1 00:00:04,019 --> 00:00:06,570 Sean Aylmer: Welcome to the Fear and Greed Daily Interview. I'm Sean 2 00:00:06,570 --> 00:00:09,330 Sean Aylmer: Aylmer. With so much of our lives online now, we 3 00:00:09,330 --> 00:00:12,090 Sean Aylmer: tend not to notice when things are just working as 4 00:00:12,090 --> 00:00:14,520 Sean Aylmer: they should. It's only when something goes wrong, that we 5 00:00:14,520 --> 00:00:16,890 Sean Aylmer: get a glimpse of just how complex the infrastructure is 6 00:00:16,890 --> 00:00:19,889 Sean Aylmer: behind the scenes. In June last year, Fastly, one of 7 00:00:19,890 --> 00:00:22,680 Sean Aylmer: the biggest cloud platform providers in the world suffered an 8 00:00:22,680 --> 00:00:26,130 Sean Aylmer: outage. Many major websites went offline for almost an hour 9 00:00:26,130 --> 00:00:28,170 Sean Aylmer: from the UK government to Amazon and the New York 10 00:00:28,170 --> 00:00:31,380 Sean Aylmer: Times. It just shows how important that infrastructure is. It's 11 00:00:31,380 --> 00:00:34,379 Sean Aylmer: a powerful reminder of how dependent we are on this 12 00:00:34,380 --> 00:00:37,800 Sean Aylmer: technology in the digital age. Fastly learned from the outage 13 00:00:37,800 --> 00:00:40,949 Sean Aylmer: and has continued to grow with customers across 75 markets. 14 00:00:41,190 --> 00:00:44,159 Sean Aylmer: In Australia, that includes Nine and Seven, the big media 15 00:00:44,310 --> 00:00:47,820 Sean Aylmer: groups, the National Rugby League and kogan. com, just a 16 00:00:47,820 --> 00:00:51,510 Sean Aylmer: few examples. Artur Bergman is the founder and chief architect 17 00:00:51,540 --> 00:00:54,720 Sean Aylmer: of Fastly. He's normally based in Denver, Colorado, but is 18 00:00:54,720 --> 00:00:57,420 Sean Aylmer: in Australia for a short visit. Artur, welcome to Fear 19 00:00:57,420 --> 00:00:57,930 Sean Aylmer: and Greed. 20 00:00:58,410 --> 00:01:00,300 Artur Bergman: Thank you very much, Sean. Thanks for having me. 21 00:01:00,690 --> 00:01:03,480 Sean Aylmer: Now, preparing for this interview, I found a stat that 22 00:01:03,480 --> 00:01:06,480 Sean Aylmer: I think is staggering. In fact, it's very difficult to 23 00:01:06,480 --> 00:01:09,060 Sean Aylmer: get your head around. Fastly's network has the ability to 24 00:01:09,060 --> 00:01:12,690 Sean Aylmer: handle more than 1. 4 trillion, with a T- R, 25 00:01:12,720 --> 00:01:16,170 Sean Aylmer: trillion requests per day. That's extraordinary. 26 00:01:16,890 --> 00:01:19,890 Artur Bergman: It's a large number. When you take any of these 27 00:01:19,890 --> 00:01:22,709 Artur Bergman: numbers and do them per day or per year or 28 00:01:22,709 --> 00:01:26,370 Artur Bergman: per month or per hour, they become very large. So 29 00:01:26,370 --> 00:01:28,800 Artur Bergman: we tend to think of it in per second, but 30 00:01:28,800 --> 00:01:30,630 Artur Bergman: even then it's pretty large numbers. 31 00:01:30,630 --> 00:01:33,990 Sean Aylmer: Yeah. So just in layman's terms, so someone like me, 32 00:01:33,990 --> 00:01:37,380 Sean Aylmer: can you take me through exactly what Fastly does? So 33 00:01:37,380 --> 00:01:39,330 Sean Aylmer: when I interact with the New York Times website, for 34 00:01:39,330 --> 00:01:41,910 Sean Aylmer: example, what part does Fastly play? 35 00:01:42,150 --> 00:01:46,680 Artur Bergman: So the fundamental physics problem that we help our customers 36 00:01:46,680 --> 00:01:52,230 Artur Bergman: overcome is speed of light. We all think speed of 37 00:01:52,230 --> 00:01:55,980 Artur Bergman: light is very fast and it is fast, but speed of 38 00:01:55,980 --> 00:02:00,630 Artur Bergman: light from Australia to the East Coast of the US 39 00:02:00,660 --> 00:02:05,760 Artur Bergman: roundtrip time is a couple of hundred milliseconds, which is 40 00:02:05,760 --> 00:02:08,160 Artur Bergman: not that fast when you're waiting for something to load 41 00:02:08,280 --> 00:02:10,800 Artur Bergman: or when you want to interact with a website. It 42 00:02:10,800 --> 00:02:14,639 Artur Bergman: gets even slower by the fact that the fiber lines 43 00:02:14,639 --> 00:02:19,260 Artur Bergman: under the ocean don't go perfectly straight between us and 44 00:02:19,560 --> 00:02:23,010 Artur Bergman: where the servers are. So what Fastly does as an 45 00:02:23,010 --> 00:02:27,510 Artur Bergman: edge cloud platform is we have servers all around the 46 00:02:27,510 --> 00:02:30,780 Artur Bergman: world. We have some in Sydney and Melbourne, for example. 47 00:02:31,139 --> 00:02:32,970 Artur Bergman: And so when you go and interact with New York 48 00:02:32,970 --> 00:02:36,330 Artur Bergman: Times, you actually end up going to the servers that 49 00:02:36,330 --> 00:02:40,980 Artur Bergman: are closest to you. And those servers will store a 50 00:02:40,980 --> 00:02:45,540 Artur Bergman: copy of the content that you're accessing and will execute customer 51 00:02:45,540 --> 00:02:50,760 Artur Bergman: code close to you to assemble your online experience, whatever 52 00:02:50,820 --> 00:02:53,370 Artur Bergman: content you're trying to view or whatever interaction you are 53 00:02:53,370 --> 00:02:55,860 Artur Bergman: trying to have so that most of it can be 54 00:02:55,860 --> 00:02:59,790 Artur Bergman: done locally, a couple of 10, 20 milliseconds away from you 55 00:02:59,790 --> 00:03:01,020 Artur Bergman: instead of couple of hundred. 56 00:03:01,800 --> 00:03:03,750 Sean Aylmer: Okay. You have just answered a question I've wondered about. 57 00:03:03,750 --> 00:03:05,609 Sean Aylmer: I used to live with a guy who laid fiber 58 00:03:05,610 --> 00:03:09,480 Sean Aylmer: optic cable for a living and he'd get on boats 59 00:03:09,480 --> 00:03:11,190 Sean Aylmer: and they'd lay out the fiber optic cable. He was 60 00:03:11,190 --> 00:03:13,950 Sean Aylmer: an engineer. And what I could never work out though, 61 00:03:14,220 --> 00:03:17,340 Sean Aylmer: was if we're trying to get the East Coast, that's 62 00:03:17,340 --> 00:03:20,700 Sean Aylmer: a long way. And fiber optic cable doesn't just go 63 00:03:20,700 --> 00:03:22,560 Sean Aylmer: straight. I mean, it's on the bed of the ocean. 64 00:03:22,560 --> 00:03:24,600 Sean Aylmer: So obviously it's going up and down. But what you're 65 00:03:24,600 --> 00:03:26,190 Sean Aylmer: actually saying is when I'm going to New York Times, 66 00:03:26,190 --> 00:03:27,960 Sean Aylmer: I'm not going to California or I'm not going to 67 00:03:27,960 --> 00:03:29,011 Sean Aylmer: New York. I'm probably going to Sydney. 68 00:03:29,011 --> 00:03:29,640 Artur Bergman: When you're going to New York Times, yes, you're going to Sydney. 69 00:03:33,270 --> 00:03:37,650 Sean Aylmer: Wow. Quite incredible. Now consumers expect instant access to data, but 70 00:03:37,740 --> 00:03:40,530 Sean Aylmer: they also expect their data to be customized. And for 71 00:03:40,530 --> 00:03:43,200 Sean Aylmer: it to be secure, obviously is a really big part 72 00:03:43,200 --> 00:03:47,220 Sean Aylmer: of it. How much does that slow down this transmission speed? 73 00:03:48,060 --> 00:03:51,900 Artur Bergman: So there is a inherent trade- off there, which is, if I 74 00:03:51,900 --> 00:03:55,140 Artur Bergman: want to make it personalized to you, I have to 75 00:03:55,410 --> 00:03:59,790 Artur Bergman: run code to make that happen. And traditionally I would 76 00:03:59,790 --> 00:04:02,130 Artur Bergman: have to run that code in my own data center. 77 00:04:03,030 --> 00:04:06,540 Artur Bergman: So either I built a lot of central data centers 78 00:04:06,600 --> 00:04:09,360 Artur Bergman: all around the world, which is very expensive and hard 79 00:04:09,360 --> 00:04:14,160 Artur Bergman: to maintain, or I would just have to decide, should 80 00:04:14,220 --> 00:04:17,909 Artur Bergman: I personalize this and encourage a latency cost? Or should 81 00:04:17,910 --> 00:04:20,970 Artur Bergman: I serve it as fast as possible? And what's really 82 00:04:20,970 --> 00:04:24,180 Artur Bergman: cool about the edge cloud platform and our edge compute 83 00:04:24,480 --> 00:04:28,800 Artur Bergman: is that it allows our customers to do that personalization 84 00:04:28,800 --> 00:04:31,830 Artur Bergman: on the edge, whether that is just the pay wall 85 00:04:32,070 --> 00:04:35,220 Artur Bergman: and all the things for you, just accessing the site, 86 00:04:35,490 --> 00:04:41,070 Artur Bergman: personalizing the content, personalizing recommendations. So because you have compute 87 00:04:41,160 --> 00:04:43,469 Artur Bergman: at the edge, so the edge being Sydney or Melbourne 88 00:04:43,470 --> 00:04:47,250 Artur Bergman: or Perth, or any of the 100 locations we have in 89 00:04:47,250 --> 00:04:50,190 Artur Bergman: the world, you can take that content that is a 90 00:04:50,190 --> 00:04:53,940 Artur Bergman: copy of it, it's cased, and you can work on 91 00:04:53,940 --> 00:04:57,060 Artur Bergman: that content, personalize it and only give out what you 92 00:04:57,060 --> 00:05:01,169 Artur Bergman: need locally. It's historically been a very difficult trade- off, 93 00:05:01,529 --> 00:05:05,400 Artur Bergman: but we're helping our customers solve that so they don't 94 00:05:05,400 --> 00:05:06,810 Artur Bergman: have to do that trade- off. 95 00:05:07,170 --> 00:05:09,359 Sean Aylmer: Talking about it, it makes me wonder, can technology actually get 96 00:05:09,360 --> 00:05:12,450 Sean Aylmer: faster? Is the speed of light the fastest we can get 97 00:05:12,450 --> 00:05:14,310 Sean Aylmer: to? So I accept the trade- offs that we're talking 98 00:05:14,310 --> 00:05:18,540 Sean Aylmer: about here, but can technology get faster than fiber optic cable? 99 00:05:18,810 --> 00:05:21,660 Artur Bergman: It can technically, but we don't know how to go 100 00:05:21,660 --> 00:05:25,680 Artur Bergman: faster than speed of light. And so there's a ratio 101 00:05:25,770 --> 00:05:29,760 Artur Bergman: of how fast light travels in fiber optic versus vacuum. 102 00:05:30,060 --> 00:05:32,520 Artur Bergman: And there is a difference there, but you're still talking 103 00:05:32,640 --> 00:05:35,909 Artur Bergman: a fairly substantial delay when you try to go around 104 00:05:35,910 --> 00:05:36,360 Artur Bergman: the world. 105 00:05:36,720 --> 00:05:38,970 Sean Aylmer: Stay with me, Artur. We'll be back in a minute. 106 00:05:38,970 --> 00:05:47,820 Sean Aylmer: My guest this morning is Artur Bergman, the founder and chief 107 00:05:47,820 --> 00:05:52,409 Sean Aylmer: architect of Fastly. Now I mentioned the Fastly outage, which 108 00:05:52,410 --> 00:05:54,990 Sean Aylmer: happened just over a year ago. It's a remarkable story 109 00:05:55,230 --> 00:05:58,110 Sean Aylmer: because of what it triggered. Can you just tell me 110 00:05:58,110 --> 00:06:01,080 Sean Aylmer: again in layman's terms, what happened, but then what you've 111 00:06:01,080 --> 00:06:04,410 Sean Aylmer: learned from that, the lessons out of that episode for 112 00:06:04,410 --> 00:06:05,490 Sean Aylmer: you as an organization? 113 00:06:08,490 --> 00:06:14,160 Artur Bergman: We have covered pretty in detail, what happened and it's 114 00:06:17,460 --> 00:06:22,830 Artur Bergman: like any complex failure. It is when multiple things go 115 00:06:22,830 --> 00:06:27,630 Artur Bergman: wrong and the compensating controls do not act as they 116 00:06:27,630 --> 00:06:30,989 Artur Bergman: should. So in this case, it was a piece of 117 00:06:30,990 --> 00:06:34,260 Artur Bergman: code that had been released fully tested. We have a 118 00:06:34,260 --> 00:06:37,860 Artur Bergman: very controlled way of releasing new code to our network. 119 00:06:38,339 --> 00:06:41,820 Artur Bergman: That takes a very long period of time. This code 120 00:06:41,820 --> 00:06:45,900 Artur Bergman: had been released and that passed all the tests. And 121 00:06:45,900 --> 00:06:50,100 Artur Bergman: then one of our customers configured a service in a 122 00:06:50,100 --> 00:06:53,789 Artur Bergman: way that no one had ever configured a service or 123 00:06:53,790 --> 00:06:57,000 Artur Bergman: at least in the previous 10 years. And then that 124 00:06:57,810 --> 00:07:02,190 Artur Bergman: triggered a crash. And it was a very large customer 125 00:07:02,190 --> 00:07:04,800 Artur Bergman: with a lot of traffic. So, that crash was very wide. 126 00:07:05,220 --> 00:07:08,430 Artur Bergman: So there was multiple steps in the process here that 127 00:07:08,430 --> 00:07:13,410 Artur Bergman: kind of missed catching this event. And we have a trade- 128 00:07:13,410 --> 00:07:16,500 Artur Bergman: off decision we, as a company have made, which is 129 00:07:16,500 --> 00:07:20,940 Artur Bergman: when our servers end up in a situation where we 130 00:07:20,940 --> 00:07:24,720 Artur Bergman: don't know fully what is going on, we have a 131 00:07:24,720 --> 00:07:27,600 Artur Bergman: choice. We have choice of trying to serve whatever content 132 00:07:27,600 --> 00:07:31,680 Artur Bergman: we can, or we have a choice to stop everything. 133 00:07:32,970 --> 00:07:36,060 Artur Bergman: Because the risk, if you don't stop is we don't 134 00:07:36,060 --> 00:07:39,720 Artur Bergman: actually know what's being served, which would be extremely bad. 135 00:07:39,720 --> 00:07:42,540 Artur Bergman: So our policy is we stop. What we really have 136 00:07:42,540 --> 00:07:49,230 Artur Bergman: learned from this is a focus on automated recovery. We've 137 00:07:49,890 --> 00:07:51,990 Artur Bergman: done a very good job in the past of trying 138 00:07:51,990 --> 00:07:54,690 Artur Bergman: to prevent these things, but ultimately if something happens, we 139 00:07:54,690 --> 00:07:59,550 Artur Bergman: need to recover very fast. And we have also accelerated 140 00:07:59,820 --> 00:08:04,110 Artur Bergman: our switch to using our edge compute product for all 141 00:08:04,110 --> 00:08:07,590 Artur Bergman: of our services, which has built- in protections so that when 142 00:08:07,590 --> 00:08:11,550 Artur Bergman: something like this happens and for customers who now have 143 00:08:11,550 --> 00:08:14,310 Artur Bergman: started to migrate, this is what will happen, only the 144 00:08:14,310 --> 00:08:18,960 Artur Bergman: request that caused the crash would have been stopped. But 145 00:08:18,960 --> 00:08:22,110 Artur Bergman: it's really that combo of that, and then the combo 146 00:08:22,110 --> 00:08:26,520 Artur Bergman: of a much more solid, underlying modern platform, which we 147 00:08:26,520 --> 00:08:30,300 Artur Bergman: think all online services over time need to move to 148 00:08:30,480 --> 00:08:34,800 Artur Bergman: using strict isolation. It's kind of like, if you remember 149 00:08:35,520 --> 00:08:38,340 Artur Bergman: a long time ago, when a tab in your browser 150 00:08:38,340 --> 00:08:41,550 Artur Bergman: crashed, your entire browser would crash. And then we got 151 00:08:41,550 --> 00:08:43,679 Artur Bergman: to a point where if one tab crashes, only that 152 00:08:43,679 --> 00:08:44,610 Artur Bergman: tab crashes. 153 00:08:45,000 --> 00:08:46,170 Sean Aylmer: Yep. Great analogy. 154 00:08:46,620 --> 00:08:49,740 Artur Bergman: Yeah. That's what we're doing on the server side. 155 00:08:50,070 --> 00:08:54,990 Sean Aylmer: Yeah. Okay. Okay. What about for individual companies needing to protect 156 00:08:54,990 --> 00:08:58,800 Sean Aylmer: their online experience? I mean increasingly digital platforms are the 157 00:08:58,800 --> 00:09:01,410 Sean Aylmer: only form of interaction between a business and consumer. So 158 00:09:01,410 --> 00:09:04,890 Sean Aylmer: it's really important. What should they be doing to make 159 00:09:04,890 --> 00:09:08,160 Sean Aylmer: sure the experience is great? They're going to be using 160 00:09:08,160 --> 00:09:10,530 Sean Aylmer: services like Fastly, but from their side, what should they 161 00:09:10,530 --> 00:09:10,949 Sean Aylmer: be doing? 162 00:09:11,400 --> 00:09:16,319 Artur Bergman: When you're designing your online presence, I think a lot 163 00:09:16,320 --> 00:09:21,179 Artur Bergman: of companies think of the functionality first and then they 164 00:09:21,179 --> 00:09:28,260 Artur Bergman: think of security, reliability, and performance after. And by switching 165 00:09:28,590 --> 00:09:32,250 Artur Bergman: that fault process and accepting that if your site is 166 00:09:32,250 --> 00:09:36,750 Artur Bergman: slow, people aren't going to use it. If your site 167 00:09:36,750 --> 00:09:40,050 Artur Bergman: isn't secure, people aren't going to use it. And certainly 168 00:09:40,050 --> 00:09:41,670 Artur Bergman: if it's not available, people are not going to use 169 00:09:41,670 --> 00:09:45,270 Artur Bergman: it. So making it just a very clear requirement that 170 00:09:45,270 --> 00:09:48,630 Artur Bergman: for this product or for this experience to be successful, it 171 00:09:48,630 --> 00:09:51,870 Artur Bergman: has to be performant, it has to be reliable, and 172 00:09:51,870 --> 00:09:54,990 Artur Bergman: it has to be secure. And that should be a high- 173 00:09:54,990 --> 00:09:58,770 Artur Bergman: level business KPI, not something that you think about after 174 00:09:58,770 --> 00:09:59,340 Artur Bergman: the fact. 175 00:09:59,640 --> 00:10:03,840 Sean Aylmer: Yep. Yep. Okay. My final question for you, Artur, is 176 00:10:04,140 --> 00:10:06,990 Sean Aylmer: where are we going with this? I have learned a 177 00:10:06,990 --> 00:10:09,960 Sean Aylmer: lot in the last 10 to 12 minutes, talking to you about 178 00:10:10,320 --> 00:10:14,910 Sean Aylmer: the speed of light and how Fastly operates its businesses. 179 00:10:15,179 --> 00:10:18,059 Sean Aylmer: Where will we be in 10 years though? Some of 180 00:10:18,059 --> 00:10:21,570 Sean Aylmer: these latencies that you're talking about and personalizations and that, 181 00:10:21,870 --> 00:10:23,280 Sean Aylmer: are we going to get a lot more of that 182 00:10:23,429 --> 00:10:27,599 Sean Aylmer: and somehow the technology will allow faster times? Where do 183 00:10:27,600 --> 00:10:28,020 Sean Aylmer: we get to? 184 00:10:28,470 --> 00:10:33,120 Artur Bergman: The underlying constant of speed of light, it's the limiting factor. And 185 00:10:33,690 --> 00:10:37,200 Artur Bergman: we just are finding new technologies to work around that 186 00:10:37,200 --> 00:10:42,090 Artur Bergman: and make it magically seem like we're more connected than 187 00:10:42,240 --> 00:10:45,450 Artur Bergman: physics sets the limit of. And I think that's really 188 00:10:45,870 --> 00:10:49,830 Artur Bergman: what's occurring is as you build new applications, as you 189 00:10:49,830 --> 00:10:54,360 Artur Bergman: build new tools, having both the business and engineering take 190 00:10:54,360 --> 00:10:58,500 Artur Bergman: into account, and this becomes extremely important when you're trying 191 00:10:58,500 --> 00:11:02,939 Artur Bergman: to become an international business. If all I'm serving is 192 00:11:02,940 --> 00:11:06,660 Artur Bergman: my customers in Sydney, I don't really need to care, 193 00:11:06,660 --> 00:11:09,900 Artur Bergman: right? But, most businesses don't just want to serve one 194 00:11:09,900 --> 00:11:13,350 Artur Bergman: city. They want to go to multiple cities and the 195 00:11:13,350 --> 00:11:17,910 Artur Bergman: rest of the world. And that's where I think building 196 00:11:18,270 --> 00:11:22,530 Artur Bergman: your new tools, building your new platforms with these kind 197 00:11:22,530 --> 00:11:26,640 Artur Bergman: of constraints in mind, coupled with the need for performance 198 00:11:26,640 --> 00:11:30,209 Artur Bergman: and security and reliability, anywhere in the world, if you do 199 00:11:30,210 --> 00:11:33,240 Artur Bergman: it right, you have a product that will scale to anywhere. 200 00:11:33,510 --> 00:11:36,360 Sean Aylmer: Yeah. Fascinating. Artur, thank you very much for talking to 201 00:11:36,360 --> 00:11:37,080 Sean Aylmer: Fear and Greed. 202 00:11:37,890 --> 00:11:39,360 Artur Bergman: Thank you so much for having me, Sean. 203 00:11:40,050 --> 00:11:43,560 Sean Aylmer: That was Artur Bergman, founder and chief architect of Fastly. 204 00:11:43,830 --> 00:11:45,990 Sean Aylmer: This is the Fear and Greed Daily Interview. Remember this 205 00:11:45,990 --> 00:11:48,660 Sean Aylmer: information is general in nature and you should seek professional 206 00:11:48,660 --> 00:11:51,960 Sean Aylmer: advice before making any investment decisions. Join us every morning 207 00:11:51,960 --> 00:11:54,000 Sean Aylmer: for the full episode of Fear and Greed, Australia's most 208 00:11:54,000 --> 00:11:57,209 Sean Aylmer: popular business podcast. I'm Sean Aylmer. Enjoy your day.