your packet takes to go from source to destination is called BGP 207 00:10:01,600 --> 00:10:02,860 border gateway protocol 208 00:10:02,860 --> 00:10:07,530 and that navigates between all the ISPs and picks the path your packet takes 209 00:10:07,530 --> 00:10:10,260 It tries to find the shortest path 210 00:10:10,260 --> 00:10:13,730 roughly in terms of how many routers does it go through 211 00:10:13,730 --> 00:10:17,030 But it doesn't have any way to differentiate between a router 212 00:10:17,030 --> 00:10:20,330 that has a ton of capacity and a router that is dying 213 00:10:20,330 --> 00:10:22,660 and dropping packets on the floor 214 00:10:22,660 --> 00:10:26,500 And it just keeps sending packets along the same path 215 00:10:26,500 --> 00:10:29,030 irrespective of whether or not the packets are getting through 216 00:10:29,030 --> 00:10:34,200 So that's a fundamental problem that you don't adapt to problems on the internet 217 00:10:34,200 --> 00:10:37,630 It actually gets worse than that when you go one level deeper 218 00:10:37,630 --> 00:10:39,900 and you look at the data 219 00:10:39,900 --> 00:10:45,160 that's supplied to BGP to find the shortest path 220 00:10:45,160 --> 00:10:48,060 One would think that you'd get the data that tells you 221 00:10:48,060 --> 00:10:51,530 okay what's the distance between this point and this point on a backbone 222 00:10:51,530 --> 00:10:54,130 and you'd send that data up into a shortest path algorithm 223 00:10:54,130 --> 00:10:57,130 which we've known how to solve for forty plus years 224 00:10:57,130 --> 00:10:58,760 and you'd get the answer 225 00:10:58,760 --> 00:11:00,400 That's not what happens 226 00:11:00,400 --> 00:11:04,000 Because of these economic problems between the ISPs 227 00:11:04,000 --> 00:11:10,430 every network sort of fabricates the data that it supplies to the whole internet 228 00:11:10,430 --> 00:11:11,300 For example 229 00:11:11,300 --> 00:11:14,400 if a network does not want somebody else using its backbone link 230 00:11:14,400 --> 00:11:17,100 between New York and Los Angeles 231 00:11:17,100 --> 00:11:19,400 which it's not getting paid for quite possibly 232 00:11:19,400 --> 00:11:23,860 they will say that the distance there is twenty instead of the true distance 233 00:11:23,860 --> 00:11:25,330 which is three 234 00:11:25,330 --> 00:11:28,060 Now when BGP computes shortest paths 235 00:11:28,060 --> 00:11:29,600 when it sees the twenty there 236 00:11:29,600 --> 00:11:32,360 well it tries to find an alternate path that's less 237 00:11:32,360 --> 00:11:36,030 So this is how networks influence traffic to go elsewhere 238 00:11:36,030 --> 00:11:39,400 so they're not used for transit when they're not getting paid 239 00:11:39,400 --> 00:11:42,130 Now everybody is playing this game 240 00:11:42,130 --> 00:11:45,930 And so what happens is the data fed into BGP is-- 241 00:11:45,930 --> 00:11:47,860 is in a lot of ways meaningless 242 00:11:47,860 --> 00:11:50,530 So your finding a shortest path based on bogus data 243 00:11:50,530 --> 00:11:52,600 So the path is not the shortest path 244 00:11:52,600 --> 00:11:55,400 And there's you know a great example you could do trace routes 245 00:11:55,400 --> 00:11:58,360 Just you know trace route to someplace and see the path it takes 246 00:11:58,360 --> 00:12:01,860 and usually you can see that that can't be the shortest path on the internet 247 00:12:01,860 --> 00:12:03,360 And there was a great example one day 248 00:12:03,360 --> 00:12:06,460 where we trace routed from our building at Akamai 249 00:12:06,460 --> 00:12:09,500 which you know you can walk to in five minutes from here 250 00:12:09,500 --> 00:12:13,000 to MIT and it went through Israel 251 00:12:13,000 --> 00:12:14,360 Okay? Now 252 00:12:14,360 --> 00:12:17,460 BGP is doing the right thing 253 00:12:17,460 --> 00:12:20,430 It is finding the shortest path based on the data it's given 254 00:12:20,430 --> 00:12:22,200 And this only way this can get-- 255 00:12:22,200 --> 00:12:27,530 this could happen to go through Israel is all the data is garbage that it's getting 256 00:12:27,530 --> 00:12:28,830 Okay now 257 00:12:28,830 --> 00:12:32,760 once you have human beings monkeying with the input data 258 00:12:32,760 --> 00:12:34,560 all hell can break loose 259 00:12:34,560 --> 00:12:35,260 Okay? 260 00:12:35,260 --> 00:12:38,330 So a great example of this was when level three 261 00:12:38,330 --> 00:12:42,660 the human being accidentally put in minus infinite 262 00:12:42,660 --> 00:12:45,530 the equivalent of minus infinite cost 263 00:12:45,530 --> 00:12:48,530 or-- or distance to go through its backbone 264 00:12:48,530 --> 00:12:50,430 Now when you do a shortest path computation 265 00:12:50,430 --> 00:12:53,760 and there's some edge that has minus infinite length 266 00:12:53,760 --> 00:12:57,800 any path using that edge has minus infinite length 267 00:12:57,800 --> 00:13:01,630 so immediately in the internet all the packets raced for level three 268 00:13:01,630 --> 00:13:03,560 Every path had to go through level three 269 00:13:03,560 --> 00:13:05,000 basically 270 00:13:05,000 --> 00:13:07,760 Well level three's peering points are small 271 00:13:07,760 --> 00:13:10,030 to protect themselves from too many packets from coming in 272 00:13:10,030 --> 00:13:11,900 Immediately they're congested 273 00:13:11,900 --> 00:13:13,530 BGP doesn't adapt 274 00:13:13,530 --> 00:13:15,430 All the packets fall on the floor 275 00:13:15,430 --> 00:13:18,830 So the internet ground to a halt for about an hour before 276 00:13:18,830 --> 00:13:20,560 they figured out what the hell went wrong 277 00:13:20,560 --> 00:13:23,400 and they corrected the information and propagated it 278 00:13:23,400 --> 00:13:25,260 Now the next day you look at the paper and it said 279 00:13:25,260 --> 00:13:29,600 ¡°Oh some tractor cut a fiber somewhere and that caused an intern- internet outage¡± 280 00:13:29,600 --> 00:13:30,300 Okay? 281 00:13:30,300 --> 00:13:32,930 Because they don't want to tell you the game that's being played 282 00:13:32,930 --> 00:13:36,530 There's not enough tractors in the world to cause all the problems 283 00:13:36,530 --> 00:13:38,500 that we've seen on the internet 284 00:13:39,600 --> 00:13:42,960 It actually gets worse that this 285 00:13:42,960 --> 00:13:47,560 The routing algorithms are subject to entire loss of routes 286 00:13:47,560 --> 00:13:50,160 and intentional theft of routes 287 00:13:50,160 --> 00:13:53,560 Now most of you would know this through the notion of black holing 288 00:13:53,560 --> 00:13:59,030 Accidentally some other guy at some ISP said that he owns your IP address 289 00:13:59,030 --> 00:14:02,160 And some-- and he typed that into his router and that got propagated 290 00:14:02,160 --> 00:14:03,130 and so all of the sudden 291 00:14:03,130 --> 00:14:06,700 all of the traffic destined for you doesn't come to you anymore 292 00:14:06,700 --> 00:14:09,530 It goes to that other ISP where some guy made a mistake 293 00:14:09,530 --> 00:14:11,400 typed in the wrong thing 294 00:14:11,400 --> 00:14:13,730 And then that ISP doesn't own that IP address 295 00:14:13,730 --> 00:14:17,400 so it just forgets about it and drops the packets on the floor 296 00:14:17,400 --> 00:14:18,900 And this happens all the time 297 00:14:18,900 --> 00:14:23,100 and it's a big economic impact to people that are running e-commerce sites 298 00:14:23,100 --> 00:14:27,430 We had one large hotel chain customer had this happen for five hours 299 00:14:27,430 --> 00:14:30,730 Comcast accidentally stole their IP address on the internet 300 00:14:30,730 --> 00:14:32,960 Nobody could get to the-- that web site for five hours 301 00:14:32,960 --> 00:14:34,630 cost them millions of dollars 302 00:14:34,630 --> 00:14:38,060 And they're still in legal-- legal fights over who has responsibility 303 00:14:38,060 --> 00:14:40,630 But that's just the accidental misuse 304 00:14:40,630 --> 00:14:45,660 Spammers do this to inject spam and blame the innocent bystander 305 00:14:45,660 --> 00:14:48,230 Okay they will steal somebody's IP space 306 00:14:48,230 --> 00:14:51,560 because nobody's really checking about who owns the IP space 307 00:14:51,560 --> 00:14:53,460 And the way you do this is very simple 308 00:14:53,460 --> 00:14:55,800 You-- you get a router or a T1 309 00:14:55,800 --> 00:14:58,860 and you just say¡°I own these IP addresses¡± 310 00:14:58,860 --> 00:15:01,960 And you communicate that to the router at the other end 311 00:15:01,960 --> 00:15:04,760 And it just spreads that information through the internet 312 00:15:04,760 --> 00:15:07,200 In fact the bad guy can spread it as-- 313 00:15:07,200 --> 00:15:08,730 he can he can actually control the spread 314 00:15:08,730 --> 00:15:10,630 He can go all way or a little way 315 00:15:10,630 --> 00:15:12,560 Once he has the IP space 316 00:15:12,560 --> 00:15:13,860 dumps on the spam 317 00:15:13,860 --> 00:15:15,830 it goes out then he goes away 318 00:15:15,830 --> 00:15:18,460 he takes he-- he gets off the net 319 00:15:18,460 --> 00:15:21,300 Now when you go to investigate who sent all that spam 320 00:15:21,300 --> 00:15:23,730 you find the guy who owns the IP address 321 00:15:23,730 --> 00:15:25,630 who has no clue what happened 322 00:15:25,630 --> 00:15:27,000 Okay? 323 00:15:27,000 --> 00:15:29,430 It gets worse than that in terms of e-commerce 324 00:15:29,430 --> 00:15:31,260 Like say you do banking on the internet 325 00:15:31,260 --> 00:15:33,230 I don't know maybe you have a fleet account here Boston 326 00:15:33,230 --> 00:15:34,100 and you have a password 327 00:15:34,100 --> 00:15:36,030 you can have your login name and your password 328 00:15:36,030 --> 00:15:38,160 I can steal that very easily 329 00:15:38,160 --> 00:15:39,200 Many different ways 330 00:15:39,200 --> 00:15:44,700 One of them is because the BGP doesn't authenticate ownership of IP addresses 331 00:15:44,700 --> 00:15:47,730 So the way I can steal your¡ªI-- your password 332 00:15:47,730 --> 00:15:51,630 is I go to your ISP 333 00:15:51,630 --> 00:15:58,160 and I send a message to your ISP saying that I own the IP address of the bank 334 00:15:58,160 --> 00:16:00,030 And I limit the spread of that information 335 00:16:00,030 --> 00:16:02,860 so it only goes to a particular ISP 336 00:16:02,860 --> 00:16:05,760 And that ISP they don't look at it they don't check 337 00:16:05,760 --> 00:16:07,430 They say¡°Okay¡± 338 00:16:07,430 --> 00:16:10,560 So immediately that ISP sends me all the traffic for the bank 339 00:16:10,560 --> 00:16:12,030 Comes to me 340 00:16:12,030 --> 00:16:13,360 So meanwhile 341 00:16:13,360 --> 00:16:15,330 I've downloaded the login page from the bank 342 00:16:15,330 --> 00:16:16,860 I just download that 343 00:16:16,860 --> 00:16:19,530 You come to me to login to do your banking transaction 344 00:16:19,530 --> 00:16:22,730 I show you the login page 345 00:16:22,730 --> 00:16:23,530 Okay? 346 00:16:23,530 --> 00:16:27,100 You type in your login identity and your password 347 00:16:27,100 --> 00:16:28,500 And then you know 348 00:16:28,500 --> 00:16:30,260 you get redirected to the site and oh you have to login again 349 00:16:30,260 --> 00:16:32,300 because there was some problem please login again 350 00:16:32,300 --> 00:16:33,260 Okay? 351 00:16:33,260 --> 00:16:36,060 You don't know that anything unusual happened 352 00:16:36,060 --> 00:16:36,930 Okay? 353 00:16:36,930 --> 00:16:38,230 The bank certainly doesn't know 354 00:16:38,230 --> 00:16:40,600 They just lost three percent of their traffic but that's noise 355 00:16:40,600 --> 00:16:41,560 They're not going to know 356 00:16:41,560 --> 00:16:44,930 And I just collected all the all the passwords for a particular ISP 357 00:16:44,930 --> 00:16:48,330 And that that's one of about five ways I can do it on the internet today 358 00:16:48,330 --> 00:16:50,200 And people don't have a clue 359 00:16:50,200 --> 00:16:52,930 okay that-- that you can-- that don't have a good understanding of 360 00:16:52,930 --> 00:16:55,360 how vulnerable things are on the internet today 361 00:16:55,360 --> 00:16:57,030 So these are some of the deeper problems 362 00:16:57,030 --> 00:17:00,360 that aren't as well understood in the internet infrastructure today 363 00:17:00,360 --> 00:17:04,760 And this is where we get started with with Akamai 364 00:17:04,760 --> 00:17:05,830 Now 365 00:17:05,830 --> 00:17:07,430 one of the issues we faced early on 366 00:17:07,430 --> 00:17:08,800 and still face today 367 00:17:08,800 --> 00:17:10,330 is-- is people have the impression well 368 00:17:10,330 --> 00:17:12,360 it's going to get better 369 00:17:12,360 --> 00:17:14,830 that there's consolidation 370 00:17:14,830 --> 00:17:17,100 that there's going to be two with three tier one networks 371 00:17:17,100 --> 00:17:18,760 that are going to own everything 372 00:17:18,760 --> 00:17:20,960 they're going to get together and they're going to work it out 373 00:17:20,960 --> 00:17:22,630 And really the reverse is happening 374 00:17:22,630 --> 00:17:24,160 When we started Akamai 375 00:17:24,160 --> 00:17:26,400 there were seven thousand networks that made up the internet 376 00:17:26,400 --> 00:17:28,400 And today there's fifteen thousand 377 00:17:28,400 --> 00:17:31,200 There's been a lot of high profile bankruptcies 378 00:17:31,200 --> 00:17:34,800 but not a lot of consolidation even among the tier ones 379 00:17:34,800 --> 00:17:38,700 The other misconception is the tier ones run the show 380 00:17:38,700 --> 00:17:39,930 And that's not true 381 00:17:39,930 --> 00:17:42,660 You know pretty much any tier one will claim a forty percent market share 382 00:17:42,660 --> 00:17:44,830 You know all two dozen of them 383 00:17:44,830 --> 00:17:46,760 And you-- you can sort of know that's not right 384 00:17:46,760 --> 00:17:48,630 You know that's not hard to check that 385 00:17:48,630 --> 00:17:49,930 But in fact 386 00:17:49,930 --> 00:17:52,530 all the tier ones put together control less than 387 00:17:52,530 --> 00:17:55,400 twenty five percent of the bits that are handed to end users 388 00:17:55,400 --> 00:17:57,500 Most of the bits are handed off to end users 389 00:17:57,500 --> 00:17:59,600 by companies you've never heard of 390 00:17:59,600 --> 00:18:01,400 And the curve 391 00:18:01,400 --> 00:18:03,130 if you want to get a ninety five percent coverage 392 00:18:03,130 --> 00:18:06,000 you're into probably close to ten thousand ISPs 393 00:18:06,000 --> 00:18:08,460 in terms of handing the bits off to end users today 394 00:18:08,460 --> 00:18:12,960 So it really is a very distributed infrastructure 395 00:18:12,960 --> 00:18:16,960 that doesn't really have any particular control over it 396 00:18:16,960 --> 00:18:18,330 Those ISPs are in the US? 397 00:18:18,330 --> 00:18:19,030 What's that? 398 00:18:19,030 --> 00:18:20,630 How many of those ISPs are in the US? 399 00:18:20,630 --> 00:18:21,930 I don't now know the exact figure 400 00:18:21,930 --> 00:18:24,560 but there's dramatic growth in traffic now overseas 401 00:18:24,560 --> 00:18:27,960 You know today China is the fastest growing sink of bits on the internet 402 00:18:27,960 --> 00:18:29,030 And Brazil 403 00:18:29,030 --> 00:18:34,060 one of the-- and also a very large growing consumer of bits on the internet today 404 00:18:34,060 --> 00:18:36,000 So it's becoming much more international 405 00:18:36,000 --> 00:18:37,960 If you look at our traffic graphs 406 00:18:37,960 --> 00:18:40,330 they're getting almost almost flat now 407 00:18:40,330 --> 00:18:41,930 You know where as you know several years ago 408 00:18:41,930 --> 00:18:44,460 at night time there was a lot less traffic in the US 409 00:18:44,460 --> 00:18:47,700 So then the-- the internet is becoming very international 410 00:18:49,230 --> 00:18:50,660 Okay so what are the implications 411 00:18:50,660 --> 00:18:52,930 what creates a business case potentially for a company 412 00:18:52,930 --> 00:18:57,100 or something that makes it interesting as a research subject? 413 00:18:57,100 --> 00:19:00,030 Well these problems makes downloads be slow 414 00:19:00,030 --> 00:19:02,230 Applications are very slow 415 00:19:02,230 --> 00:19:04,360 They're unreliable 416 00:19:04,360 --> 00:19:06,960 The infrastructure's inherently not scalable 417 00:19:06,960 --> 00:19:08,100 the architecture of the internet 418 00:19:08,100 --> 00:19:10,830 And Tim recognized this way back in '95 419 00:19:10,830 --> 00:19:13,760 four years before Victoria's Secret's fashion show 420 00:19:13,760 --> 00:19:16,160 You know he said ¡°This is going to be an issue¡± 421 00:19:16,160 --> 00:19:21,260 It's insecure and-- and just much more than people really realize 422 00:19:21,260 --> 00:19:22,900 It's expensive 423 00:19:22,900 --> 00:19:25,530 If you want to provision an e-commerce site 424 00:19:25,530 --> 00:19:27,930 or something that's going to do business or protect your brand on the internet 425 00:19:27,930 --> 00:19:32,060 you've got to spend a lot of money building up your centralized infrastructure 426 00:19:32,060 --> 00:19:33,330 And 427 00:19:33,330 --> 00:19:37,430 you know another misconception is that broadband solves the problem 428 00:19:37,430 --> 00:19:38,230 Okay? 429 00:19:38,230 --> 00:19:41,230 Now actually broadband makes the problem worse 430 00:19:41,230 --> 00:19:43,260 Because now you have people who are sinking bits 431 00:19:43,260 --> 00:19:47,600 they expect to sink bits at twenty times the rate they were before 432 00:19:47,600 --> 00:19:48,500 Right? 433 00:19:48,500 --> 00:19:51,500 Instead of you know fifty kilobits a second they want to go to a megabit a second 434 00:19:51,500 --> 00:19:53,730 That puts more load on the central infrastructure 435 00:19:53,730 --> 00:19:55,460 and on the peering relationships 436 00:19:55,460 --> 00:19:57,830 So it creates more of a problem 437 00:19:57,830 --> 00:19:59,400 It solves the last mile problem 438 00:19:59,400 --> 00:20:02,760 but not the the internal infrastructure of the internet 439 00:20:04,300 --> 00:20:07,000 Okay so what was the idea at a high level behind 440 00:20:07,000 --> 00:20:09,260 what Akamai was going to do and back then the research? 441 00:20:09,260 --> 00:20:12,400 Alright? So Tim comes in and says ¡°We've got this problem¡± 442 00:20:12,400 --> 00:20:15,830 okay well-- well what do we think we might try to do to 443 00:20:15,830 --> 00:20:17,230 you know deal with that? 444 00:20:17,230 --> 00:20:21,200 And the idea at a high level is very simple 445 00:20:21,200 --> 00:20:24,600 Use a distributed infrastructure and architecture 446 00:20:24,600 --> 00:20:27,460 instead of a centralized infrastructure and architecture 447 00:20:27,460 --> 00:20:29,130 And that's an easy idea for us 448 00:20:29,130 --> 00:20:32,300 because we did distributed algorithms on networks for a living 449 00:20:32,300 --> 00:20:33,730 that's what we liked to do 450 00:20:33,730 --> 00:20:37,160 So even if it wasn't a good answer that's what we were going to do 451 00:20:37,160 --> 00:20:39,160 okay because that's sort of what we did 452 00:20:39,160 --> 00:20:40,900 We'd write a paper on it if it-- you know 453 00:20:40,900 --> 00:20:42,660 and that would be the end of it if it weren't useful 454 00:20:42,660 --> 00:20:46,200 So the idea is-- is to take a zillion servers 455 00:20:46,200 --> 00:20:50,500 place them all around the world across the last mile from the end users 456 00:20:50,500 --> 00:20:54,030 Place these servers in those fifteen thousand ISPs 457 00:20:54,030 --> 00:20:58,100 And instead of serving the content from across the world or across the country 458 00:20:58,100 --> 00:21:01,700 get it into these-- these servers here and serve it locally 459 00:21:01,700 --> 00:21:03,630 And if every once and a while you needed to go back 460 00:21:03,630 --> 00:21:05,130 and get something from the origin 461 00:21:05,130 --> 00:21:08,300 we'll you'd build some kind of architecture to do that 462 00:21:08,300 --> 00:21:10,360 But now all the interaction is local 463 00:21:10,360 --> 00:21:12,000 That's the high level idea 464 00:21:13,800 --> 00:21:17,900 And the-- the idea that is that this would be better 465 00:21:17,900 --> 00:21:19,660 because it would be faster 466 00:21:19,660 --> 00:21:22,000 Everything's local so it's going to be faster 467 00:21:22,000 --> 00:21:25,630 It's going to be more reliable because you've got a distributed infrastructure 468 00:21:25,630 --> 00:21:29,000 If some of the servers die who cares you've got 10000 more 469 00:21:29,560 --> 00:21:31,260 It would be scalable 470 00:21:31,260 --> 00:21:34,230 You wouldn't have to worry about scaling up that central infrastructure 471 00:21:34,230 --> 00:21:36,200 because you've already built this large platform 472 00:21:36,200 --> 00:21:39,300 and an individual could use more of it on demand 473 00:21:39,300 --> 00:21:43,100 It would be secure because you couldn't even get the-- 474 00:21:43,100 --> 00:21:45,760 the end users couldn't get to an origin infrastructure anymore 475 00:21:45,760 --> 00:21:49,160 They'd only have to deal with these server spread around the world 476 00:21:49,160 --> 00:21:53,100 and they would be beefed up to provide a shield to an origin infrastructure 477 00:21:53,100 --> 00:21:54,900 And it would be cost effective 478 00:21:54,900 --> 00:21:58,360 You could make a a pitch to somebody who owns a website saying 479 00:21:58,360 --> 00:21:59,860 ¡°Don't worry about provisioning 480 00:21:59,860 --> 00:22:02,000 Don't worry about redundant data centers 481 00:22:02,000 --> 00:22:05,700 We'll get it all back up for you and replicated in our distributed network¡± 482 00:22:05,700 --> 00:22:06,860 And it's easy to manage 483 00:22:06,860 --> 00:22:10,060 We take care of all that hard stuff using the math 484 00:22:10,060 --> 00:22:12,200 okay the distributed computing 485 00:22:14,600 --> 00:22:16,500 Okay So now 486 00:22:16,500 --> 00:22:18,600 well how do you make that really work? 487 00:22:18,600 --> 00:22:21,630 You know and-- and back then you know back in '95 488 00:22:21,630 --> 00:22:24,200 you went out into practice and you said 489 00:22:24,200 --> 00:22:26,060 ¡°We're going to build this gigantic distributed network 490 00:22:26,060 --> 00:22:29,130 and it's going to serve all the web pages from all the leading websites ¡± 491 00:22:29,130 --> 00:22:31,400 and they'd have a great yuck yuck at that right? 492 00:22:31,400 --> 00:22:32,930 That-- that's pretty funny 493 00:22:32,930 --> 00:22:35,760 And there's a reason you guys are in the ivory tower 494 00:22:35,760 --> 00:22:37,960 instead of out here in industry 495 00:22:37,960 --> 00:22:42,660 So we-- we spent a lot of time actually doing the theory 496 00:22:42,660 --> 00:22:44,160 figuring out how to make it work 497 00:22:44,160 --> 00:22:49,660 and a lot of our early effort was how do you coordinate all those servers 498 00:22:49,660 --> 00:22:52,260 that live in this unreliable internet 499 00:22:52,260 --> 00:22:57,700 How do you direct an end user who wants yahoo's web page to the server 500 00:22:57,700 --> 00:22:58,930 that's really near by them 501 00:22:58,930 --> 00:23:01,500 that has yahoo's web page and isn't overloaded 502 00:23:01,500 --> 00:23:02,800 and all those kind of things 503 00:23:02,800 --> 00:23:04,160 And you got to do that in real time 504 00:23:04,160 --> 00:23:05,200 You can't 505 00:23:05,200 --> 00:23:07,330 you know the request doesn't come in when they click the browser 506 00:23:07,330 --> 00:23:09,230 and you think for a few minutes to give them the answer 507 00:23:09,230 --> 00:23:11,900 It's got to be ready-- ready to go real time 508 00:23:11,900 --> 00:23:14,900 So there's a lot of effort in coordinating that 509 00:23:15,260 --> 00:23:19,130 You know this is not a-- probably we thought about in the early days 510 00:23:19,130 --> 00:23:21,900 but as a-- as a company you got to bill for this 511 00:23:21,900 --> 00:23:24,130 You know the guy who has the web site has to-- 512 00:23:24,130 --> 00:23:26,730 you have to send him a bill to be in business 513 00:23:26,730 --> 00:23:29,100 And actually 514 00:23:29,100 --> 00:23:33,030 we're billing today for thirty billion things a day 515 00:23:33,030 --> 00:23:36,000 It's-- you take the-- the phone company's billing problem 516 00:23:36,000 --> 00:23:37,730 and multiply by ten or more 517 00:23:37,730 --> 00:23:40,200 and that's the billing problem we have today 518 00:23:40,200 --> 00:23:42,160 Because you're-- we're billing for every single hit 519 00:23:42,160 --> 00:23:45,160 and there's you know many tens of billions of hits a day 520 00:23:45,160 --> 00:23:47,930 That required a ton of effort in terms of 521 00:23:47,930 --> 00:23:50,600 thinking ahead of time to manage that problem 522 00:23:50,600 --> 00:23:54,400 So today we manage that with a team of maybe half a dozen people 523 00:23:54,400 --> 00:23:56,960 you know for billing And it required a lot of thought to 524 00:23:56,960 --> 00:24:00,060 how do you get a large scale distributed billing process 525 00:24:00,060 --> 00:24:01,530 Also there's reporting 526 00:24:01,530 --> 00:24:03,360 Once you take over a customer's website 527 00:24:03,360 --> 00:24:05,160 and you're serving from these servers all around the world 528 00:24:05,160 --> 00:24:07,330 they want to see what's happening to their content 529 00:24:07,330 --> 00:24:09,030 and you got to report on that in real time 530 00:24:09,030 --> 00:24:12,160 And that's a huge information gathering problem 531 00:24:12,160 --> 00:24:15,260 That is very complicated 532 00:24:15,260 --> 00:24:18,330 How do you make it really work with operations? 533 00:24:18,330 --> 00:24:22,730 That your running this thing so it's-- it's not just the paper anymore 534 00:24:22,730 --> 00:24:25,300 There's-- first there's code which was very scary 535 00:24:25,300 --> 00:24:29,060 and then the code has to work and you got to maintain it 536 00:24:29,060 --> 00:24:32,760 So a lot of research there 537 00:24:32,760 --> 00:24:35,560 When you're handling content from web sites 538 00:24:35,560 --> 00:24:39,360 people change their content all the time in unpredictable ways 539 00:24:39,360 --> 00:24:41,530 And you got to make sure the content is always fresh 540 00:24:41,530 --> 00:24:43,430 so you're never serving anything stale 541 00:24:43,430 --> 00:24:46,030 without having to always go back and get it 542 00:24:46,030 --> 00:24:51,800 Because the whole point is it's-- it's in this server local to the end user 543 00:24:51,800 --> 00:24:53,630 Of course then you have things like streaming 544 00:24:53,630 --> 00:24:57,600 live streaming and webcasting where you've got to manage-- 545 00:24:57,600 --> 00:24:59,860 you don't you don't have the content stored there ahead of time 546 00:24:59,860 --> 00:25:01,760 it's coming live from a source single source 547 00:25:01,760 --> 00:25:05,630 that you've got to distribute in a fault tolerant way through the internet 548 00:25:05,630 --> 00:25:08,660 You know and multicast won't cut it because if you cut it 549 00:25:08,660 --> 00:25:12,830 if you if you break the link somewhere everybody downstream in that tree is dead 550 00:25:12,830 --> 00:25:15,000 So you got to build something that's fault tolerant 551 00:25:15,000 --> 00:25:17,900 to go through the internet for your-- for your streaming 552 00:25:19,100 --> 00:25:24,960 If we're going to have the entire interaction be at a server near the end user 553 00:25:24,960 --> 00:25:27,360 you've got to be able to assemble the page dynamically 554 00:25:27,360 --> 00:25:30,360 because pretty much all the web sites today are personalized and dynamic 555 00:25:30,360 --> 00:25:33,000 The content you get depends on where you're located 556 00:25:33,000 --> 00:25:35,700 who you are what your cookies say about you 557 00:25:35,700 --> 00:25:37,500 and the time of day 558 00:25:37,500 --> 00:25:40,730 So it's a s-- the HTML is assembled dynamically 559 00:25:40,730 --> 00:25:44,700 Most pages are not cacheable today 560 00:25:45,700 --> 00:25:47,130 And finally 561 00:25:47,130 --> 00:25:48,860 you're going to want to have the applications 562 00:25:48,860 --> 00:25:50,960 run on this server near the end user 563 00:25:50,960 --> 00:25:52,660 not at the origin infrastructure 564 00:25:52,660 --> 00:25:54,560 Which is a whole another level of 565 00:25:54,560 --> 00:25:56,600 complication to have the apps running 566 00:25:56,600 --> 00:25:59,660 out at-- out at the edge server instead of an origin infrastructure 567 00:25:59,660 --> 00:26:04,130 So these are tremendous research challenges which we spent years working on 568 00:26:04,130 --> 00:26:06,660 never thinking about a company quite happily cranking away 569 00:26:06,660 --> 00:26:11,060 writing our papers and writing theses and so forth now 570 00:26:11,060 --> 00:26:12,600 Now 571 00:26:12,600 --> 00:26:15,800 what converted this into practice for-- 572 00:26:15,800 --> 00:26:19,860 for Akamai was the fifty K competition 573 00:26:19,860 --> 00:26:23,060 This is a-- a competition at MIT run by the Sloan school 574 00:26:23,060 --> 00:26:25,860 It's a student run business plan competition 575 00:26:25,860 --> 00:26:29,060 and the winning team gets actually-it's 35000 dollars 576 00:26:29,060 --> 00:26:31,730 I think not fifty it's fifty total- 577 00:26:31,730 --> 00:26:35,700 and they get tremendous involvement among the community at MIT 578 00:26:35,700 --> 00:26:39,460 There's one to two hundred teams get together to write a business plan 579 00:26:39,460 --> 00:26:40,860 for the fifty K 580 00:26:40,860 --> 00:26:43,200 And so let me say a little bit about that process 581 00:26:43,200 --> 00:26:47,930 because that's what in our case made the tech transfer happen 582 00:26:50,830 --> 00:26:55,930 So it started with Danny Lewin and Preetish Nijhawan 583 00:26:55,930 --> 00:27:00,460 Danny was my student over in at then LCS 584 00:27:00,460 --> 00:27:03,460 and Preetish Nijhawan was his next door neighbor 585 00:27:03,460 --> 00:27:05,100 and he was in the Sloan school 586 00:27:05,100 --> 00:27:07,330 And Danny wrote a prize winning master the- 587 00:27:07,330 --> 00:27:11,060 masters thesis on how should content be delivered on the internet 588 00:27:11,060 --> 00:27:14,830 you know addressing a bunch of the research challenges that I just outlined 589 00:27:14,830 --> 00:27:18,030 And one night they were having beer and-- and Danny was-- 590 00:27:18,030 --> 00:27:20,730 they were commiserating about how Danny was never going to 591 00:27:20,730 --> 00:27:23,960 have a career and never pay off his student loans 592 00:27:23,960 --> 00:27:26,560 I think he had like 30000 dollars in student loans 593 00:27:26,560 --> 00:27:27,830 He was married 594 00:27:27,830 --> 00:27:33,260 He had two kids in private school and he was a theoretician 595 00:27:33,260 --> 00:27:36,260 That's like you know the worst possible combination 596 00:27:36,260 --> 00:27:39,800 because you know Preetish he's at the Sloan school 597 00:27:39,800 --> 00:27:42,560 He already had a job lined up quarter million bucks a year 598 00:27:42,560 --> 00:27:43,860 He's-- he's a happy camper 599 00:27:43,860 --> 00:27:44,860 Poor Danny 600 00:27:44,860 --> 00:27:46,360 he was in computer science 601 00:27:46,360 --> 00:27:47,930 but he was a-- a theoretician 602 00:27:47,930 --> 00:27:51,360 a theoretical computer science and the job prospects there were much 603 00:27:51,360 --> 00:27:54,600 much less you know promising to the point 604 00:27:54,600 --> 00:27:57,400 where he wasn't even sure he was ever going to pay off those student loans 605 00:27:57,400 --> 00:27:59,160 Now Preetish said 606 00:27:59,160 --> 00:28:00,630 ¡°Hey Danny¡± in an effort to cheer him up 607 00:28:00,630 --> 00:28:03,100 ¡°there's this thing called the fifty K competition 608 00:28:03,100 --> 00:28:05,030 You're doing some cool stuff in your thesis 609 00:28:05,030 --> 00:28:06,760 Write it up into a business plan 610 00:28:06,760 --> 00:28:10,600 You might win and you might get 50000 dollars¡± 611 00:28:10,600 --> 00:28:13,030 So Danny was all excited 612 00:28:13,030 --> 00:28:14,930 They needed the-- somebody with gray hair 613 00:28:14,930 --> 00:28:17,230 so the two of them came and corralled me and they talked me into it 614 00:28:17,230 --> 00:28:21,530 So we wrote a three page business plan and sent it off 615 00:28:23,030 --> 00:28:26,930 It turns out there's a-- a early round of the fifty K called the one K 616 00:28:26,930 --> 00:28:29,460 And they have ten categories in the one K 617 00:28:29,460 --> 00:28:32,300 And lo and behold we won 618 00:28:32,300 --> 00:28:35,600 the software category based on our three page business plan 619 00:28:35,600 --> 00:28:37,430 Now it wasn't a thousand dollars we got 620 00:28:37,430 --> 00:28:40,330 since there's ten categories we got a hundred bucks 621 00:28:40,760 --> 00:28:44,000 So we went out to have some more beer with a hundred dollars 622 00:28:44,000 --> 00:28:47,030 Now the three of us are having beer and Preetish 623 00:28:47,030 --> 00:28:48,560 you know who's involved in the Sloan school said 624 00:28:48,560 --> 00:28:49,830 ¡°Hey guys this is pretty good 625 00:28:49,830 --> 00:28:52,030 We-- we won the software category here 626 00:28:52,030 --> 00:28:54,930 We got to go with a full round and-- and we have prospects here¡± 627 00:28:54,930 --> 00:28:56,630 You know he-- Preetish was a little clued into 628 00:28:56,630 --> 00:28:58,260 what was going on out in the real world 629 00:28:58,260 --> 00:29:00,030 and-- and the fifty K and so forth 630 00:29:00,030 --> 00:29:02,300 And so we said ¡°Okay¡± 631 00:29:02,300 --> 00:29:06,660 And we got together and we wrote a-- a twenty five page business plan 632 00:29:06,660 --> 00:29:10,930 And-- and back then we measured the quality of our plan based on its length 633 00:29:10,930 --> 00:29:15,200 We had no idea what really a business plan meant 634 00:29:15,930 --> 00:29:17,030 Now 635 00:29:17,030 --> 00:29:22,830 somehow we made the final six out of the 120 or 130 teams they had 636 00:29:22,830 --> 00:29:27,400 And Preetish came and said ¡°Guys this is-- this is really big¡± 637 00:29:27,400 --> 00:29:29,160 He's just totally excited 638 00:29:29,160 --> 00:29:30,430 And you know 639 00:29:30,430 --> 00:29:32,830 Danny and I and the and--the people who had joined us by that time 640 00:29:32,830 --> 00:29:34,560 we were doing this as an academic project 641 00:29:34,560 --> 00:29:36,330 We could learn about how to write a business plan 642 00:29:36,330 --> 00:29:39,100 We went out and got books: ¡°how to write a business plan¡± 643 00:29:39,100 --> 00:29:43,030 You know we started to talking to potential customers 644 00:29:43,030 --> 00:29:43,960 Now back then 645 00:29:43,960 --> 00:29:47,800 we had this crazy idea we were going to sell our technology to ISPs 646 00:29:47,800 --> 00:29:49,460 And so we went out and talked to a couple 647 00:29:49,460 --> 00:29:52,060 And then we got very discouraged because we realized even back 648 00:29:52,060 --> 00:29:53,660 then they were about to go broke 649 00:29:53,660 --> 00:29:55,200 and there are fires burning left and right 650 00:29:55,200 --> 00:29:57,600 and there's no way they were going to pay for our stuff 651 00:29:57,600 --> 00:30:00,930 But then we got the idea we would talk to people with websites 652 00:30:00,930 --> 00:30:02,400 You know we called up CNN 653 00:30:02,400 --> 00:30:04,100 We called up Yahoo 654 00:30:04,100 --> 00:30:07,100 And because we were MIT and not selling them anything 655 00:30:07,100 --> 00:30:08,830 they were happy to talk to us 656 00:30:08,830 --> 00:30:11,200 And they said ¡°Well that's sort of cool stuff¡± 657 00:30:11,200 --> 00:30:15,530 you know and¡°if you changed it this way yeah we might try that out¡± kind of level 658 00:30:15,530 --> 00:30:17,630 So it got us a little more interested 659 00:30:18,130 --> 00:30:21,060 And then I started to get the impression this might've been big 660 00:30:21,060 --> 00:30:24,260 when some business guy from California 661 00:30:24,260 --> 00:30:27,200 showed up in my office wanting to join the team 662 00:30:27,200 --> 00:30:28,000 you know 663 00:30:28,000 --> 00:30:29,830 What? 664 00:30:29,830 --> 00:30:31,500 You know so I said¡°Okay¡± 665 00:30:31,500 --> 00:30:33,730 And by now we had like twenty five people on the team 666 00:30:33,730 --> 00:30:38,460 And then I knew it was really big when a guy from Harvard business school 667 00:30:38,460 --> 00:30:41,830 came all the way down Mass Ave to join the team 668 00:30:41,830 --> 00:30:44,260 And then-- then I knew we were on to something 669 00:30:44,260 --> 00:30:46,730 you know at that point 670 00:30:47,030 --> 00:30:50,430 So the team got up to about thirty people at this point 671 00:30:50,430 --> 00:30:51,800 and it was really getting pretty active 672 00:30:51,800 --> 00:30:55,130 And we wrote up a much better business plan 673 00:30:55,130 --> 00:30:59,500 It was fifty five pages at that point 674 00:30:59,500 --> 00:31:01,760 And we went into the final six 675 00:31:02,160 --> 00:31:03,860 And we lost 676 00:31:03,860 --> 00:31:06,930 And not only did we lose we lost badly 677 00:31:06,930 --> 00:31:09,100 They-- they announced first second and third 678 00:31:09,100 --> 00:31:12,060 First was a nonprofit organization 679 00:31:12,060 --> 00:31:14,600 And this is a business plan competition 680 00:31:14,600 --> 00:31:17,630 I am-- I am sure we were dead last 681 00:31:17,630 --> 00:31:20,730 They just don't announce four five and six to avoid the embarrassment 682 00:31:20,730 --> 00:31:22,960 And this was a slap in the face 683 00:31:22,960 --> 00:31:26,200 This was a group of people who weren't used to losing 684 00:31:26,200 --> 00:31:28,430 and weren't used to being embarrassed 685 00:31:28,430 --> 00:31:31,330 you know And we were embarrassed when we saw the quality of the other presentations 686 00:31:31,330 --> 00:31:33,300 the other five teams They were good 687 00:31:33,300 --> 00:31:34,430 They knew what they were doing 688 00:31:34,430 --> 00:31:37,700 And we realized we didn't know what the hell we were doing in terms of a business 689 00:31:37,700 --> 00:31:40,960 You know we could write great papers that very few people would ever read 690 00:31:40,960 --> 00:31:42,500 but when it came to a business 691 00:31:42,500 --> 00:31:44,660 we had a lot to learn 692 00:31:44,660 --> 00:31:47,760 Now the fact that we lost notwithstanding 693 00:31:47,760 --> 00:31:50,060 the VCs came to us after the fifty K 694 00:31:50,060 --> 00:31:51,860 And they sort of see what goes on 695 00:31:51,860 --> 00:31:55,230 And they said¡°Guys look this is MIT It's the internet it's hot 696 00:31:55,230 --> 00:31:56,360 We'll make a company 697 00:31:56,360 --> 00:31:59,430 we'll flip it six months everybody goes home happy¡± 698 00:31:59,430 --> 00:32:01,830 Well we didn't want to do that 699 00:32:01,830 --> 00:32:06,800 And we actually made an active decision not to form a company in the spring of '98 700 00:32:06,800 --> 00:32:09,800 You know Danny wanted to be a professor 701 00:32:09,800 --> 00:32:12,230 even though he was never going to pay off his student loans 702 00:32:12,230 --> 00:32:14,130 I liked being a professor 703 00:32:14,130 --> 00:32:16,260 The students by and large liked what they were doing 704 00:32:16,260 --> 00:32:19,060 and this had been an academic project 705 00:32:19,060 --> 00:32:21,800 So we-- we said¡°No we don't-- we don't want to do that¡± 706 00:32:21,800 --> 00:32:23,530 But we were intrigued 707 00:32:23,530 --> 00:32:25,260 you know by the idea 708 00:32:25,260 --> 00:32:28,530 And we said would you know if you really felt we could commercialize this 709 00:32:28,530 --> 00:32:30,300 and could be really successful 710 00:32:30,300 --> 00:32:32,400 long term successful it's worth doing 711 00:32:32,400 --> 00:32:35,300 to get the technology out there and to make a difference 712 00:32:35,300 --> 00:32:41,430 And so what we did is we spent the summer with a program at MIT in-- 713 00:32:41,430 --> 00:32:42,560 in the lab computer science 714 00:32:42,560 --> 00:32:45,460 It was then called ¡°hacker haven¡± or ¡°hacker heaven¡± 715 00:32:45,460 --> 00:32:48,230 And this is a program set up in the lab we get-- 716 00:32:48,230 --> 00:32:52,630 there's maybe fifty really smart undergrads that love coding 717 00:32:52,630 --> 00:32:55,560 And you take them for the summer and you put them on a project 718 00:32:55,560 --> 00:32:59,460 And we one of the projects where here we-- we wrote all these great papers 719 00:32:59,460 --> 00:33:03,330 we have these cool algorithms but we really hadn't written any code 720 00:33:03,330 --> 00:33:07,900 Let's make some prototypes to see how a system would work for content delivery 721 00:33:07,900 --> 00:33:10,900 And a bunch of the students thought that was cool so they signed up 722 00:33:10,900 --> 00:33:14,730 And meanwhile we'd tal-- one of the folks we called up to see would 723 00:33:14,730 --> 00:33:18,300 they ever buy this service from us was Paramount Digital Entertainment 724 00:33:18,300 --> 00:33:21,300 Now Paramount owns Star Trek 725 00:33:21,300 --> 00:33:24,260 And they said ¡°Hey look If you guys could do this 726 00:33:24,260 --> 00:33:28,500 we'll let you distribute the Star Trek TV series over the internet 727 00:33:28,500 --> 00:33:31,800 And by the way here's a bunch of Star Trek paraphernalia¡± 728 00:33:31,800 --> 00:33:34,060 So I got my captain Kirk tie 729 00:33:34,060 --> 00:33:36,500 and my Doctor Spock stuff and all the cool hats 730 00:33:36,500 --> 00:33:38,030 The kids went nuts 731 00:33:38,030 --> 00:33:40,730 You know they-- they twenty four hours a day they're writing code 732 00:33:40,730 --> 00:33:41,600 you know 733 00:33:41,600 --> 00:33:44,800 It just¡ªthat-- that little tie and the hat with Captain Kirk was all it-- 734 00:33:44,800 --> 00:33:46,960 and Doctor Spock was all it took 735 00:33:46,960 --> 00:33:48,960 So they're cranking away over the summer 736 00:33:48,960 --> 00:33:52,430 and by the end of the summer we actually had two prototypes that showed that 737 00:33:52,430 --> 00:33:55,000 hey this really can work in code 738 00:33:55,000 --> 00:33:57,100 And we thought some more about the business plan 739 00:33:57,100 --> 00:33:59,330 and we actually have a plan that we started believing in 740 00:33:59,330 --> 00:34:01,100 as opposed to the fifty five page plan 741 00:34:01,100 --> 00:34:04,930 where we sort of understood the number of pages in the plan 742 00:34:04,930 --> 00:34:08,100 And so we decided to take the leap 743 00:34:08,100 --> 00:34:11,030 got together at the end of the summer and we incorporated 744 00:34:11,030 --> 00:34:14,830 And my colleagues at LCS were-- were much happier 745 00:34:14,830 --> 00:34:16,860 especially those that lived next to me 746 00:34:16,860 --> 00:34:20,160 because I had these thirty people involved in my office 747 00:34:20,160 --> 00:34:22,130 And it was a zoo at LCS while 748 00:34:22,130 --> 00:34:25,130 you know people were writing code and there was this fifty K plan 749 00:34:25,130 --> 00:34:26,230 and it was very exciting for us 750 00:34:26,230 --> 00:34:27,960 but you know disruptive 751 00:34:27,960 --> 00:34:30,700 And once we incorporate and decide we're going to go commercial 752 00:34:30,700 --> 00:34:32,230 then we leave MIT 753 00:34:32,230 --> 00:34:34,460 And we had to get offices outside of MIT 754 00:34:34,460 --> 00:34:37,630 And we had to actually pay for it ourselves 755 00:34:37,630 --> 00:34:40,930 negotiate with MIT over the intellectual property 756 00:34:40,930 --> 00:34:42,730 because we'd done the stuff at MIT 757 00:34:42,730 --> 00:34:44,830 MIT owned the rights to it 758 00:34:44,830 --> 00:34:47,430 And so we had to make a relationship where by MIT took some equity 759 00:34:47,430 --> 00:34:50,530 in the company we formed and then we got the exclusive rights to it 760 00:34:50,530 --> 00:34:53,860 And a process that in the end worked-- worked very well 761 00:34:54,630 --> 00:34:56,400 And we had to buy some servers 762 00:34:56,400 --> 00:35:00,060 you know so instead of just on paper servers we had real servers 763 00:35:00,060 --> 00:35:02,230 We got a real business person 764 00:35:02,230 --> 00:35:06,130 that the original team of twenty people were all MIT folks 765 00:35:06,130 --> 00:35:09,730 In fact they're mostly MIT undergrads that formed the company 766 00:35:09,730 --> 00:35:11,130 So everybody that was involved in the project 767 00:35:11,130 --> 00:35:13,160 we said¡°Look Come on board you can join the company¡± 768 00:35:13,160 --> 00:35:14,600 Everybody got equity 769 00:35:14,600 --> 00:35:17,860 The only rule we made was that you had to finish your degree 770 00:35:17,860 --> 00:35:20,160 So we let students take a year off and they had to comeback 771 00:35:20,160 --> 00:35:22,800 and we let them keep vesting while they came back to finish their degree 772 00:35:22,800 --> 00:35:25,160 because we didn't want to you know screw people up 773 00:35:25,160 --> 00:35:26,130 you know that way 774 00:35:26,130 --> 00:35:30,030 And fortunately everybody did go back eventually get their degrees at MIT 775 00:35:30,030 --> 00:35:33,730 One of things we-- misconception we had early on was 776 00:35:33,730 --> 00:35:36,400 hey once we just say we're forming the company 777 00:35:36,400 --> 00:35:39,100 the VCs are going to send us the checks our way 778 00:35:39,100 --> 00:35:40,400 And it turned out not to be that easy 779 00:35:40,400 --> 00:35:41,400 They would take the check 780 00:35:41,400 --> 00:35:43,560 but actually getting the terms to be 781 00:35:43,560 --> 00:35:46,500 what we considered to be reasonable took a long time 782 00:35:46,500 --> 00:35:49,660 In fact we had to go the angel route to start 783 00:35:49,660 --> 00:35:54,330 And the even the day we were going to close the first round with the VCs 784 00:35:54,330 --> 00:35:56,500 one of the VCs pulled out that day 785 00:35:56,500 --> 00:35:58,460 We had the signed term sheet We had the handshake 786 00:35:58,460 --> 00:36:00,860 Everything was done we're going to sign 787 00:36:00,860 --> 00:36:03,530 And we learned the lesson that you don't deal with a junior partner 788 00:36:03,530 --> 00:36:04,860 Because the senior partner decided 789 00:36:04,860 --> 00:36:06,730 he was going to teach a lesson to the junior partner and said 790 00:36:06,730 --> 00:36:08,600 ¡°This is the kind of deal you don't want to do¡± 791 00:36:08,600 --> 00:36:12,100 And so the day we're going to close he pulled out and we were very glad that 792 00:36:12,100 --> 00:36:13,230 you know Battery stayed with us 793 00:36:13,230 --> 00:36:15,630 Battery Ventures was our lead VC with Todd Dagres 794 00:36:15,630 --> 00:36:16,660 and they stayed with us 795 00:36:16,660 --> 00:36:20,200 They closed the deal without the partner which was hard for them to do 796 00:36:20,200 --> 00:36:22,330 and we didn't get the second partner until much later 797 00:36:22,330 --> 00:36:26,400 Now eventually that-- the VC who pulled out did regret the decision 798 00:36:27,230 --> 00:36:29,830 So we finally closed the first round months later 799 00:36:29,830 --> 00:36:31,460 We were-- we were stressed at that point 800 00:36:31,460 --> 00:36:33,360 so we had spent hundreds of thousands of dollars of our money 801 00:36:33,360 --> 00:36:37,260 and our families money at that point so we were glad to get that done 802 00:36:38,130 --> 00:36:39,330 I'll go briefly through this 803 00:36:39,330 --> 00:36:43,400 As now we're getting to the stage we're outside of MIT 804 00:36:43,400 --> 00:36:46,800 We deployed servers at that time into Exodus data centers 805 00:36:46,800 --> 00:36:49,960 Jonathon Seelig who joined the team from the Sloan school 806 00:36:49,960 --> 00:36:51,230 got Ellen Hancock 807 00:36:51,230 --> 00:36:53,930 who was CEO of Exodus to give us three free months 808 00:36:53,930 --> 00:36:56,600 of bandwidth and colo a decision she regretted 809 00:36:56,600 --> 00:36:58,930 because she spent much of the next several years trying to 810 00:36:58,930 --> 00:37:02,960 kill Akamai as a company because she felt it was competitive 811 00:37:03,300 --> 00:37:07,200 Our first internet test we wanted to actually test to see if we could 812 00:37:07,200 --> 00:37:08,600 you know if the thing could work 813 00:37:08,600 --> 00:37:09,660 So we did it on Exodus 814 00:37:09,660 --> 00:37:11,700 We said¡°We're going to drive our servers to the max¡± 815 00:37:11,700 --> 00:37:12,860 They said ¡°No problem¡± 816 00:37:12,860 --> 00:37:13,930 So we drove them to the max 817 00:37:13,930 --> 00:37:17,930 We took out Exodus and a bunch of the internet went down when we ran that test 818 00:37:17,930 --> 00:37:22,360 That's the last time we took down the internet from the test of our servers 819 00:37:22,360 --> 00:37:27,060 The first live object was an object on a back page of the Disney website 820 00:37:27,060 --> 00:37:32,000 We got about one hit every five minutes across the world on-- on that site 821 00:37:32,000 --> 00:37:35,300 And we had a dramatic ramp up in traffic 822 00:37:35,300 --> 00:37:38,460 The first big break-there were actually two breaks in the same day 823 00:37:38,460 --> 00:37:41,730 came to us-was March Madness in '99 824 00:37:41,730 --> 00:37:44,560 And we'd been trying very hard to get into ESPN 825 00:37:44,560 --> 00:37:47,660 which was a shop that had-- 826 00:37:47,660 --> 00:37:50,660 was viewed as a leader in doing things in the internet-- 827 00:37:50,660 --> 00:37:53,330 in the internet back then 828 00:37:53,330 --> 00:37:56,700 And we'd been trying to get them to test out-- our process 829 00:37:56,700 --> 00:37:59,030 to show we could be faster and more reliable 830 00:37:59,030 --> 00:38:00,460 March Madness came 831 00:38:00,460 --> 00:38:02,430 ESPN went down 832 00:38:02,430 --> 00:38:04,200 And I remember the day they called us up and said 833 00:38:04,200 --> 00:38:06,200 ¡°Hey you know that test you want us to try? 834 00:38:06,200 --> 00:38:07,730 We'd like to try it out¡± 835 00:38:07,730 --> 00:38:08,700 We said ¡°Okay 836 00:38:08,700 --> 00:38:09,560 You know 837 00:38:09,560 --> 00:38:12,900 here's how you set up your URLs so they point to Akamai and we'll deliver them¡± 838 00:38:12,900 --> 00:38:15,100 So they did that in about fifteen minutes and they said 839 00:38:15,100 --> 00:38:16,060 ¡°Okay 840 00:38:16,060 --> 00:38:19,660 By the way can you guys take a thousand hits a second?¡± 841 00:38:19,660 --> 00:38:23,360 And we're doing one hit a minute on the system--the live system 842 00:38:23,360 --> 00:38:25,100 So we said ¡°Sure¡± 843 00:38:26,430 --> 00:38:27,630 You know 844 00:38:27,630 --> 00:38:29,500 yeah And this is where theory helped 845 00:38:29,500 --> 00:38:31,430 because you know normally you build a live system 846 00:38:31,430 --> 00:38:34,130 you scale by a factor of two another factor of two 847 00:38:34,130 --> 00:38:36,860 then you sort of redesign you do another factor of two 848 00:38:36,860 --> 00:38:38,560 And you know the theory had been built to go 849 00:38:38,560 --> 00:38:40,400 to you know millions of hits a second 850 00:38:40,400 --> 00:38:41,830 and it actually worked 851 00:38:41,830 --> 00:38:43,130 And so they went to 1000 hits a second 852 00:38:43,130 --> 00:38:44,400 and they said¡°Okay this is good¡± 853 00:38:44,400 --> 00:38:45,730 And they put the entire site on us 854 00:38:45,730 --> 00:38:48,130 and all of the sudden its 3000 hits a second 855 00:38:48,130 --> 00:38:49,300 So we scaled up 856 00:38:49,300 --> 00:38:53,200 you know five or six orders of magnitude within-- within an hour 857 00:38:53,200 --> 00:38:54,360 And made their site faster 858 00:38:54,360 --> 00:38:57,100 and that was a huge break for us as a company 859 00:38:57,100 --> 00:39:03,730 That very night was the night that the Star Wars trailer was going to be released 860 00:39:03,730 --> 00:39:07,230 Now this is-- this was¡ªthis¡ªthe-- a big event back then 861 00:39:07,230 --> 00:39:09,700 People would love to get the download of the trailer 862 00:39:09,700 --> 00:39:13,060 Now Steve Jobs had paid for the rights to have 863 00:39:13,060 --> 00:39:16,200 the trailer to be exclusively distributed on QuickTime 864 00:39:16,200 --> 00:39:18,930 exclusively on the Apple site because he was trying to 865 00:39:18,930 --> 00:39:23,230 you know make a dent with QuickTime against Real and Windows Media back then 866 00:39:23,230 --> 00:39:27,500 And we didn't know any of that but Entertainment Tonight has a website 867 00:39:27,500 --> 00:39:30,600 They came to us and said¡°Look Would you distribute the trailer for us?¡± 868 00:39:30,600 --> 00:39:32,430 We said¡°Sure¡± 869 00:39:32,430 --> 00:39:34,660 Entertainment Tonight it turns out had a bootleg copy 870 00:39:34,660 --> 00:39:35,800 They weren't supposed to have it 871 00:39:35,800 --> 00:39:38,060 Only you know Apple was supposed to have it 872 00:39:38,060 --> 00:39:42,430 So the trailer was released at nine pm in the evening 873 00:39:42,430 --> 00:39:44,760 And by nine pm and thirty seconds 874 00:39:44,760 --> 00:39:46,060 Apple was down 875 00:39:46,060 --> 00:39:47,900 all the other bootleg sites were down 876 00:39:47,900 --> 00:39:51,960 and Akamai was cranking out Star Wars trailers just fine 877 00:39:51,960 --> 00:39:56,400 So Steve noticed this and wasn't so happy 878 00:39:56,400 --> 00:39:57,860 but we got a lot of great press 879 00:39:57,860 --> 00:39:58,800 because all of the sudden the-- 880 00:39:58,800 --> 00:40:01,900 the web noticed that this-- this thing called Akamai 881 00:40:01,900 --> 00:40:04,560 and their delivering and-- and nobody else was 882 00:40:05,100 --> 00:40:08,200 Shortly there after on April first 883 00:40:09,830 --> 00:40:11,500 we're cranking away in our offices 884 00:40:11,500 --> 00:40:13,400 you know writing code and stuff like that 885 00:40:13,400 --> 00:40:14,760 And our president then 886 00:40:14,760 --> 00:40:17,460 Paul Sagan comes out of his office and said 887 00:40:17,460 --> 00:40:18,030 ¡°Okay 888 00:40:18,030 --> 00:40:22,860 which one of you bozos is pretending to be Steve Jobs trying to buy the company?¡± 889 00:40:22,860 --> 00:40:25,230 And we're looking around and he goes 890 00:40:25,230 --> 00:40:28,960 ¡°Oh no!¡± and he runs back into his office to call back Steve 891 00:40:28,960 --> 00:40:31,800 you know because he-- he hung up on him 892 00:40:31,800 --> 00:40:34,000 And it really was Steve but he thought it was April Fools joke 893 00:40:34,000 --> 00:40:36,000 because you know we always do these April fools jokes 894 00:40:36,000 --> 00:40:37,330 but it was the real thing 895 00:40:37,330 --> 00:40:40,000 Now in the end we-- we did not sell the company to Apple 896 00:40:40,000 --> 00:40:42,360 but Apple became our first strategic investor 897 00:40:42,360 --> 00:40:43,660 And today to this day 898 00:40:43,660 --> 00:40:45,830 we have a great relationship with Apple and we-- we take 899 00:40:45,830 --> 00:40:47,330 you know we handle iTunes 900 00:40:47,330 --> 00:40:50,660 their e-commerce site and-- and pretty much everything they do 901 00:40:51,000 --> 00:40:54,700 Alright let me go quickly through the bubble mania 902 00:40:54,700 --> 00:40:56,930 Very quickly in '99 things exploded 903 00:40:56,930 --> 00:40:59,560 We-- we added on a tremendous executive team 904 00:40:59,560 --> 00:41:03,700 We took on Cisco and Microsoft as strategic investors 905 00:41:03,700 --> 00:41:08,130 We went public very quickly on October twenty eighth 1999 906 00:41:08,130 --> 00:41:11,960 It turned out my birthday was the night we-- we priced the IPO 907 00:41:11,960 --> 00:41:14,960 Before you knew it we had a thirty five billion dollar market cap 908 00:41:14,960 --> 00:41:17,230 on about five million dollars of revenue 909 00:41:17,230 --> 00:41:20,600 And that was excessive even for the bubble 910 00:41:20,600 --> 00:41:22,730 That-- that was-- that was out there 911 00:41:22,730 --> 00:41:24,800 We-- we had bizarre problems 912 00:41:24,800 --> 00:41:29,760 We had these eighteen year old kids who were multi multi multi millionaires 913 00:41:29,760 --> 00:41:32,160 and how do you deal with that help them to adjust to that? 914 00:41:32,160 --> 00:41:34,430 Of course by the time we figured that out we had the reverse problem 915 00:41:34,430 --> 00:41:36,700 How do we help them adjust to the reverse problem? 916 00:41:38,130 --> 00:41:42,000 So we you know AOL bought Time Warner we-- we bought a bunch of companies 917 00:41:42,360 --> 00:41:43,860 We had massive expansion 918 00:41:43,860 --> 00:41:44,760 no budgeting 919 00:41:44,760 --> 00:41:46,530 You just-- you just bought 920 00:41:47,030 --> 00:41:49,800 We grew to 1400 employees overnight 921 00:41:49,800 --> 00:41:52,700 And then the bubble burst 922 00:41:52,700 --> 00:41:54,400 And we had just the reverse process 923 00:41:54,400 --> 00:41:56,400 We-- we downsized to five hundred employees 924 00:41:56,400 --> 00:41:58,360 It was very painful 925 00:41:58,360 --> 00:42:00,230 We had to revamp our entire customer base 926 00:42:00,230 --> 00:42:02,830 Almost all of our customers went broke 927 00:42:03,200 --> 00:42:04,900 It was a nightmare 928 00:42:04,900 --> 00:42:07,530 So we had to get on stable customers enterprises 929 00:42:07,530 --> 00:42:09,460 and the government onto the system 930 00:42:09,460 --> 00:42:11,860 We-- we had massive pricing pressure 931 00:42:11,860 --> 00:42:15,060 You know the cost of bandwidth dropped by a factor of a hundred 932 00:42:15,060 --> 00:42:16,400 over a several year period 933 00:42:16,400 --> 00:42:17,900 A hundred 934 00:42:17,900 --> 00:42:19,060 It's unbelievable 935 00:42:19,060 --> 00:42:21,360 And that put a lot of pressure on us because we were compared 936 00:42:21,360 --> 00:42:22,830 sometimes to bandwidth 937 00:42:22,830 --> 00:42:26,060 A had to put a--we had to put in a budget into the company 938 00:42:26,060 --> 00:42:28,500 focus on cost controls and profitability 939 00:42:28,500 --> 00:42:31,200 We had hundreds of millions of dollars of real estate 940 00:42:31,200 --> 00:42:35,660 we were never going to use and was under water by a factor of five 941 00:42:35,660 --> 00:42:38,960 some of which had been purchased by MIT in the interim 942 00:42:38,960 --> 00:42:41,230 which we eventually worked out a-- a deal on 943 00:42:41,230 --> 00:42:43,030 So we had to get rid of all the real estate 944 00:42:43,030 --> 00:42:45,330 We had to move and we had to refinance our debt 945 00:42:45,330 --> 00:42:47,230 And our market cap went from thirty five billion 946 00:42:47,230 --> 00:42:49,600 to seventy five bil- seventy five million 947 00:42:49,600 --> 00:42:52,630 You know it's remarkable decline 948 00:42:52,630 --> 00:42:54,700 Somehow we survived 949 00:42:54,700 --> 00:42:56,360 And I-- I should mention the worst 950 00:42:56,360 --> 00:42:58,960 you know far worst thing is our cofounder 951 00:42:58,960 --> 00:43:02,900 Danny Lewin was killed on American flight eleven 952 00:43:02,900 --> 00:43:05,330 He was on the flight that crashed into the World Trade Center 953 00:43:05,330 --> 00:43:06,630 Absolutely devastating 954 00:43:06,630 --> 00:43:10,360 you know he was the heart and soul of the company and it was it was horrible 955 00:43:10,360 --> 00:43:14,000 A tremendous individual was killed 956 00:43:14,000 --> 00:43:17,960 So it was very very tough during those years 957 00:43:18,960 --> 00:43:20,660 Fortunately we survived 958 00:43:20,660 --> 00:43:23,300 Today we're a-- basically a profitable company 959 00:43:23,300 --> 00:43:26,100 We will be I think we're projecting to the street-- 960 00:43:26,100 --> 00:43:30,230 the street's projecting twenty or twenty five cents a share this year in 2004 961 00:43:30,230 --> 00:43:33,560 We're generating substantial free cash flow today 962 00:43:33,560 --> 00:43:36,700 Revenues are growing substantially 963 00:43:36,700 --> 00:43:40,400 We're still to this day have a very heavy focus on R and D 964 00:43:40,400 --> 00:43:43,200 We still have a ton of MIT and students from other universities 965 00:43:43,200 --> 00:43:47,760 and professionals from other--other companies working at Akamai at the R and D today 966 00:43:47,760 --> 00:43:51,300 And the market cap has recovered from its lows 967 00:43:51,300 --> 00:43:54,900 You still need a double logarithmic plot to track our stock over time 968 00:43:54,900 --> 00:43:58,230 It's-- it's been pretty wild 969 00:43:58,230 --> 00:44:01,300 Just a snap shot of where we are today and then I'll finish up 970 00:44:01,300 --> 00:44:05,030 We actually have 15000 servers deployed all around the world today 971 00:44:05,030 --> 00:44:09,760 They're in 2400 different locations in 1100 networks 972 00:44:09,760 --> 00:44:12,600 So we haven't got all 15000 yet but we're in 1100 973 00:44:12,600 --> 00:44:15,600 It's a unique concept to this day in the internet 974 00:44:15,600 --> 00:44:18,160 to be in all these different networks 975 00:44:18,160 --> 00:44:21,060 On a daily basis we're over forty gigabits a second 976 00:44:21,060 --> 00:44:24,600 Last week we had a peak of about seventy five gigabits a second 977 00:44:24,600 --> 00:44:28,760 Combination of news events and a software release 978 00:44:28,760 --> 00:44:33,360 And we're serving about fifteen percent of the world's web traffic today 979 00:44:33,360 --> 00:44:35,830 On-- on average basis 980 00:44:35,830 --> 00:44:38,360 We have over 1100 customers 981 00:44:38,360 --> 00:44:43,260 A lot of the biggest and best known sites on the web today are Akamai customers 982 00:44:43,260 --> 00:44:46,260 They use our service to be faster and more reliable 983 00:44:46,260 --> 00:44:48,800 This is a-- a typical sales tool 984 00:44:48,800 --> 00:44:52,560 It shows the performance of the web site without Akamai 985 00:44:52,560 --> 00:44:56,130 And that's a time plot of how long it takes to download a page or an object 986 00:44:56,130 --> 00:45:00,030 so without Akamai is it takes a lot of time up here 987 00:45:00,030 --> 00:45:02,500 and with Akamai it takes a lot less time 988 00:45:02,500 --> 00:45:07,600 And then the rest of that pitch is time-- speed matters on the internet 989 00:45:07,600 --> 00:45:12,430 And we've just done some important case studies for us 990 00:45:12,430 --> 00:45:15,600 as a business with a large hotel customer 991 00:45:15,600 --> 00:45:16,860 They switched to Akamai 992 00:45:16,860 --> 00:45:19,660 We made their site faster and they did a study 993 00:45:19,660 --> 00:45:22,660 that showed that produced thirty million dollars more a year of revenue for them 994 00:45:22,660 --> 00:45:24,430 by increasing the conversion rate 995 00:45:24,430 --> 00:45:28,630 If your web site is faster you get a higher conversion rate 996 00:45:29,360 --> 00:45:31,030 A very large e-commerce provider 997 00:45:31,030 --> 00:45:32,630 which I think we're going to be able to announce 998 00:45:32,630 --> 00:45:34,160 hopefully in the next couple of weeks 999 00:45:34,160 --> 00:45:35,760 the improved reliability 1000 00:45:35,760 --> 00:45:39,300 and the a substantial decrease in their exit rates 1001 00:45:39,300 --> 00:45:40,560 is produced 1002 00:45:40,560 --> 00:45:44,160 They quote a fifty million dollars a year in improved revenue 1003 00:45:44,160 --> 00:45:46,430 And then even the smaller e-commerce sites 1004 00:45:46,430 --> 00:45:48,900 actually those we can get a bigger improvement in conversion rate 1005 00:45:48,900 --> 00:45:51,360 because we make them that much faster 1006 00:45:51,360 --> 00:45:54,560 getting a substantial improvement in their profitability 1007 00:45:54,560 --> 00:45:57,130 by making the website be faster more reliable 1008 00:45:57,130 --> 00:45:59,930 And that was the original business plan 1009 00:45:59,930 --> 00:46:03,460 you know way back when you known coming out of MIT 1010 00:46:04,260 --> 00:46:06,160 A long range vision where are we headed? 1011 00:46:06,160 --> 00:46:08,430 Today we're focused on the internet 1012 00:46:08,430 --> 00:46:11,030 We're moving now into the extranet 1013 00:46:11,030 --> 00:46:13,900 Longer term we'll go into the intranet 1014 00:46:13,900 --> 00:46:15,860 Today we do intranet work for the government 1015 00:46:15,860 --> 00:46:18,300 and a very few enterprises 1016 00:46:18,300 --> 00:46:22,330 And we want to distribute 1017 00:46:22,330 --> 00:46:24,160 We do that content delivery today 1018 00:46:24,160 --> 00:46:25,760 We do page assembly today 1019 00:46:25,760 --> 00:46:27,630 We're starting with business logic 1020 00:46:27,630 --> 00:46:29,360 running the apps on our platform 1021 00:46:29,360 --> 00:46:33,100 and then evolving into web services and eventually to databases 1022 00:46:33,100 --> 00:46:38,800 and combining all these into an on demand global computing utility 1023 00:46:38,800 --> 00:46:40,700 And the marketing is similar to IBM 1024 00:46:40,700 --> 00:46:42,530 and they're our largest reseller 1025 00:46:42,530 --> 00:46:45,360 So it's a very you know synergistic relationship 1026 00:46:45,900 --> 00:46:49,000 Okay let me stop for any questions 1027 00:47:00,130 --> 00:47:01,330 Yes? 1028 00:47:02,360 --> 00:47:09,430 1029 00:47:09,430 --> 00:47:11,460 Right 1030 00:47:11,900 --> 00:47:12,860 What's that? 1031 00:47:12,860 --> 00:47:15,330 The question is how we play in pervasive computing 1032 00:47:15,330 --> 00:47:17,830 and in the wireless space and so forth 1033 00:47:17,830 --> 00:47:20,560 We're working very closely with some of the wireless providers 1034 00:47:20,560 --> 00:47:24,400 you know in terms of providing the bits to them in an efficient way 1035 00:47:24,400 --> 00:47:27,800 Today we are deployed into their networks as far as the IP goes 1036 00:47:27,800 --> 00:47:30,900 As long as it's IP enabled you'll find our servers there 1037 00:47:30,900 --> 00:47:33,830 We will sense what device you have 1038 00:47:33,830 --> 00:47:36,900 Is it a low bandwidth device or a high bandwidth device? 1039 00:47:36,900 --> 00:47:39,360 And automatically give you the right kind of content 1040 00:47:39,360 --> 00:47:42,000 So you get the richer content if you're a high bandwidth device 1041 00:47:42,000 --> 00:47:44,860 You get the text version if you're a low bandwidth device 1042 00:47:44,860 --> 00:47:45,760 So it-- 1043 00:47:45,760 --> 00:47:48,660 it's automatically getting the right version of the page to you 1044 00:47:48,660 --> 00:47:50,730 based on on your bandwidth 1045 00:47:50,730 --> 00:47:54,730 In terms of you know grid computing and computing everywhere 1046 00:47:54,730 --> 00:47:58,130 we're probably the largest example of a grid computer today 1047 00:47:58,130 --> 00:47:59,800 The difference is we own the servers 1048 00:47:59,800 --> 00:48:03,200 It's not on servers in¡ªon-- in your home 1049 00:48:03,200 --> 00:48:05,200 and so it's a more controlled environment 1050 00:48:05,200 --> 00:48:08,630 which-- it allows us to have better performance 1051 00:48:08,630 --> 00:48:12,700 better security than a situation where we don't own the servers 1052 00:48:12,700 --> 00:48:16,360 and they're in people's homes which is much less reliable 1053 00:48:16,930 --> 00:48:18,300 Yeah 1054 00:48:18,300 --> 00:48:24,730 Tom you had a program for small educational institutions 1055 00:48:24,730 --> 00:48:27,230 colleges and junior colleges 1056 00:48:27,230 --> 00:48:31,760 that you would put the Akamai servers in their local network 1057 00:48:31,760 --> 00:48:33,900 if they would allow you to run that part 1058 00:48:33,900 --> 00:48:36,860 while they got all sorts of other benefits 1059 00:48:36,860 --> 00:48:38,760 How did that survive the bubble? 1060 00:48:38,760 --> 00:48:40,130 It worked great 1061 00:48:40,130 --> 00:48:44,400 So this is a program where we will place our servers into ISPs 1062 00:48:44,400 --> 00:48:45,630 that have the end users 1063 00:48:45,630 --> 00:48:48,660 and one class of ISPs is the university networks 1064 00:48:48,660 --> 00:48:50,800 So we have servers in MIT's network 1065 00:48:50,800 --> 00:48:52,460 We have servers in other in 200 other 1066 00:48:52,460 --> 00:48:55,360 I think campuses around the world 1067 00:48:55,360 --> 00:48:59,030 And that enables us to get free bandwidth and colo 1068 00:48:59,030 --> 00:49:00,530 so it saves us on cost 1069 00:49:00,530 --> 00:49:03,100 It enables us to get better performance for our end users 1070 00:49:03,100 --> 00:49:07,000 because we're close to the end users in university systems 1071 00:49:07,000 --> 00:49:09,800 The universities do it because it saves them money 1072 00:49:09,800 --> 00:49:13,930 They don't have to pay the upstream provider to keep downloading the content all the time 1073 00:49:13,930 --> 00:49:16,700 And it makes the performance better on their networks 1074 00:49:16,700 --> 00:49:18,560 because the content is sitting there the-- 1075 00:49:18,560 --> 00:49:19,760 that they need 1076 00:49:19,760 --> 00:49:21,300 And that program has worked very well 1077 00:49:21,300 --> 00:49:24,760 And it's-- that's the kind of thing that helped us through the bubble 1078 00:49:24,760 --> 00:49:29,630 because it helped us on our cost side and maintained our performance advantage 1079 00:49:29,630 --> 00:49:30,660 Yeah 1080 00:49:30,660 --> 00:49:31,830 Yeah 1081 00:49:31,830 --> 00:49:34,060 How did you come up with the name Akamai? 1082 00:49:34,060 --> 00:49:36,300 Danny had a friend who was in marketing 1083 00:49:36,300 --> 00:49:38,500 and we said¡°We're looking for a name¡± 1084 00:49:38,500 --> 00:49:42,400 In fact our early name was Cachet when we were in the fifty K competition 1085 00:49:42,400 --> 00:49:44,630 and then we realized caching was going to die 1086 00:49:44,630 --> 00:49:47,560 as an industry and decided we'd better get a different name 1087 00:49:47,560 --> 00:49:50,130 And he said Hawaiian names will be in 1088 00:49:50,130 --> 00:49:52,960 So we got a Hawaiian- English dictionary 1089 00:49:52,960 --> 00:49:55,330 put fifty Hawaiian words on the board 1090 00:49:55,330 --> 00:49:58,460 and Akamai means clever 1091 00:49:58,460 --> 00:50:01,100 intelligent and cool in Hawaiian 1092 00:50:01,100 --> 00:50:04,860 So in Hawaii you say very Akamai as a compliment 1093 00:50:04,860 --> 00:50:08,130 And we did a search which came up clean so we took the name 1094 00:50:08,130 --> 00:50:09,460 Then a year later 1095 00:50:09,460 --> 00:50:11,500 we realized the search hadn't been done right 1096 00:50:11,500 --> 00:50:14,400 and there was a company called Akamai dot net in Hawaii 1097 00:50:14,400 --> 00:50:17,230 and we had to buy them to preserve the name 1098 00:50:20,530 --> 00:50:21,900 Yeah 1099 00:50:21,900 --> 00:50:24,860 Your distribution model seems very much like 1100 00:50:24,860 --> 00:50:31,330 what we would call in the logistics space a ¡°logistics distribution model¡± 1101 00:50:31,330 --> 00:50:33,130 It's basically an model 1102 00:50:33,130 --> 00:50:37,000 where you have points of distribution at a localized level 1103 00:50:37,000 --> 00:50:39,200 Was any of that thinking ever involved 1104 00:50:39,200 --> 00:50:42,030 when you when started putting these ideas together? 1105 00:50:42,030 --> 00:50:45,360 Not from that space because we didn't know anything about that space 1106 00:50:45,360 --> 00:50:48,500 And I don't even recognize the terms today 1107 00:50:48,500 --> 00:50:50,700 But that the same ideas that are useful in that space 1108 00:50:50,700 --> 00:50:54,360 I'm sure are the ideas that made us-- made this appeal to us 1109 00:50:54,930 --> 00:50:56,000 Yeah 1110 00:50:56,000 --> 00:50:58,400 some idea of your pricing model essentially 1111 00:50:58,400 --> 00:51:02,460 how small a company can for your services? 1112 00:51:02,460 --> 00:51:05,200 Anybody who cares about their web presence 1113 00:51:05,200 --> 00:51:10,000 So you know there's companies that have no traffic on their website at all 1114 00:51:10,000 --> 00:51:12,660 Tiny traffic but they care about their brand 1115 00:51:12,660 --> 00:51:14,960 and they want it to be up and be reliable 1116 00:51:14,960 --> 00:51:16,960 And so then they would be an Akamai customer 1117 00:51:16,960 --> 00:51:20,560 If you don't really care about your web presence and you have tiny local traffic 1118 00:51:20,560 --> 00:51:23,660 well then you probably wouldn't be an Akamai customer 1119 00:51:23,660 --> 00:51:27,000 You know I think our smallest customers would pay 1000 dollars a month 1120 00:51:27,000 --> 00:51:29,100 for the very smallest to start 1121 00:51:29,100 --> 00:51:31,630 so that would be a a barrier right there to a lot of companies 1122 00:51:31,630 --> 00:51:33,000 so I guess 1123 00:51:33,960 --> 00:51:35,360 Yeah 1124 00:51:35,360 --> 00:51:36,830 At the beginning of your talk 1125 00:51:36,830 --> 00:51:40,000 you were talking about security issues of BGP 1126 00:51:40,000 --> 00:51:45,230 but I kind missed any comments 1127 00:51:45,230 --> 00:51:46,100 Right Okay 1128 00:51:46,100 --> 00:51:50,360 so we actually have an offering today that we call BGP Watch 1129 00:51:50,360 --> 00:51:52,260 Because we're in these 1100 networks 1130 00:51:52,260 --> 00:51:54,660 we can all the BGP feeds from those networks 1131 00:51:54,660 --> 00:51:56,360 And so we can actually flag 1132 00:51:56,360 --> 00:52:00,400 if somebody starts advertising an IP address for one of our customers 1133 00:52:00,400 --> 00:52:01,700 And we can tell our customer 1134 00:52:01,700 --> 00:52:04,400 ¡°Hey all of the sudden somebody just took your IP space 1135 00:52:04,400 --> 00:52:05,900 You better go check it out¡± 1136 00:52:05,900 --> 00:52:07,730 Because we can look for changes that way 1137 00:52:07,730 --> 00:52:09,030 And we do this for several 1138 00:52:09,030 --> 00:52:12,260 you know high profile government customers and-- and now in the financial vertical 1139 00:52:12,260 --> 00:52:14,460 obviously some of those customers care very much 1140 00:52:14,460 --> 00:52:16,430 if somebody's taking their IP address 1141 00:52:16,430 --> 00:52:18,760 And this is something they can't get any other way 1142 00:52:18,760 --> 00:52:22,830 You can't get this from MCI because MCI may not even see it 1143 00:52:22,830 --> 00:52:25,030 And it's only because we're in this large footprint 1144 00:52:25,030 --> 00:52:27,060 that we can actually see what's happening 1145 00:52:27,060 --> 00:52:30,160 in the the routing infrastructure at that-- at that level 1146 00:52:30,160 --> 00:52:31,230 Yeah 1147 00:52:32,000 --> 00:52:35,900 Have you learned any new surprising things from your operations data 1148 00:52:35,900 --> 00:52:41,400 that would challenge the underlying theory or stimulate new theoretical work? 1149 00:52:41,800 --> 00:52:45,460 Nothing that would challenge the underlying theory 1150 00:52:45,460 --> 00:52:46,300 You know 1151 00:52:46,300 --> 00:52:48,430 there are interesting things we learned 1152 00:52:48,430 --> 00:52:49,400 For example 1153 00:52:49,400 --> 00:52:52,330 the internet does not satisfy the triangle inequality 1154 00:52:52,330 --> 00:52:55,560 You know he triangle inequality says that if I take two sides of a triangle 1155 00:52:55,560 --> 00:52:58,000 it's longer than taking the direct path 1156 00:52:58,000 --> 00:52:59,630 That's not true on the internet 1157 00:52:59,630 --> 00:53:01,100 And that's you know 1158 00:53:01,100 --> 00:53:04,100 so for example the direct path given by BGP 1159 00:53:04,100 --> 00:53:07,700 the only path often is bad 1160 00:53:07,700 --> 00:53:08,930 And there are better paths 1161 00:53:08,930 --> 00:53:10,930 if only the internet would give them to you 1162 00:53:10,930 --> 00:53:12,930 And so one of the things we did in response to 1163 00:53:12,930 --> 00:53:15,630 that was build an overlay network on top of the internet 1164 00:53:15,630 --> 00:53:17,630 so that when the direct path is not working 1165 00:53:17,630 --> 00:53:20,200 what we'll do is encapsulate the packets at the source 1166 00:53:20,200 --> 00:53:22,360 send them to an intermediate Akamai server 1167 00:53:22,360 --> 00:53:23,360 across BGP 1168 00:53:23,360 --> 00:53:25,830 and then unwrap them and send them to the destination 1169 00:53:25,830 --> 00:53:28,600 thereby taking a a longer path around the triangle 1170 00:53:28,600 --> 00:53:31,660 which in some cases is a lot faster 1171 00:53:31,660 --> 00:53:34,200 In fact today if you're going between the US and Asia 1172 00:53:34,200 --> 00:53:38,800 the direct path is usually a bad path compared to what you can get 1173 00:53:38,800 --> 00:53:39,760 Yeah 1174 00:53:39,760 --> 00:53:43,130 Why do you think that some of the companies like Sockeye that focused 1175 00:53:43,130 --> 00:53:45,530 optimization haven't been that successful? 1176 00:53:45,530 --> 00:53:48,800 Well it's a brutal economic climate out there 1177 00:53:48,800 --> 00:53:52,830 I think ninety nine point something percent of the companies went broke 1178 00:53:52,830 --> 00:53:56,360 And Sockeye had some great ideas and great technology 1179 00:53:56,360 --> 00:53:59,860 They eventually got bought out by Internap 1180 00:53:59,860 --> 00:54:02,800 Internap is selling that kind of technology today 1181 00:54:02,800 --> 00:54:04,930 I think ultimately it will be successful 1182 00:54:04,930 --> 00:54:06,260 but it's-- it's just a-- 1183 00:54:06,260 --> 00:54:10,060 it's a brutal economic climate today and it's extremely hard to get profitable 1184 00:54:10,060 --> 00:54:13,430 And today you got to get profitable or your-- or your gone 1185 00:54:14,130 --> 00:54:15,430 Yeah 1186 00:54:15,430 --> 00:54:16,330 Okay 1187 00:54:16,330 --> 00:54:19,500 Can you say a little bit about the choice between a service based business 1188 00:54:19,500 --> 00:54:21,130 and a product based business? 1189 00:54:21,130 --> 00:54:22,730 Was it an accident of the customers 1190 00:54:22,730 --> 00:54:24,600 that you happened to be able to get at the time 1191 00:54:24,600 --> 00:54:26,830 or was this more of a conscious decision? 1192 00:54:26,830 --> 00:54:27,500 Right 1193 00:54:27,500 --> 00:54:29,960 So with the customer base it had to be a server based-- 1194 00:54:29,960 --> 00:54:32,630 once we switched to the content provider as the customer 1195 00:54:32,630 --> 00:54:34,130 it had to be service based 1196 00:54:34,130 --> 00:54:37,300 You can't-- because we had to have the servers spread all around the world 1197 00:54:37,300 --> 00:54:39,230 Somebody's got to run and operate that 1198 00:54:39,230 --> 00:54:42,860 And so the only way really to sell that is as a service 1199 00:54:42,860 --> 00:54:45,200 When we go behind the firewall 1200 00:54:45,200 --> 00:54:48,060 now you're sort of building a mini Akamai behind the firewall 1201 00:54:48,060 --> 00:54:49,860 You can think of operating that as product 1202 00:54:49,860 --> 00:54:52,160 And we have thought about that in the-- 1203 00:54:52,160 --> 00:54:55,230 the cases where we're doing that today we're actually selling it as a service 1204 00:54:55,230 --> 00:54:57,060 We will come manage it behind your firewall- 1205 00:54:57,060 --> 00:54:59,760 or in the government space- we will come and manage it 1206 00:54:59,760 --> 00:55:03,200 It's-- and that's because it's better for us do it that way 1207 00:55:03,200 --> 00:55:05,030 That's just the way we're set up as a company 1208 00:55:05,030 --> 00:55:06,430 There are competing vendors 1209 00:55:06,430 --> 00:55:10,200 you know Cisco will sell you products and say you can be an Akamai 1210 00:55:10,200 --> 00:55:11,760 just buy my hardware 1211 00:55:11,760 --> 00:55:14,760 and they've have limited success at doing that 1212 00:55:14,760 --> 00:55:16,000 Okay Thanks very much Tom 1213 00:55:16,000 --> 00:55:17,830 That was great 1214 00:55:24,900 --> 00:55:32,300 at two pm which is eleven minutes from now hope you can be there see you everyday