a16z Podcast - a16z Podcast: An Open Source Business Model That Works

Episode Date: July 31, 2014

Open source software has permeated practically every nook of the software world. The biggest companies and largest-scale systems all lean heavily on open source code. Yet, with the exception of Red Hat, no one has built a great business on top of open source software. That’s because what companies should be selling isn’t necessarily the software, or even support, says a16z’s Peter Levine, but a service that leverages open source. One example of such a service is DigitalOcean, which has built a cloud environment tailored to the needs of developers. DigitalOcean CEO and cofcounder Ben Uretsky joins Levine from his New York City HQ for a discussion about open source-as-a-service, how companies of all sizes should think about leveraging open source, and whether we’ll start to see a slew of specialized clouds geared toward different verticals.

Hello, this is Peter from Andresen Horowitz, and I'm, it's my honor to be joined by Ben, who's the CEO of DigitalOcean. Ben, how you doing? Doing really well. Thanks, Peter, for having me on. Good. Yeah, great. So one of the things, Ben, that I've spoken quite a bit about recently in sort of the data center of the future and the future of, enterprise technology is the notion of open source economics finally working and specifically kind of contrasting open source being used as a service like what digital ocean does versus let's say the red hat model of you know on-prem free software and then charging for for you know at the most charging for support and maintenance as the economic model and my belief is that there's really a very beneficial model by having open source as a service, open and

proprietary source as a service, again, as a way to really drive economic value out of open source. And I'm just sort of curious. I mean, you guys are, you've built a company on this. So I'm curious on your thoughts and how you guys think about that. Yeah, definitely. So we built our cloud on top of KVM's virtualization technology. which in and of itself is an open-source project, and we deliver a ton more value than just the virtualization engine.

So we've created an entire interface and API to interact with the infrastructure that we're providing. And we've also put thousands of servers into a data center. Now, all of this is powered by KVM, and we've turned it into a infrastructure as a server, provider. Without, you know, the work that was put into the open source project, we'd never be able to build our company. So I think this is a big, widespread trend that we're seeing in the open source community. First of all, the projects themselves are becoming much more relevant to what people are doing today. But at the same time, there's also a proliferation of new

projects that are coming out every month and every year. So database as a service essentially has grown significantly over the last decade with a whole bunch of NoSQL projects that are just getting their start. And what's interesting is as an individual user, it would take you a very long time to figure out how to leverage that project. You have to learn the ins and out There's a ton of best practice that goes along with it. And one thing of the open source community hasn't really done a great job of just yet is giving you a primer and getting you started on best practice implementation from day one. And I think that's where this open source as a service provider really can step in.

And hopefully they've taken the time out to perfect the technology on their end, which is based on that open source project, but it's pre-packaged for an end user's need. So now, let's say I do have a need for a clustered, scale-out, no-SQL database. I can just go to one of these open-source providers and consume that service without taking all the time to build it on my own. So this is definitely a prevailing trend that we're seeing more and more.

Yeah, so it's interesting. you know maybe to contrast what like it may not seem obvious to folks that um it may it's not even completely obvious to me that open sources out there i can readily pick up open source pieces and i can put those pieces in my on-prem installation so like if i'm thinking about this why what's the benefit of as a service versus, you know, just a company themselves, I'll say cobbling together open source components. They're readily available, free to use. Maybe, you know, are there pros and cons to each of those? And how might people think about adopting, you know, open source as a service versus putting open source on-prem? Yeah, that's a great question.

A customer that I was just speaking to recently is taking this new philosophy that they're building inside of their business. And what they're looking for is the opportunity to outsource all of these open source projects to a specialized provider. And their basic premise is we don't want to be responsible for the operational challenges that come with scaling. and running that technology. So if I consume it as a service, I have a very fixed rate under which I operate, and it's going to cost me, you know, X cents or dollars per time period or per volume or whatever the case might be. And it really is a fundamental decision that a company needs.

needs to make? Do you outsource these pieces to specialize providers and hope that they provide a better level of service and you can develop internally? You have actually someone's neck that you can choke, especially because you're paying the money for that service. Or do you build it in house? And with this specific customer, they've already built a lot of this as a first pass on premise, you know, taking that open source project and deploying it to their own infrastructure, and then realizing that there's so much complexity that goes into managing each one of these discrete pieces. And more importantly, if you just take the open source project, you don't have any commercial support that you can rely on. So when the system breaks, it's up to your

operational people to fix it. And, you know, that was the decision point that they used to say, in our particular case, we would much rather outsource this to someone else. We believe that our total cost of ownership actually winds up being less. And more importantly, if something breaks, we know that we're not on the hook. So I think the example that he used was in this particular leveraging AWS's RDS product, which is basically MySQL as a service. And he basically said, look, if the database is broken, there's nothing that we need to do because we know that the service provider will go ahead and fix the problem for us. Once the issue is resolved, our business is back online and operational, whereas the

opposite approach would be he needs to get in touch with his operations. people and and make sure that they're responsible for fixing the problem. So it's a fundamental shift in responsibility and cost. And does it make sense to outsource it to a specialized provider or deal with the operational challenges of running it in-house? Yeah. It's really interesting. I, um, in addition to open source being offered as a service, I've also spoken about and written about this whole notion of data center as a service, where, you know, you just spoke about these databases being offered as a service. And if you think about it, every part of the old on-prem ecosystem is now being made available as a service, whether it's storage, security,

infrastructure, all databases, to your point, are all now being made as a service. And I'm just curious like you guys are in this business and you're you're absolutely you know driving that trend why what what what is the catalyst that has made this happen now versus at some point in the past and why are we seeing all these sort of infrastructure components now being offered in the cloud and you know what's driving that trend yeah there are few things that are coming together at the same time. So we have virtualization that really came out in the mid-2000s and became mainstream. I think that really sparked a lot of change in the data center industry. Okay, virtualization kind of became mainstream. So what? Like, what did that do to catalyze this

on-prem to off-prem movement? Because I think you're on to something, and I'd love for you to clarify that? Yeah, well, part of the issue became that you have this substantially larger infrastructure to manage, because now a single physical machine could run 100 virtual machines. And so organizations or businesses needed to become much better at managing all of these systems of their deploying. And what happened is the open source community caught up to this need, and they'd built a whole bunch of tools that allowed that provisioning process to work a lot smoother.

So now you can actually provision these machines on the fly. You can maintain them inside of some type of inventory management system, and you have full control and visibility over what's happening. So virtualization created the opportunity for more servers to exist and there are physically in the world, and then the open source tools came along to allow you to manage that effectively. And I think those are a pretty big catalysts to drive the beginning for, you know, this utility type of model where people can come in. Because part of the problem is in the old world, essentially you would typically provision one dedicated server per customer. And you wouldn't always have a need for the full resources that the machine provides.

So there was a mismatch between what the service provider says. and then what the customer needs. And so by virtualizing it and being able to deliver a much smaller slice, you've now matched supply and demand. And that allowed people to really consume these, you know, services on an as-needed basis because the tools and the technology enabled provisioning of really thin slices in an automated fashion and everything kind of just, you know, supported itself. Whereas in the old model, you have people that would get a phone call and being told,

hey, we need you to go ahead and rack and stack a server with this many processors, this much memory, and these disks, that doesn't really allow this automated approach to work. Interesting. So this fluidity of sort of the back-end system as enabled by virtual machines really created this operational model that, that, that, that, worked very well for multiple customers getting maximum use out of the, out of the system in the back end. And that's probably that, you know, that's what you're saying is really the catalyst.

That's, yeah. Yeah, it's also interesting, it's also interesting to know that in this particular case, both extremes are serviced really well by this model in terms that a really small customer could now get started and enjoy the benefits of a world-class enterprise. infrastructure, you know, for just a few cents an hour. And at the same time, a large scale enterprise can really scale out their operation without incurring all the CAPX requirements that go into building out a huge data center. And more importantly, as business dynamics change and the workload changes, they can quickly adjust, you know, in a matter of seconds, rather than being stuck with hundreds of machines that they no longer have a use for.

So it created a lot of efficiencies from, you know, from a small customer all the way into the enterprise. Yeah, yeah. Let's maybe shift gears a little bit and talk about kind of the needs of the developer. I mean, your business is really, you know, kind of anchored around a large, large number of developers using your cloud to, you know, do their work. why wouldn't developers just go to a you know sort of standard public cloud why have a developer oriented cloud what are the differences and you know what happens over the course of time do the developers continue to use a developer specific cloud or move to something more generic i'm just sort of curious how you think about different and then can we envision different clouds for

different verticals, you know, maybe developers, what you guys are doing, but we have, you know, a cloud for rocket scientists, right? And, you know, I don't, you know, so. Yeah, exactly. So, like, how do you think about that? And I'm just sort of curious how you see the world evolving and, you know, kind of the importance of having a developer-oriented cloud versus a generic cloud and how, you know, how you guys think about that. Yeah. Well, prior to DigitalOcean, we feel that there was just the generic cloud. And the analogy that we like to use is the, you know, today's cloud providers are really focused on the technology. And they present that to their users in the most direct manner possible. And what that allows for is kind of what we spoke about previously, this really large. They allow infrastructure to develop. It gives you all the knobs and controls that you would need to take over.

Obviously, there's a bunch of operational work that the customer, the user, needs to do in order to take advantage of this public, generic cloud, but there's no limit to what they can accomplish. The challenge that comes up is as you build larger and larger scaling systems, the complexity increases linearly, if not even exponentially, to deliver that to the market. So what we recognize is that there was an opportunity for us to focus on people rather than on technology. And because at the end of the day, everyone is using the same virtualization engines. There are two or three popular ones, and it's kind of like, you know, Silicon used to build processors. No one has a substantially faster CPU than anyone else today.

And the same thing happens in virtualization. Everyone's virtualization engine is fairly similar in its capabilities and in its performance. And so it boils down to what are the core values that you're focusing on? And as I mentioned, most cloud providers today are trying to focus on technology. And they think by introducing the next feature, introducing the next service, they can win a substantial part of the market. And that is a fairly accurate statement. What they don't realize is that there's an opportunity to focus on the people.

And who actually uses cloud and hosted infrastructure? Well, developers, right? They are one of the fastest growing segments of users online. There are 20 million developers worldwide today, and that grows at a couple million people per year. And what we realize is that the actual human experience of trying to wrangle that infrastructure is very painful. And that's the problem that DigitalOcean tries to solve is by presenting a simple, intuitive, easy-to-use interface to bring that infrastructure under control. We use the same virtualization engines as the other cloud.

But where we spend a lot of our time is on designing the interface so that people can go ahead and provision these systems rather than it taking large enterprises and large teams of ops people and software engineers to figure out, well, how do I build the auto-scaling group and how do I manage these thousands of servers across multiple regions? What DigitalOcean is setting out to do is providing the easiest way for people to manage that without having to. to hire even more people to run their operations in the cloud. Against that backdrop, could you imagine other vertically oriented public clouds for other folks out there? Could you imagine something beyond developers? Let's say not for DigitalOcean, but in the industry in general. Does the public cloud verticalize around people, as you're saying, like you guys have done,

or is there only one which is developers and then there's everyone else? I think there's opportunity for other verticals to exist and depending on how far you want to stretch the cloud analogy, you could say that companies such as Squarespace or VIX, you know, these companies that provide build your website in the cloud using these tools. You know, they cater to a specific segment of, the market as well, but they're certainly not an infrastructure provider. They're more of a cloud website, if I may use that term. And even if we go back to infrastructure itself,

I think that there is an opportunity for another focus to exist. But it has to be so fundamental for it to create a real connection with the users. to stand out, right? If we say that current cloud, the generic providers are focused on technology, and we all use that, and we focus on people, and obviously all customers are people at the end of the day, even if it's a large enterprise like Nike, there's still someone that's making that decision. Finding another bucket that's as fundamental as those two, I think is possible, but there aren't many of them, and, you know, it takes a lot of of thought and a lot of focus. And what winds up happening quite often is that these buckets

actually exist many times over, but they're so small that in the grand scheme of things, we are not aware of these companies. So is there a third large bucket that a infrastructure provider could focus on? I'd like the thing that there's always an opportunity to innovate, to figure out the next step. You know, when we were building, DigitalOcean in 2011, this is the dilemma that we face. How do we build a company that stands out in the sea of all the other clouds that already exist? And it took us many months to be able to figure out our position around simplicity and focusing

on users that are developers. So I think if someone else took the time out to really digest the market and understand what's happening. There is certainly an opportunity for a new entrance. The challenges that quite often we find companies reacting to existing competition or to market conditions rather than being able to pierce through that, you know, fog and truly create something new. Yeah. Let me just sort of in rapid fire here, some competing things, you know, we talk about internally. Linux containers or virtualization? What do you think in the future?

In the future? Yeah, and the not too distant future. Linux containers started the whole trend. There's a bunch of talks nowadays about the hypervisor is dead. I don't think we're there just yet. Containers have a very poor security model, and that's really what's holding us back, but they're certainly more lightweight.

The real difference between the two is the container has to take on the properties of its parent hypervisor. So you're very restricted. You can't today run a Windows container inside of a Linux server. And that's what virtualization allows you to do. So it kind of depends on the use case, but for general public cloud computing, I think the container is the future. We just need to fix the security model.

Okay. Flash versus spinning disk. Well, Flash, I think that's a no-brainer. Okay. so disk goes away completely i think this is great for uh archival and and backup media because of its cheap cost and and fairly good performance and reliability especially in low i.o environment so i'd say you know second tier long long term storage so desk becomes the new tape yeah okay intel or arm intel i wouldn't bet again

them. Okay. Fair enough. There you go. Well, good. So with that, Ben, thanks a lot. We're going to, you know, call it a wrap. And I'll have you on at some point again. I'll love to see how all this materializes in the future. So we'll talk with you soon. Thanks.