How the Cloud Reshaped the Internet
Paris Marx is joined by Dwayne Monroe to discuss what it’s like to work in a data center, how the cloud came to hold a dominant position, and the consequences of its control by companies like Amazon, Microsoft, and Google.
Dwayne Monroe is a cloud technologist and aspiring Marxist theorist of technology, with twenty years of experience architecting large-scale computational systems. Follow Dwayne on Twitter at @cloudquistador.
Support the show
Venture capitalists aren’t funding critical analysis of the tech industry — that’s why the show relies on listener support.
Become a supporter on Patreon to ensure the show can keep promoting critical tech perspectives. That will also get you access to the Discord chat, a shoutout on the show, some stickers, and more!
- Dwayne wrote about cloud computing for Logic Magazine. He’s also written about a public cloud and the metaverse on his blog.
- Amazon’s cloud infrastructure in the eastern United States experienced a major outage in December 2021.
- Residents in various parts of the world have been questioning the logic of building data centers, including in the United States and New Zealand.
- In July, the network of Canadian telecom giant Rogers went down, leaving millions without service.
Paris Marx: Dwayne, welcome to Tech Won’t Save Us.
Dwayne Monroe: Thank you very much.
PM: I’m really looking forward to chatting with you. I’ve been wanting to do an episode on the cloud, because I think it is this really important infrastructure that we maybe sometimes underestimate its importance. And I’ve been paying attention to your work for a while. And you had this fantastic article in Logic Magazine recently, where you look broadly at the cloud and the impacts that it has had, but then also look specifically and refer back to your personal experience in working in data centers and working with these technologies, implementing cloud solutions, all these sorts of things. And so I think that also gives us an important perspective when we are considering the impact that this technology and these systems have had. And so to start our conversation, I was hoping that we could talk a bit about that more physical, more personal experience that you have in dealing with these things, since you’ve worked in these data centers. So can you start by telling us a little bit about what it’s actually like in those spaces?
DM: Of course. So first of all, let’s define what we mean by data center. A data center is a warehouse, essentially, where computers, known as servers — which are like people’s home computers, their laptops, and what have you, but just more powerful, more memory, more processing ability and so forth — are concentrated together to provide a service. And other equipment is collected with those servers, such as storage equipment and cooling equipment, which is absolutely unnecessary, because computers generate a tremendous amount of heat, because they’re quite wasteful, actually. There was a recent paper, I think, from Intel that talked about how wasteful computation actually is — we’re not really getting all of the computation for our bucks, so to speak. But anyway, when I started my career in the 1990s, one of the things I had to do was, as we said, rack and stack the servers. What does that mean? Well, that means that, in the case of a large company, such as a pharmaceutical firm that had a massive data center, the size of some people’s suburban cul-de-sacs or bigger. Servers would come in from Dell or Compaq before Dell, and IBM and even Sun Microsystems back in the day.
And my job was to take that physical equipment and put it into racks and wire it up for a network connection to the corporate network, and to access to the internet eventually, and make sure that everything was properly balanced and work with the cooling engineers to make sure that the cooling balance was good. And so it was a very physical activity. And often you would hurt yourself, as I mentioned in the Logic Magazine article. You would cut your hands; your knees would hurt because you were on your knees having to rack these things. And the rooms are quite cold, and because as I said, they have to be. And so there was nothing ethereal at all about working with these systems. And today, when we talk about cloud, we’re still talking about data centers like Microsoft and Amazon and Google — this is what they’ve got. They’ve got data centers, and are just renting out access to the computing and the database capacity. But behind the scenes, there are people doing today what I was doing earlier in my career, different equipment, but it’s the exact same principle. So it was a very industrial, I would say, in a way, kind of an industrial task, even though the way that it’s presented is as if it’s science fiction, or magic or something.
PM: I thought it was really fascinating to get that description of the data centers and of what that work was like. There are a few points that stood out to me. I think the first was when you described how there were injuries that came of this work, how it was kind of like a factory, in a sense, a factory of computing, and how the space was really cold. When we think about Amazon as one of the main cloud providers with the Amazon Web Services (AWS), it almost seemed similar, but different in a way to its warehouses — this industrial scale, really large. People are getting injuries, obviously, it’s a very different type of work. But then, in the case of the data centers, it has to be kept really cold for this computing. But then it seemed like in the warehouses, Amazon keeps it particularly warm for the robots that are in the warehouses there.
DM: I’m not quite sure what they’re about there, what the actual technical explanation for that is. I’ve always suspected that it’s because they don’t want to have the expense of actually properly cooling such large warehouse spaces for the people who work there. And that some of the robots that they use have probably a higher tolerance for the heat. Or perhaps they even have cooling rooms for them. It wouldn’t shock me at all if they did something like that — the very limited capacity robots, because they simply move things from A to B. They can’t actually do the warehouse work. Try as they might to make that make that happen, it’s not actually possible. But yes, in the concentration of computing power in a data center, cooling is absolutely necessary, because otherwise, the machines would burn themselves out.
And fire was of course a concern, because that could occur. And earlier, there was a method for suppressing fire called Halon gas, which I do not believe is in use anymore. It’s been replaced by another gas that’s not quite as injurious to human life. Because with Halon dispensed, it would take all the oxygen. I believe it would convert as nitrogen. I could be wrong about that — it’s been a number of years. But I think it would actually remove the oxygen from the air therefore making fire impossible. And one particular evening, I think, and of course, probably was a Friday — because everything seemed to happen on Friday, when you want to go out — there was a false fire alarm and the gas started descending. And my manager just grabbed my shirt, and dragged me out of the room as these doors. It was like a movie with the lights, the klaxons and these giant doors closing and the gas just descending. And so as I said, nothing ethereal about that at all, because we quite literally could have died.
PM: No way!
DM: In a data center — just asphyxiated.
PM: And for most people using the internet, even unknowingly interacting with these data centers, and now the cloud computing, which are just bigger data centers, really, you would never know that something like that actually happens in those spaces.
DM: That’s precisely right. And that there are people looking at screens and so forth, monitoring. There’s an area of work called Site Reliability Engineering, and one of its tasks is correctly architecting a solution or a system. So it’s resilient against a failure, but also monitoring. So there are people just burning hours, spending hours of their lives just making sure these systems continue to function. And although I no longer do that, I moved my career into the more abstract area of it. But I know that’s happening. But when I connect to say, and as your interface and I create solutions, I know that behind the scenes, if I spin something up that, it’s happening in a data center, and there are people actually behind that curtain. Even the term cloud is quite hilarious, because it’s obviously designed as a marketing hook to make you think that there’s just this glowing thing that’s providing this computational service. So yes, it was boring work, but also, in some senses, quite interesting.
And I also had a chance to see the scale at which computation can occur because the consulting firm that I worked for, in those days, their focus was on Fortune 500 and 1000 firms. So the data centers that I worked in were really just quite big. I mean, one of them I actually helped to build it from from the ground up. Well, not the building, but turning it from a giant, empty space into a data center. And we would ride our bikes around before all the racks were in place, and all the servers were delivered in pallets, giant trucks, we woukld just ride our bikes around because there was so much space. And before the the Leibert, which is the name of a company that produces these massive cooling units, before they were in place. And it was just normal air conditioning, we would just ride our bikes around. But within a few months, it was filled with hundreds, if not thousands, of computers just humming away.
PM: It’s quite different from what you’re describing, but one time, I helped set up a Best Buy store. And I think it’s just weird to see these large, warehouse-y spaces — obviously, a Best Buy store would not be as large as a data center that you’re talking about — but just empty, and then slowly being filled in with the various, I don’t know, objects that fill it up with stuff.
DM: All sorts of things! That is the background, and then that is what informs my view. And I think that that also was what prevents me from being, as I’ve said, in other conversations from being fooled, this applies even to things that seem esoteric, machine learning, or what have you, or GPT-3, or DALL-E, or what have you. I’m seeing data centers, others see: Oh, artificial general intelligence is on the way! But what I’m seeing are computers in racks generating heat. And that’s why my question is always: Okay, fine, the mathematics is interesting. You’re not producing intelligence, but the mathematics is interesting, as even Chomsky would say. But tell me what’s going on in that data center, because that tells me what the actual cost is, what the power consumption is, the industrial process that informs what you are presenting as this ethereal, again, science fiction-y kind of thing that you’ve built.
PM: I want to get to some of those questions. But first, I do want to just talk a little bit more about your experience in seeing this transition play out. And in particular, you talked in your article about how these companies, some of them, at least, had no cloud policies, at least in the in the early days, as cloud was emerging.
DM: That’s right.
PM: Maybe this is an obvious question. But why would that be? Why would some companies not want to use these cloud solutions?
DM: So imagine that you are a vice president, or a CIO at a major corporation in 1995. Or better yet, because it wasn’t an option then, 2005, 2006, when Amazon Web Services first became available. Now you have just overseen the spending of millions of dollars for maybe a refresh cycle. And by refresh cycle, that’s when a collection of servers have reached the end of their life. And you’ve received a budget from the board. And they said: Okay, here’s 8 million; buy new servers, and you’ve done your dog and pony show to the board explaining the value, which was all funny money back then.Tthere was no way to really, in any accurate sense, determine what it would be. But this is what CIOs would do. And so you’ve done all that, and this is your bread and butter. This is the basis of your power within the organization. And someone comes along and says: Well, you know what? Goodbye to all that. We can actually relocate that whole function to somewhere else. Now, companies had colocation, and I’d worked with companies that did colocation and what that means is rather than staging the servers in a data center that you owned, you would rent space from another company such as RackSpace, which is a major player in that industry, but you still had the ability to visit, to see it. And so there was a physical connection to the investment.
So at the executive level, that was the concern — the loss of power. But at the level of the technologist, there also was a concern about a loss of relevance and therefore a loss of jobs. So at the executive level, as I said, that the nature of the business or the way that the business have been run for, my God by then, probably since long before I started, in the 60s or 70s, there was a way that information technology was managed. And this was a challenge to that. And so that’s why companies would have a no-cloud policy. In addition, security teams often would be very concerned, often the very legitimate concerns is: Well, how do we know that the system has been properly secured? Their own systems probably weren’t properly secured, but at least there was the potential to properly secure it. And in those early days before the major cloud vendors had their ISO certifications and so forth. They’re very security certifications, there was no way to know. So there were some legitimate concerns, some power concerns, some political concerns, and some job concerns was a mixture of things there at that time.
PM: I appreciate you outlining that, and particularly going through the different layers of it. Because I think that makes a lot of sense. You can understand why people would be reluctant to embrace something new, that’s going to change the dynamics of power change the way that things work, especially when you’re used to a certain way of doing it.
DM: I’ll give you a very brief anecdote to tell you how — when I first became a consultant focused entirely on what we call cloud transformation — in a meeting with CIO with members of the IT team, various people in a conference room all together on those days long before COVID. People said to me quite openly, but what happens to my job? And the CIO, in this particular client says: Yeah, but I don’t have any control. They were just quite upfront about it, what their concerns were, and my job as a consultant, it was to assuage their fears, and massage them, which as the technology stack grew more sophisticated and more capable, became easier. And also, it was less of a sales pitch, because they are capable, and you can do things with them. But I have many memories of these conversations. And some of these conversations still occur today. But the political winds have changed in the industry. And now CIOs are getting fired for being in the way, whereas before, they might have had problems with their jobs if they said: Oh, yes, of course, let’s do that.
PM: I think it makes a lot of sense. Now, in the article, you also describe how cloud does come into these organizations. And as you’re talking about you were dealing with a lot of larger organizations. So I guess my question probably has two parts. First of all, how does then cloud enter into these larger organizations like the ones that you’re talking about? And is that kind of generalized across the board? Or would that work differently for smaller companies? We see all the startups that form that are kind of taking advantage of the cloud, in the early 2010s, and things like that. And so I guess they probably have a different relationship to it. Would that be right?
That is correct. So for larger enterprises, as we call them, and I’ll just use an example. Let’s say you’re a massive paperclip company, and you had a massive — maybe I’ll use the word massive, one more time for good measure — and you had a large investment in your own data center, or perhaps, colocation cloud came in, as I said in the article, through the back door. Because people needed to solve problems and the process of procuring equipment and getting it approved. And all of that was often quite slow. So different departments would say: We need this. And so people would say: Well, I can’t really get it to you now, because I haven’t got approval, but I could deploy it on AWS. Or I could deploy it on Azure, or Google Cloud Platform. And in the beginning, it was it was only really Amazon Web Services in the early days, because they were the only company actually offering anything serious in this area. So for enterprises to be coming through the back door, typically through development teams or engineering teams that were trying to solve problems. That was how I was introduced to it, as I said in the article. For a few years later down the road, smaller firms that needed a lot of computing power, but didn’t have the capital could begin to spin up solutions, create solutions on the cloud and punch above their weight from a computational perspective.
And also mid-range firms. I helped a friend with a medical billing company once, like a $500 million company, so large enough, but not global enterprise scale. They had a data center that probably was the size of somebody’s apartment building, maybe a little less than that. For them, cloud was a way of saying: Well, this is not our core function. We’re spending millions that we could be spending someplace else. We’re spending millions on computation and high salaries for people. And what if we could simply offload that function to these cloud providers? Now, there’s a misunderstanding because it turns out that you still need people who were decently paid. And that probably came as a shock to many organizations, who thought: Oh, I’ll just give this away. But at least they no longer had to worry about ordering servers and having data centers of any type or doing a colocation contract or any of that.
Today, however, I would say the mindspace has changed dramatically, as I said, and so CIOs of larger organizations are expected to have a cloud strategy. The startups, of course, the use case is still exactly what it was in the past. Many of the so-called machine learning or AI startups that are spinning up all kinds of nonsense. 10 years ago, they would not have been able to do that, because they would have had to have made major investments in computing equipment they could not have afforded unless they got like a billion dollars in Round A funding or something. So that’s an impact. And the mid-range firms, the value proposition, as they say, for them is absolutely being able to have the service that they need without the overhead of having to manage the actual hardware.
PM: I think you can see why that happens. And I guess I wonder are there many large companies today that would still have their own data centers, and by extension, does the economy of scale offered by something like AWS, or Google Cloud, kind of reduce the cost of I guess managing your data or whatnot, if you move it on to one of these cloud platforms, rather than having your own data center?
DM: The answer to that is no. In the beginning, the sales pitch was if you do this, by this I mean get rid of your data center and move everything to a public cloud. And there’s a distinction made between what’s called a private cloud, which something that you would build within your own data center using a technology such as OpenShift, for example, which is a platform that allows you to create an abstraction layer to the computing power that you have, in your own data center. The sales pitch originally was if you move everything to public cloud, it’ll be instantly cheaper. But this is not true. So I’ll give you an example: one of my clients in the past had 3000 servers in their own data center hosting databases — they were a rather large fashion house. And so they had a pretty massive investment in databases, both for tracking what’s happening in various stores around the globe, and various other things that they did. They were convinced by one of the major cloud vendors to move everything to their infrastructure. It was Amazon as a matter of fact, and actually their bills went up. And the reason is because with cloud infrastructure, you are billed for everything. Everything is a trackable metric that generates a cost.
Now on the positive side, you can track those costs, you can see them in real time, which is new. That’s something that wasn’t possible — isn’t possible — with on premises, as we call it, data center resources. You’d have to, what we used to call, sweat the assets. You’d buy it, and then you’d have to just use that equipment for five to 10 years, because the company made the investment. With cloud, you can see that: Oh, I’m spending 3 million a month, for example, and I see exactly how that’s being generated. So not instantly cheaper, but it is possible to be cheaper, through a proper architecture. But it is also possible to move at greater speed. And this gets me back to what I was mentioning a little while ago about these AI startups. So they can just spin up using the services. I’ll now use Azure Cognitive Services because I know that quite well, which offers language models and text to speech and an all these various services as application programming interfaces or APIs. That is the same as services that you can plug into and just consume as a utility. So this gives them the ability to operate in a way that they could not have done earlier. And so for them, for companies in that space, it’s possible to not necessarily be cheaper, but because they don’t have the overhead, they can offer a service more rapidly. And then reach the idea is reach profitability at a faster pace.
PM: So a number of benefits: move faster, you can take this part of your business that is not core to your business away, have someone else manage it, who theoretically is better at doing that. And you can also have access to these tools and these applications that you wouldn’t have just on your own programs.
DM: That’s right. Even earlier in my career, only the very largest organizations — I consulted over a decade ago at a power utility in Philadelphia. And that company, of course, had access to an Everest of capital. And so they wanted to build on premises, a system that allowed them to determine what a power consumption forecast would be for the Eastern Seaboard. Because Philadelphia is part of, of course, the grid and generates quite a bit of power, actually. And I think it’s called PJM. And so they have that money. But if we started a little company, we wouldn’t have had that cash. But today, we could start up a little company and say: Well, we’ll do this analysis for you, and then rather than having to invest in all this extremely expensive equipment and having to house it somewhere in a data center, we can simply maybe hire some developers, and then those developers can plug into the API, as I mentioned, of the analytical services made possible by the cloud provider.
PM: So I guess I’m starting to plan the Tech Won’t Save Us AI startup right now [both laugh]. But, I think that gives us really good insight into how these services rolled out on the end of these companies and experience from the perspective of people who are in these companies. I would like to zoom out. And I think it’ll allow us to talk a bit more about the AI startups and things like that, as well. But I want to kind of go back to the moment that cloud services emerge. What are the companies? We’ve already talked about some of them, obviously, but I guess just to kind of lay it out for people, what are the companies that are that are leading this? And what is the moment when these technologies, these solutions really start to be made? Why does this happen?
DM: So the story is told that Amazon was the first and that Amazon had developed a method for abstracting computer services internally, so that they would be able to fulfill their own needs — for their growing need — for computational power, and that they had developed a method internally for offering that as a service. So that if your line of business needed computational power, whatever purpose, that rather than you having to order your own servers and then have the requisition added to the Amazon’s data centers, and so forth, that you would simply have your developers to build against this computational infrastructure that was commonly available. And at some point, it’s said that Bezos, who knows, but someone within Amazon said this would actually be something that would be interesting to customers. The first service that they offered was called — is called because it’s still available — S3, which as I recall, stands for Simple Storage Service. So storage located somewhere else. Now, it may seem strange — like why is that important? Well, the reason that’s important and would be compelling to people, is that one of the chief problems for applications is where do I store my data? Obviously, you have databases, but there are other types of things or objects that you may need to store images and other types of things. Files, let’s say for example, if you’re pulling up a record could be stored within a database. But it also could be stored in a separate storage container, electronic storage container for those who aren’t familiar.
So Amazon, by offering this service, created something that became instantly attractive to developers. Because they could say: I can develop my application. Now at the time, it might still be on premises mostly at the computational layer. But I can point it towards this storage that was elastic, that is to say it could expand with my needs. So rather than having to go over and over and over again to the till, or hat in hand to management and say: Oh, well, I thought we only needed $2 million worth of storage. But turns out we need $5 million worth of storage, you could simply continue to use this S3. So it was a clever move on Amazon’s part because that did become the hook — so that when you added other services such as elastic compute, or as it’s now EC2 which are virtual machines, which just means that it’s just a virtualized version of the very servers that I was talking about, at the operating system level, which is what controls the computer. When you offered higher level services, now you have the foundation, which was storage. So now people were hooked because they saw the value. And as we mentioned earlier, the developers got excited. And engineering people also got excited, and they could see all the solutions they could build, again, without having to once again go to management and say: Oh, well, I miscalculated.
Which happened all the time. You would always get it wrong, you would always do a forecast of what you needed. And you were always wrong, because either you overdid it and so everyone’s mad at you, because you said: Oh, we need $10 million. And well, most of the CPUs on these servers are not even at 1% utilization — why the hell we spend $10 million? Or you didn’t do enough. And this comes up in my article in Logic, of course, where this particular company, the book ordering system, fell down every year, because they just wasn’t enough in heartbreak capacity to meet the need, and people yelling and working late and all sorts of nonsense going on. So the solution to that problem was to have the storage and then to have those higher level services. So Amazon was first, Microsoft had what they called Windows Azure service, which Steve Ballmer, who was the CEO at the time, man who some geeks might know as a dude who likes to scream a lot. One of the early managers and executives at Microsoft, during the gates days. Windows Azure, which it was just Microsoft Windows Server, and they weren’t quite serious about it.
But then under the leadership of Satya Nadella, they became extremely serious about it. And then it became a contender. Google came in third with its Google Cloud Compute, and they decided to turn their attention and their resources to building a competing infrastructure to what Amazon and Microsoft had done. And those are the three major players. Now there’s also Alibaba in China, there is Oracle, which has, it pains me to say, because Larry Ellison is a sinister character. But Oracle’s database service is actually useful to people. IBM has a cloud service, which everyone makes fun of. So there are various companies in the space, but the three major players from North America would be in order of dominance: Amazon, Microsoft and Google.
PM: And would it mainly be that Amazon pioneered this because they saw it was something that they needed internally, and then decided to spin it off as more of a public facing business as well, or rather, it’s still within Amazon, but to offer it as a different service beyond their e-commerce platform?
DM: I think so.
PM: But then, I guess, when it comes to Google and Microsoft, were they basically just seeing that and saying: Oh, wow, this is working for Amazon, we need to copy it, or was there a different sort of set of motivations there?
DM: Yes and no. So with Amazon, yes, they definitely benefited from the the first adopter effect, or the first one out of the gate effect. And then they did, to be fair, actually offer a number of services, as we said, that were quite compelling, like the service primitives — as we geeks say — that were quite compelling. Microsoft fumbled for a bit in the space, particularly when Ballmer’s vision was just to have access to versions of Windows servers in the cloud. But then when they realized under Nadella, that well, actually, we have a pretty powerful story, as they say, to tell — which is Microsoft still dominates in the corporate data center. They still dominate on the corporate desktop. And companies, like here in Europe, I think it’s 70%.
Companies have a very deep investment in Microsoft. And so what Microsoft strategy became was: What can we do to our story? Microsoft’s story would be, their marketing hook would be, you already know us from your on-premises, your data center. Now we can we can take that into public cloud. And that’s why their products such as more recent versions of Windows server are “cloud aware,” so that they can immediately take advantage of in the case of Azure, it’s a Blob storage, which is the equivalent to Amazon’s S3. And so they totally retooled their offerings to be “cloud aware.” So that became their selling point for corporations.
And Google, focused on what they consider to be the sweet spot, which is analytics. So although they offer some of the other services that their competitors do, it’s services such as BigQuery, and other database and analytics focused offerings that have become, I think the the most compelling use case for many organizations, when it comes to Google. And for some of the largest customers I’ve had, there’s been a multi-cloud strategy, but not one in which a single application spans multiple clouds, but one in which it’s the right tool for the right job. So, Azure, Microsoft Azure is used for core services, Amazon is used for customer facing stuff, and GCP is used for analytics. And those bills are insane. But yes, I’ve seen that with some of the largest enterprises, and actually designed solutions that use that strategy.
PM: I guess then you have some degree of — security is not the right word — but I guess redundancy and stuff, facing using different clouds.
DM: To some degree. Although there’s a fallacy that it’s a method of preventing one of the things that people talk about in this industry, and what is lock in. But it’s not, because you’re using different platforms for different things. And although there are technologies such as containers, which is a like a lightweight version of a server, and into other technologies, such as Kubernetes, which allow you to, you know, to orchestrate containers and move them from place to place. These things exist, but nothing is ever as simple or as easy as it’s advertised to be. And I’ve witnessed a number of projects, in which someone swoops in and says: Oh, you just go from cloud to cloud. And it never, it never ever works out actually, the way that people present it as as working out.
PM: That makes perfect sense! I do want to build on this line of questioning. But I remembered one other question that I had about what we were talking about earlier. So I want to briefly return to that. And that question is, you were talking about, and you were writing in the article about what it was like to work at these data centers pre-cloud and in the transition to cloud. As these data centers become larger, as there’s more and more focus on these cloud data centers, these massive data centers controlled by Amazon and Microsoft and Google, how does that change the work in these data centers? When you go from, say, a data center that’s controlling the work for one Fortune 500, or Fortune 1000 company to working in one of these data centers that are part of this cloud infrastructure that are these massive, massive sites for computing? What is the difference there?
DM: Obviously, the scale is much greater. The largest data center I worked in was for a pharmaceutical firm. And they had more than one. They had multiples, and they’re actually spread across the United States and one in Madrid, which I had the pleasure of visiting. But even so, not as large in scale as what these three companies have done. They have spent many billions of dollars year-on-year to expand their data center footprint. If anyone can just Google, for example, AWS data center, and you’ll see this big map of Amazon, or I’m sorry, or Azure, or GCP. And you’ll see maps across the globe, showing where these data centers approximately are. And so for a site reliability engineer, and for a server engineer in these cloud data centers, of course, things are much more tightly controlled. And as I said, they’re much larger. And they don’t use computing elements, the way that we did in the data centers in the 90s and early 2000s. Whereas we would get an entire server with, you know, fans and memory and so forth, from Dell, or from another company, and then rack and wire it up. They are using computing elements, distributed amongst — they still use racks, of course — but they use competing elements in a proprietary configuration, to save space, and to be able to cram as much computing power into the spaces as possible.
All the principles remain exactly the same. However, it’s just a change in scale. And the fact that of course, they are serving many, many, many customers around the globe, whereas, for example, the farmer that I worked for, mostly it was them. They owned it all. And all the computing power was for their purposes — no one else’s purposes. Whereas these cloud providers, they become kind of the all roads lead to Rome. There’s multiple rooms, but all roads are leading to them. And so they have a scale of complexity, a scale of concern about security, both physical and digital security, concerns about power. It is not an exaggeration to say they’ll be orders of magnitude greater than — even though it’s the same principles — but orders of magnitude greater than what I was doing in the 90s and early 2000s. Which is why they are known, the category of computation that they fill is called a hyperscale.
PM: And how does this hyperscale, how does this movement to cloud computing to these major data centers controlled by Amazon, Microsoft and Google benefit these corporations and enhance their power over computing over the internet over what we do online?
DM: Well, anyone who is old enough to remember what was called the Microsoft tax, has the ability to already suss this out. And what was the Microsoft tax? Well, there was the idea that Microsoft, through their monopoly or near monopoly over desktop computing, in corporate environments, and in other aspects of corporate compute in the 90s, and into the early 2000s, that their power and their wealth was derived by essentially exacting a tax on almost every business, whether you were a mom-and-pop with a laptop in your store, or a midsize firm, like the one that the medical billing company I mentioned, or you were a massive pharmaceutical firm. In each case, you had to find a way of playing ball with Microsoft. Because they had you in a critical way, because you needed this computational power in order to to run your business. So take that reality, and just turn it to 11,000. Because now, these companies not only have say, control at the operating system level, or the desktop level — the entire computational infrastructure is taken into their hands. So they become a part a critical part of global infrastructure in a way that is actually quite new. And one of the things I say to people is, how long do you think this can really go on? How long do you think, say, an Amazon can run these the systems? 100? Do you really think it’ll be 100 years, and they’ll still have these computers running? Get out of here.
So putting this in corporate hands, obviously, is a problem, because more and more companies and therefore individuals are subject to the profit, incentives and concerns, and indeed, to some extent, the whims. Now, of course, they do provide a service. And so it’s not as if they’re just going to turn it off. Because, I don’t know, Nadella had a bad morning or something. That’s not going to happen. But it’s more a matter of such level of power should not be in private hands. I have nothing against the idea of abstracting computational power and offering it as a service. That’s fantastic. And that’s also not new — mainframe’s offering this, Jesus, when I was a baby. So that’s not new. What’s new is the concentration of this in private hands. So that I think that is something that the each of us should be mindful of. Here in Europe, there are individuals who are who are calling for the creation of truly public cloud, that is to say, abstracted computation, database and storage offered as a public good. But it’s difficult. It’s difficult, because there is lobbying by of course, these majors. It’s difficult, because it’s expensive. And there are people who believe that somehow these companies have magical powers. And that no one can do what they did, which, of course, is nonsense. They didn’t come from the future — they’re just people like you and me. But I think that’s really what’s needed. I think, that in order to get a control, get control over power usage, to get control over the politics of this, because it’s not tenable to have three North American companies have so much sway over the direction of computation.
PM: I think you’re completely right. What you’re describing really brings to mind a number of articles that I’ve been reading recently, like concerns in the United States about the location of these data centers, and the kind of power that these companies can exert when they choose where to locate them, the subsidies that they get for them, but also like, you know, the the power that they require and the water and what they extract from these communities where they’re located. I was reading a story recently in New Zealand, where there are proposals for a bunch of new data centers, I believe in the north of the country. And the communities were grappling with like: Is this really what we want? We’re already struggling to hit our renewable energy goals. And if these companies come in, they’re going to use up a ton of power, and it’s going to make it more difficult for those to achieve. I remember reading in Africa as well, concerns about do we want these major companies to come in and build their data centers? And then kind of control the African intranet rather than controlling it ourselves. So there are all these concerns with allowing these major companies that have this degree of power over, really the internet.
DM: And people have criticized the PRC (People’s Republic of China) for being quite strict on this. And there are many reasons to criticize Beijing, as there was many reasons to criticize any government anywhere. However, strategically, it makes perfect sense to say no to Facebook. It makes perfect sense to say no to these companies dominating your computational infrastructure. Another reason why this concerns me is fragility, as I mentioned. So for example, there was an outage of a portion of AWS infrastructure. I think it was the ‘us-east-1’ region. And for those who don’t know, because of the scale, the infrastructure of cloud providers is divided into regions. So the us-east-1, which I believe is based in Virginia, in the United States, there was an outage of a portion of the infrastructure, which brought down the services provided by a number of private entities, schools, I believe maybe a hospital was impacted. Now, in the past, all of these organizations would have had their own computational infrastructure, which that’s wasteful in its own way, the way we did it in the 90s and early 2000s. However, you could see that cracks are starting to show, I think, in the site reliability. As we mentioned earlier of these hyperscalars, because operating at a hyperscale begins to expose the infrastructure to the effects of complexity theory.
Because if you’re getting more and more complex, there’s more and more compute, there’s more and more need to maintain, there’s more surface area for things to go wrong. And it’s inevitable that things will go wrong in various places, having an impact upon all the companies, some of which are providing critical services. I believe there may even have been some 911 outages — just all sorts of things that that happen when even a portion of the infrastructure of these providers goes down. And there’s no recourse other than the private recourse of: Well, I signed a service agreement with you. These are very capitalist sanctions, but not like a public sanction. How do we make sure that this computational infrastructure is serving our needs and is redundant? And if it costs more, we as a populace might say: Okay, that’s fine, because I don’t want my hospital to go down when I’m getting my X-ray. They should be able to scan that x-ray and put it someplace, put the data someplace, for example. So that I think, is the primary concern that I certainly have, which is the concentration of power, political power, and socio-technical power, because computation, as I’ve said many times is a command control technology. And if you are in command of it, then you have command and control over a large portion of a society.
PM: As I was rereading your article, in preparation for the interview, I was thinking of that as well. Because I don’t know if you saw the news recently, but one of Canada’s major telecom providers went down the other day, it was down for 15 hours, and phone service, cable service, internet service was lost to hospitals to 911. Everything. It’s basically one of the dominant players. And one of the responses to that was that we need more competition in telecom. But then I was like, but there’s, there’s kind of a fundamental infrastructure here, an infrastructure that kind of links the whole country. And is competition on infrastructure really going to make a difference? Or do you just need to build more redundancy into the system? Because I don’t think you’re going to get much competition on a national telecom infrastructure like that. And so I guess when you’re thinking of cloud is the response that we need to de-center it and have less centralization of these infrastructures, or to have more redundancy built into it so that these sorts of outages don’t have the same degree of impacts if it all goes down in this way?
DM: That’s a technical matter. And certainly those companies do whatever they can, because obviously they don’t want it to go down. It’s just happening because failure is an emergent property of complexity. And it’s just inevitable that it would happen. So I don’t fault them for the failure. But it’s the failure of government to control — to oversee what this is. I’m not aware of any oversight of any sort really, of cloud services. I’m not. I’m sure there are some federal statutes in the United States and certainly, here in Europe, there’s GDPR. And then California, there’s the California Data Protection Law. So there’s some measures of control and there are privacy, security certifications and so forth, they have to go through. But what I mean is active oversight of this critical infrastructure — the way that there was decades ago, many decades ago — oversight of critical infrastructure, such as you mentioned, telecom. The very reason AT&T was broken up. Now, that didn’t turn out the way that people hoped.
But the very reason that happened was because there was concern, and I think it might have been under Sherman Antitrust, but there was concerned about as this private entity, Bell, having so much control over something that was so important to the nation. And in those days, of course, there still was the vestige of what I would call the World War Two era federal command and control bureaucracy before the neoliberal era. And there was still people who remembered when the government would just say to Chrysler: I need 10,000 tanks. Well, I don’t want to build tanks, I don’t care if you don’t want to build tanks, you’re building tanks. And so there was still a vestige of that kind of person in government. Those people are all gone now, of course — they’ve retired or they’re dead. And now we have the total neoliberalization. of government’s approach, which means that well, these companies must be fantastic. They’re all geniuses. Isn’t Elon Musk a genius? The lack of questioning, the lack of doubt. And the assumption that there’s some special sauce going on in Silicon Valley, which, those of us who who monitor the space carefully note that theory’s not, of course.
PM: It’s a bit beyond our conversation. But that’s basically how I feel but the telecom issue in Canada as well. I think focusing on competition is the wrong way to look at it. And what we really need is just greater role for the government and actually ensuring that these technologies and these systems serve the public, instead of just these these major telecom companies. So I totally agree. You were touching on the AI piece of this, and how cloud computing allowed kind of the explosion of all these kinds of AI companies promising that these thinking computers were going to change so many aspects of how we live for the better and what have you. And we’ve seen little of that I think actually play out? What do you make of — because you’ve written a bit about this on your on your blog —what do you make of the way that these AI companies, these artificial, and companies have been framing themselves in this technology over the past decade or so? And how the cloud has kind of allowed that to be possible? And how it has misled the public on what these technologies can actually do?
DM: What you have is a convergence of several factors. One being the false narrative of the creation of thinking machines — the other being the false narrative that the systems that are being deployed are flawless, as happens with, say, facial recognition, or companies coming out and saying, We can detect who’s cheating on a test, if they’re taking a test through Zoom or Teams or what have you. So there’s the hype, but then there’s the ability to deploy systems, as we mentioned earlier, at scale, that’s made possible by a cloud provider. So the cloud providers are either purposely or not an enabler or a force multiplier of the proliferation of the systems which are built on hype. Then there’s the gullibility of not just government, but corporations, because the atmosphere that we’re in, in which people believe that these systems are flawless, and that we are further along than we actually are, sort of acts as a cultural sales pitch for these companies. And also the fact that governments and companies are looking to trim their head counts. So you can say: Oh, well, I don’t need to have so many people examining documents or working at the DMV; I can, I can just use facial recognition or whatever cobbled together system they come up with.
For example, you’ve written quite a bit about these so-called automated supermarkets and of its self-checkout and what a bit of nonsense that actually is. But it’s presented as if somehow it’s futuristic. Somehow it’s better. No one can explain other than, oh, I’m in and out in a minute. Well, the extra five minutes it took you to to stand in line with a person. Is that really so terrible? What are you doing? Are you protecting the Earth from an incoming asteroid? What are you doing that so important? Like: Oh, my God, I had to stand in line behind someone! Oh, okay — it’s fine. So it’s a combination of these factors moving together, to enable these companies to to get funding from VCs, who see money to be made. Whether it’s long term or not, they don’t really care, they’ll get a return on their investment. And as I said, the gullibility of the customer base, and the hype cycle that is acting as, as I said, a cultural sales pitch for this. But were it not for the cloud providers, there would be far, far fewer of these organizations. There would be some, but far fewer, because the VCs would have to make the investment in the hardware, and the engineering capacity, and the data center real estate. It’s a different type of investment — a different type of commitment.
It’s not as lightweight or lightweight seeming as: Hey, Paris and Dwayne have this great idea, and then we got this guy from Caltech. And he’s a genius somehow, and our facial recognition software will be able to predict whether this person is going to commit a crime in five months. There are people who believe that right. And this is a little bit off our topic, but I also blame, what I will call, the effects of science fiction. I’m a science fiction fan, and have been since I was a kid, but I think that some people’s brains have been rotted, not because science fiction exists, but because there’s no counterbalancing narrative to: It’s nice that that you saw that nice story about a supercomputer took over the world. But here’s why that that can’t happen. And the very companies that we’re talking about take advantage of this, as I say, culturally, presented a sales pitch for what they’re for what they’re selling, and perhaps even bolster it to some extent. So that’s one of the reasons why we have this massive problem, this big problem with the proliferation of these AI systems that are having negative impacts on people’s lives. Because it’s far too easy, far too easy to just spin up things, and then just sell them into the world.
PM: Yeah, I completely agree with you on the sci-fi piece as well. The tech people in Silicon Valley who are making use of that, who are trying to realize it, but then also the broader public who has consumed it for so long and just kind of believes like: Okay, this is the direction that things should go. And it really allows people to kind of take their guard down when thinking about these ideas or approaching these narratives that we’re receiving from the tech industry,
DM: Or also not just take their guard down, but sometimes be fatalistic, and I’ll give you an example. Every time you see a video of a Boston Dynamics Big Dog, right, and not the Big Dog — Spot and the way that they film it is as if it’s autonomously moving around the streets of New York patrolling. Everyone thinks about that same “Black Mirror” episode, “Metalhead” I believe it was. Everyone thinks about that episode. And so the fiction of that episode then causes people who should know better in many cases to think: My God, you know, it’s here, this autonomous death machine is here. And then you’re at pains to say: No, if the camera were to pull back a bit, you would see the person with the remote control, actually directing the Spot unit. I think there was Eric Adams in New York, there was a picture of, which I think Boston Dynamics probably wasn’t happy about, because you could see him walking up the stairs with the Spot directing it. It’s a drone — it’s a remote control device.
The engineering piece, of course, is the legs. That is an engineering accomplishment. But it’s not possible to build a truly autonomous system that can perceive and say: Oh, I’m going to kill that guy, and not that guy. No, that cannot be done. But the science fiction milieus has a convinced people that this is happening. So not just the dreams, but also the nightmares are assumed to be 100% possible. And this puts those of us who challenge this in an interesting and a difficult spot. Because not only do we have to challenge the the narrative of Silicon Valley, we also have to help our friends snap out of it, and say: No, this isn’t you’re not dealing with the thing that you saw in that movie or that show. It’s still a threat, but for a different reason than the reason that you’re thinking.
PM: This is this is one of the reasons why I really liked Alex Rivera’s “Sleep Dealer” as well, because it shows the exploited humans behind the supposedly acting machines, in a way that I think I think a lot don’t. But on that question about futures, I feel like one of the elements of the futures that some of these tech companies are selling us now are really dependent on this massive scale up in computing power that they assume would want to happen and that they want to realize like they want more and more of this computing for more computing to govern more of our lives effectively. You can see this I think, with streaming video, streaming media have that has required a lot more computing power in recent years. And how there’s now a push for streaming video games, which are expected to account for a lot more. And Facebook’s vision for this metaverse that would require this massive expansion in computing power. What do you make of these visions by these tech companies for these kinds of futures or to have us even more subjected to all this computing power and the increased requirements for it?
DM: They are absolutely trying. However, I think that this infrastructure will collapse upon itself, of its own weight. Meta for example is certainly realizing that the task that they set for themselves is extraordinarily complex. And you may recall, and I did mention this in I think, a piece that I wrote on my blog, that an Intel executive did an actually quite good, technical breakdown of why the metaverse as described by Meta is not achievable with current technology, and the amount of of new computer equipment and computing equipment of a different type that would be required, and the scale that would be required. And all that someone has to do is to ask themselves a couple of basic questions like the Oculus Rift — what was the the maximum number of people who were using that? Probably a couple of 100,000 people. And then you think about all that was required, there was a bandwidth and hardware and manufacturing to make that happen. And now you’re talking about expanding that to a billion people? I mean, come on. Obviously, you’re talking about something that has never been achieved, and certainly would not be achieved by a completely wacky company. And not just wacky, but sinister — like Meta.
But then there’s another part of this, which is that computers are built or are created based upon industrial process. There are minerals; there’s silicone; there is metal; there’s plastic; there are people mining things; there is a supply chain for computation that is completely obscured. And so what you’re talking about is ramping up the extractive industries that support it. You’re talking about racking up or increasing the power consumption through real estate consumed. Tthat is not sustainable, and is not the least bit achievable. And I think that 50 years from now, whatever shape the world is in, one thing that we will not be talking about is some of the technologies that these companies are deploying, because they will have totally fallen apart. And then we’ll be the digital versions of ancient Roman cities.
PM: [Laughs]. Except the servers will be shut down, so we won’t be able to even go find them.
DM: The servers will be shut down — exactly! [both laugh]. It’d just be ghost towns, and we know this was coming. This day is definitely coming.
PM: Absolutely. And I really appreciate you drawing attention to the material piece of this quest of the company, because I think it’s so important and doesn’t get nearly the attention that it deserves. I appreciate you giving me your time to have this conversation. I want to end with a final question. You’ve talked about the need for government to actually step in and have some regulation on cloud computing. And actually look at what kind of impacts this is having, and ensuring that it actually serves the public. And you’ve also talked about the need for potentially of a public cloud. So I wonder, what do you think our response to this growth of cloud computing, and it’s controlled by, largely, these three major companies? What is our response to that? And how do we ensure, whatever form cloud computing takes, that it’s a form that actually benefits the public rather than just the kind of goals of these major corporations?
DM: I wrote a piece about this a little while ago called “Public Cloud As a Public Good,” in which I lay out at least the beginning, I think, of an idea. I think the first step would be, well — not to put too fine a point on it — nationalization of the some of this infrastructure. I think that would be the first step. And then the skilling up of ordinary citizens who are interested. There are people who are neglected and in neglected places who would be excited, and very engaged to participate in a community computational infrastructure. I know that I, as a kid growing up in West Philly, which wasn’t nearly as wild as Will Smith would have you believe. But there were ups and downs. And it was exciting when I when I got interested in in computing — it expanded my world. And, I got to meet all sorts of interesting people through computation. And I think back to that time, in which my young self, want did I want to do? Solve problems for people using this machinery — that’s what it’s really for. And so I think that’s what’s required is is the nationalization and the creation of parallel systems from the ground up are built with people’s needs in mind.
One idea that I lay out in the piece that I wrote on my blog was that the effects of climate change are global, but they also manifest differently in different places. And so we need a method for ingesting data, analyzing data, determining what these local effects are, in order to determine what mitigation strategies should be. And computation should play a role in that but computation that is a public good. One of my inspirations for this is the the Cybersyn project that Stafford Beer oversaw in Allende’s Chile. I think that provides an also a great model for what computation could be, particularly now that we have orders of magnitude more computing power. This future has been intercepted by the dictates the demands of Silicon Valley. My position is that the future has been halted by the tech industry. It’s presenting itself as giving us the future but what it’s really doing is preventing the future. And so in order for us to actually craft a future for computation, we have to circumvent or somehow deal with them. That’s pretty much my feeling on that!
PM: I completely agree with you. And I think that is a great place to leave it as well. Dwayne, thank you so much for taking the time. I’ve really enjoyed this conversation. Thank you!
DM: Thank you so much!