a16z Podcast - a16z Podcast: Datacenter of the Future

Episode Date: June 18, 2014

The datacenter has long been -- there's no nice way to put this -- a bit of a snoozer. Expensive boxes running expensive software. No more, says a16z General Partner Peter Levine. Along with Chris Dixon, Levine lays out a vision for the datacenter of the future. Building on the technology established by companies like Facebook and Google, Levine and Dixon describe a software-led transformation of the datacenter, one where the mobile supply chain and fast-moving companies are reimagining everything -- from the underlying architecture to new business models. Be prepared to get in the weeds, hear Levine talk about the next opportunity, “hosted instances,” Dixon describe the “the dream within the dream,” and discover why the datacenter is about to get exciting.

Hi, it's Chris Dixon. This is the A16Z podcast. I'm here with my partner, Peter Levine, who, so Peter, you focus a lot on data center technology, sort of when people talk about the cloud, what is this cloud, someone has to build it. It seems like it's an exciting time in that area. Can you kind of explain why? Yeah, it sure is. And I'm Peter. Hello, everyone. It really is an exciting time. You know, data center, which tends to be the plumbing of the internet, has been one of these pretty state areas for a long time. And all of a sudden, there's been this unlocking of lots of innovation happening in the data center, largely on the back of, and not uncoincidentally, on the back of the whole mobile revolution.

Like as every human on the planet starts to use a mobile device, you know, tweets and emails and applications, all that information has to go somewhere. and it goes to back-end data centers, it goes to the cloud. And the cloud is comprised of data centers and computers, and that whole infrastructure is now transforming to support massive numbers of users that are out there, and eventually Internet of Things to support all the endpoint devices that are out there. Can you kind of explain just, like, so in the old days or whatever, five years ago,

if you would go as a company and you would rent, space from rack space or whatever some physical data center you would install your own servers you would buy software you'd install it there like oracle and all this kind of software right and now and then what happened is people started sort of use sales force and they manage their own data center because software is so software as a service is the first enabler right yeah um and then what you really had then these giant scale companies like facebook and google who who had such a new level of scale that no one had to actually design new infrastructure right and so can you kind of explain like how that is evolved yeah um i think it's tremendously interesting what facebook and google and

twitter and other very large scale out companies are doing is really interesting because it's the consumer internet companies that are driving this back end infrastructure they have they have to they have to right i mean they're supporting billions of users this whole notion of multi-tenancy you think about all these users they're all kind of you know independent it's kind of crazy to think facebook is more sophisticated on networking right now than Cisco, right? I mean, is that fair to say? Yeah. It's very fair to say. I think they, they, Facebook and Google have absolutely pushed the needle on backend data center infrastructure. And it's interesting. It's all about software. It's about commodity components kind of being put together in these data centers. So it's much less about

thinking of a big vendor coming in with a refrigerator size switch or storage device. It's much more about aggregating very inexpensive components at massive scale using a layer of software on top that can unlock the value. So in the old days, you would go to EMC for your storage and they make a custom piece of hardware and software. You go to Cisco for your networking. You'd go to, what, HP, Dell for your computer. For your servers. You'd go to Oracle for your databases. And all those custom hardware. Now people are saying, no, let's just take complete commodity, Linux, and it's all software. Absolutely. We like to say the mobile supply chain is eating the data center. Right? And it's rather interesting that as, you know, look, mobile, the innovation in mobile and the

innovation in hardware on mobile devices has been phenomenal over the past several years. We will be able to take many of those in Facebook and Google are starting, setting an example here, to take many of those components, whether it's arm chips or flash memory and take those hardware components and aggregate those together with open source and proprietary software. to create the next data center of the future. So my smartphone is connecting to the cloud, and the cloud, the data center is literally going to be a bunch of reconfigured smartphones.

I, I, that's a great, look, I have this visualization that the data center is a billion smartphones connected together through very sophisticated software. And it doesn't, you know, may not have a screen on it or whatever, but it's, it is, it is those. For example, arm processors. Arm processors. It turns out that power is the most important,

is the scarcest resource in the data center, same way that it is on the handset? Power and cooling are critical elements in designing a data center these days. So to the extent that we can go design a data center that uses much less power, requires much less cooling, is optimizing on floor space. Those are the ingredients for massive scale-out data centers, and those are the drivers. So to the extent we can use mobile components for that, it's definitely going to solve that problem. So how does this then relate to are investing so you know i know probably what a quarter maybe even a third of our investments are around data center technology can you can you kind of walk through like different areas of

storage networking compute yeah so on a on a on a on a on a broad level if we take facebook and google those companies are becoming the blueprint for data centers of the future it used to be where companies would look at wall street for example at fidelity and morgan stanley and say hey we want to copy, we're going to build a data center like them right now. So Fidelity would build something and maybe some of it would be custom. And then companies would come along and say, you know, let's take that, factor it out and sell it to other people. And, and right, companies would come, exactly. That was a startup model. Startups would go. Startup would go go look at Fidelity or, or Morgan Stanley and go build a stack of things. So in the old days,

if you were an entrepreneur looking to do a data center company, you'd go and learn what Fidelity's doing, package it up as as as a product. As merchant market. as merchant market products to make it available to the rest of the world. And now you go look at what Facebook is doing and Google's doing and you say, okay, instead of me as a company having to need 10,000 engineers to keep this up and going, how can we create the merchant market products to make that type of data center available to data centers around the world, to companies around the world who want to go build this next generation of system infrastructure.

And so a lot of the companies that we invest in, whether it's storage, networking, compute, databases, analytics, security, all follow the model of how do we unlock the value of commodity hardware components with sophisticated software to bring this next generation scale out data center to everyone who's building that. So maybe we can talk about a few of these companies. So like on a networking level, NICERA was a company that we invested in. Can you explain what they do? Yeah, so Nicera, a company that was acquired by VMware, they really pioneered this whole notion of software-defined networking and really unlocking this sort of hard coding of networking to something that's much more flexible.

Anytime you build something in software, it becomes inherently more flexible, so they unlock that. And what's the customer benefit of that? The customer benefit of that is that no longer am I locked in. into particular silos of networking or having to lock down a server to a particular underlying piece of storage or a user to a particular server, it's very fungible. And so if I'm creating a service-oriented architecture, I need to be very flexible in terms of how my network connects to different parts of my physical infrastructure, and that the software layer actually enables us to go do that.

So Nysera is now part of VMware, and VMware is taking that to market. We are on the networking side, one of our current investments is Cumulus Networks, which is building very much along the theme of replacing network switches. What they are doing is taking commodity hardware, putting a layer of open source and closed source software on top of commodity servers and turning that into a new switch that is as fast as is, you know, as capable as some of the most proprietary systems that are out there today from the likes of Cisco and Juniper. And then what about storage?

So on the storage side, we see much of the same thing, where, again, it's all about software being built on top of commodity components, whether it's Flash or, you know, Flash has become a huge, trend in storage while unlocking that flash and using commodity flash as the basis of an enterprise storage array is quite compelling for the for the market to the extent that we can build you know take software and unlock the value of commodity flash we can literally see in the not too distant future the cost of flash storage devices being the same price as you know spinning media and that's going to be very compelling.

It's fast. It's, you know, reliable. So I think that the transformations on that side are really interesting as we start to use, again, these commodity components for the basis of the systems. Why weren't people using commodity components, let's say, five or ten years ago? It wasn't, the performance wasn't there? It's just one of these sort of Moore's lost things? So one could argue that we did use commodity components.

You had mainframes. Mainframes were replaced by PCs. So five, ten years ago, it was the PC revolution. So x86 chips and, you know, disc drives from a PC and Microsoft software and Oracle did. That was the, that was quote unquote, commodity. Commodity now is even way more commodity in that in the same way that PCs replace mainframes, I see mobile devices replacing PCs. So it's really, you know, in 10 years ago, people,

we're talking about commodity components in the data center, meaning X-86 and PCs. Well, now we take that a step further, power cooling, the price of these components coming down that much further. It's just the further commoditization of enterprise hardware that is continuing. And so how do you think the world's going to look in five or ten years in the sense of is every large company going to be building their own data center using the software? Are they going to be outsourcing it to $8? and Amazon and Google, like the stuff that we're seeing now was, like every startup we meet, right?

Yeah. It would be strange to not be running your stuff on AWS. Yeah, yeah, yeah. Is that how enterprises are going to look in five or ten years? Look, I think a lot moves to the cloud. I think there's the big debate of does everything move to one? But there's sort of private or public cloud, right? There's private, you know, there's certainly, there's...

Private cloud is Goldman Sachs has their own, quote, cloud. Right, right. So I think certainly the notion of private and public clouds will be the, the dominant compute sort of style in the future. So there won't be a question as to whether you do a cloud, quote-unquote cloud-style infrastructure that stuff we're talking about, it's so much better, but it'll be a question as to whether you own it and control it. Correct, correct.

And that will be mostly around security? Security, data, you know, kind of within. Regulatory stuff, like HIPAA and whatever. All of, you know, and so, you know, sort of DNAs of the company themselves on how they view this. but every data center of the future will be a scale-out data center, right? Applications are going to change. The user base is going to need that as, you know, again, mobile,

and then we start to think about Internet of Things as, you know, collecting all this information and analytics that needs to be done on that, all point to this notion of the scale-out data center, whether it's public or private. I see. So you think it'll be, and the difference will just be, as you said, depending on the DNA of the company or maybe some security issues or something, but there's not a whole lot of other,

there won't be other reasons why you would go private or? I mean on-prem, on-prem private. You know, this is a, it's certainly a, you know, if you have a private data center right now on-prem, I think it'll take a while, so a lot of people will say, well, of course I'm going to continue to do that. I think that the economics and the way that data centers are going to be built

is going to drive everyone to private or public clouds and the notion of the on-prem data. center starts to disappear. But, you know, we'll see what happens. And you mentioned Facebook and Google leading the way. Amazon's not leading the way? I mean, that's the people think of cloud. They think of Amazon. I think that the Google and Facebook infrastructure is now, in my opinion, more advanced than what Amazon is. Amazon was created several, many sort of already generations ago relative to this commoditization curve. Can you get some examples of the kinds of things that you do that are old technologies? Um, you know, um, virtual machines for example. I think, uh, yeah, that's a good one. Um,

uh, you know, Amazon, Amazon, from what I know is, uh, based on a virtual machine architecture. And virtual machines, um, are, were, you know, the way to encapsulate information and encapsulate sort of, uh, uh, you know, kind of, uh, uh, data. I see that we've already moved to the next generation where virtual machines, certainly within, which is containers, which certainly in a Google or Facebook environment, it's more about containers and less about virtualization. Now, why is that? Well, containers use far less compute resources. There, you know, it used to be where I had to support Windows and Linux on a machine, therefore virtual machines are important. Now it's only Linux on both sides. as the as the base operating system and running within these containers so we can simplify the environment by as an example through containers and that unlocks a lot of a lot of additional power for compute for applications as opposed to locking that power down for for the virtual machine

infrastructure so what are the areas in the data center that we're like that we're looking at for new investments like what do you consider well one of the new investments that we recently made was a company is a company Mesosphere, and I'm particularly excited about that particular architecture and that it literally is what I call the one-over-v-v-emware technology company. And when I say one-over VMware, this is not about carving up machines and carving up resources. It's about aggregating resources together. And I think that the notion of aggregation of processes, compute resources, storage, and bringing that all together against this massive scale-out kind of data center is going to be the way that that environments run in the future.

So that's one area. So I think of it as, so tell me this is wrong, but I think of it as if, so in the cloud architecture, the data center becomes a lot like a server used to be, and therefore a single server. So now the entire data center is in sense one server, and all the things you used to do at the server level and the operating system are now to be done to the data center level. And I think of Mesos is sort of like maybe the scheduling aspect of the operating system. But it is a platform for the aggregation of these resources beyond just scheduling.

So one could imagine applications actually being written to the Mesos API and it will coordinate the right resources from the right places coming together, not unlike an operating system does on an individual server. So you can say, I want, you know, to increase the capacity of web servers 50%, and databases 10%, and it'll just sort of... And MESOS will go out and get that for you and make it available to the application without the user doing anything. In the old days, you had to have like a team of sysadmins, you know, whatever, spinning up

servers and configuring. And your analogy of like single server to data center being the server is exactly the analogy. Within a given server, I could always, you know, I could tune things up and move. things around. But at the data center level, very, very difficult to do and then things change and it's, you know, sort of a shit show of stuff that goes on and it's crazy, right? And so as you expand out to, thousands or, you know, we talk about a billion cell phones, imagine trying to manage all that without something like a Mesos helping to coordinate at all. So what other areas

you think like are, you know, that we haven't made investments in yet that you think are interesting? I'm interested in, I think the whole container world is, uh, is particularly interesting. So containers, let me do my layperson's take on containers too. So virtual machines in the data center world have probably been one of the biggest inventions of the last 10 to 20 years.

VMware, massively successful company that pioneered it. It basically allows you to run an operating system on top of an operating system, which is great for both running windows on top of Linux, but also in a way to partition resources effectively. So you can set or say, right? I mean, it lets you change things in software that pass you to change in hardware.

Like, you want to run more web servers or something. But the problem was you're running an operating system on top of an operating system, which is a preposterous architecture, right? Like, I mean, it's a total hack. It's a fun. It's a cluge. It's a clue-chee hack. Well, it's a complete replica like that operating system, the base level operating system

is a complete, it's a complete reproduction of the server itself. It's inception like, and that everything is like happening a dream within a dream. And it's incredibly. You know, resource intensive. It's very resource intensive. Now, so containers are you get the benefits of that, but you're, you're not doing a dream within a dream. You're doing one layer of operating systems. You're doing one dream, right?

Yeah, it's one, it's sort of one layer. The operating system effectively reproduces itself, right? So you have multiple instances of the same operating system running on a system. And look, I think the notion of, the notion of having a content. and being able to move things. And this was the best of the virtual machine world was that I had this entity that I could move around. And it was comprised.

So a developer now, what happened also, right, is in the old days, you wrote a program, you wrote a program, right? Now you write a program, you got 50 different libraries, you've got a database, you've got a web framework, and just getting the thing deployed. So you take your software and I want to put it on a new machine is like a day of work of like downloading,

figuring, installing. And so containers let you put. Package it all up in one single file, and boom, you can shoot it out anywhere you want. And that's been, that was, that is the value. So for virtual machines, that was this, you know, had similar characteristics. One file I could go deploy in multiple places. Containers offer that without all the overhead of this virtualized infrastructure. And so I think that that's really, it's a rare thing where it seems like it's a no-brainer

dramatic improvement over virtual machines and they're just, VMs are over. I mean, it's going to take a while. It certainly will take a while. But there's just, there's not, there's not really tradeoffs here. The, the, well, the tradeoff in the past was, and I mentioned this before, is that what virtual machines allow you to do is run different operating systems on top of the base operating. That was a big, it doesn't matter anymore.

So while it's a no, you know, Linux is so people don't understand, some people don't understand, it's so thoroughly one. It's pervasive, right? It's one. And so if I happen to be running Linux on top of Linux, like, hey, that's just fine. In the quote-unquote old days, I had to run versions of Windows. I had to run different instances of Linux. And so that mixed environment created the opportunity and rationale for server virtualization.

So look, I think the world is shifting in this new direction. And, you know, it's going to unlock, I think, a lot of value out of, you know, these data center infrastructure components. So around containers, so we have Mesosphere, like what are the other interesting kind of, Start-off opportunities around containers. You mentioned containers as being an... So one of the things that I'm particularly interested in is the hosting of a variety of infrastructure components. So we made an investment in a company DigitalOcean, right?

And what they do is they take a variety of open source components and they're a cloud for developers. And what I find really interesting about their environment is they are offering to developers the hosted instance of many open source projects. So I'm a big believer that we're moving from a world of sort of, you know, this on-prem kind of, you know, virtualization running on-prem to these hosted instances of, let's say, hosted instance of Docker, for example, or hosted instances. of databases. And so I think, in my mind, to the extent that we look to invest in the not too distant future, it'll be about investing in hosted instances of particular open source technology. So a hosted Docker might be an interesting thing. And this is partly because it provides a good business model for open source? Yes. One, it provides great value to the

end customer because I no longer have to deploy this. stuff on-prem, right? And then the business of open source, this completely rationalizes open source as a viable business model. In the past, open source was a questionable business model. Yes, Red Hat made a, you know, nice way at it. I wrote a blog on this. Like, you know, there are some companies who will make good money with open source.

But there haven't been other significant, there been acquisitions, but there haven't been there haven't been standalone successes other than Red Hat that have really been able to quote unquote monetize open source. Given that it's, you know, if you compare it to its proprietary software analogs, it's, you know, Microsoft, et cetera. You just made a couple of minutes ago, you made the point that Linux has taken over the world, right? There's no need for anything else. Well, if you think about the economic benefit that Red Hat has derived over the complete transformation of Linux, it pales in comparison to the impact that Linux has had. in the world. And so, look, open source has been fantastic. It's been a fantastic vehicle for

innovation. But in terms of economic value back to the developer of open source and back to the provider of open source, I believe that once open source can be provided as a service, it is then able to unlock full market value and full potential. And to me, that's a really interesting sort of transformation. Does that mean the company, like historically with open source, the interesting company has been the company that had the core developers of the project. So for example, MySQL, the company had the MySQL developers, right? Yeah, yeah, yeah, yeah.

And so in the future, you think that there's a cool new open source project X, the key developers for X will be in a company, and that company will provide a service-based version of that software. That's what you think will be the future of the open source. Yes, I think that to me is the interesting part of the next gen open open source models will be that particular, that particular type of deployment. And, you know, GitHub is a good example of this. Like, you know, they're the epicenter of all open source development, of all development, right? So you could have taken Git and you could have sold it MySQL style or you could do GitHub, which seems to be a much more interesting business.

Duh, right? Like, yes. So that model is exactly what I believe is going to happen at every segment of the open source market will be around. You know, GitHub has that network effect. Yes. Do the other, I mean, that, I wonder if it's really the beginning of a pattern or if it's such an exception. I think, look, GitHub certainly has the network effect of developers, but I would argue that if they had just gone off and supported Git and done it as an on-prem thing with some collaborative hoo-ha on top of it,

it would never have been as successful as what it is now. So sure, the collaboration adds to it, but look, I mean, I think there are a number of companies who I believe will have very great promise of offering sort of the sassified versions of open source that will be the future, just because it's going to be very compelling for them to continue to innovate and for users to actually use that in ways that we just haven't seen before. Okay. All right. Great. We're out of time. Thanks. Okay. Thanks.