A16z partner Peter Levine on why mobile phones are the future of the datacenter
Andreessen Horowitz general partner Peter Levine has a long history in enterprise IT, dating back to his early career as an engineer and EVP at storage company Veritas. He is best known as CEO of server virtualization pioneer XenSource, which Citrix acquired for $500 million in 2007, after which Levine served as SVP and GM of Citrix’s data center and cloud computing business.
In this interview, which has been edited for clarity and style, Levine shares his opinions on the future of computing, including the notion that everything will change once the economics and architectures of mobile phones and datacenters become nearly indistinguishable from each other.
SCALE: What has you excited in enterprise IT right now? What are the next hot areas?
PETER LEVINE: One of the observations that I have about enterprise IT is that there is a tremendous number of innovations occurring by younger folks. It used to be people would work at Cisco or Oracle for a lot of years, leave there and go start a company to do enterprise IT. Now what we’re seeing is a lot of folks coming out of Ph.D. programs, or straight out of universities, to go do enterprise IT. I think that is yielding tremendous interesting innovations.
If you look at just in our own portfolio, three really interesting projects — Spark, Mesos and Tachyon — all came out of Berkeley AMPLab. This is an example of where the hyperscale datacenter IT community is actually being led by a younger generation. I think that is upending many of the conventional thoughts that we have seen in the industry. That’s very exciting.
Whenever we see an interesting project come out of a university, where it’s a Ph.D. student or young entrepreneur working on it, we’re going to take a look at it. It’s more like fresh views and out-of-the-box thinking is driving this new wave of innovation in enterprise IT.
Do you think younger entrepreneurs miss out on any valuable experience by not working at large tech vendors first?
I think that the new entrepreneurs are thinking out of the box and they’re reinventing how enterprise IT is supposed to work. I’m more interested in making IT simple and making it perform and making it consumable, which has not been paramount in ideas coming out of folks who have long backgrounds in this space. I think the combination of the mobile era and a younger view on this is yielding new products that actually lend a consumer orientation toward IT, which is very new thinking.
One might view it as hampered because you don’t have the IT background. I view it exactly opposite and think of it as liberated thinking. I love it. I think it’s changing the market right now.
“I think with the proliferation of mobile devices, and then eventually the Internet of Things, we literally have supercomputers in our pockets and supercomputers that will hang on telephone poles and in light bulbs.”
The move to mobile architectures in the datacenter
What new technologies are emerging as a result of this? If you look five years ago, it was broadly big data and cloud computing, but there has been a lot of evolution lately.
Those trends that you mentioned were things that one should have looked at five years ago to skate to today’s puck. When I think about skating to the puck five years from now, I think about things like machine learning, as an example. It’s a natural extension of analytics and big data, but done in ways where software inherently becomes more intelligent through its use of data. Starting to deploy machine learning as opposed to business intelligence is one area it might take off.
There’s also a very interesting trend, relating to the relationship between datacenters and endpoint computing. If you look back over the history of computing, it started as mainframes or terminals. As PCs or work stations became prevalent, computing moved to the edge and we had applications that took advantage of edge computing, and the CPU and processing power at the edge. Cloud computing brought things back to the center. There has been an ebb and flow in enterprise IT, of centralized versus distributed.
I think with the proliferation of mobile devices, and then eventually the Internet of Things, we literally have supercomputers in our pockets and supercomputers that will hang on telephone poles and in light bulbs. That latent capacity will be used for applications running at endpoints, which will change the dynamics of cloud computing just like the PC did. I think that that’s going to be a very profound shift, that more and more computing will be done at the edge as opposed to always relegating it back to the datacenter.
That sounds like a feasible vision, but most of my apps are still cloud services.
I am starting to see the seeds of new projects that are starting to look at that. I think it’s very early, but I can see a world where endpoint distributed computing becomes more popular, just like we saw when corporate IT shifted from mainframes and onto workstations and PCs. Are there hot projects right now? Not yet, but I’m starting to see university work being done in this area.
Then, there’s the discussion of what happens to datacenter architectures. If you believe that the mobile phone is the next supercomputer, which I do, you can imagine a datacenter that is modeled after, literally, hundreds or thousands or millions of mobile phones. They won’t have screens on them, but there’ll be millions of lightweight mobile-phone processors in the datacenter.
“Where’s the boundary between the datacenter and the endpoint? It’s an amorphous blob of stuff and applications can move all around it.”
Over time, we’ll see commonality of architectures and applications that will become a very effective model for running apps and doing things where the boundaries between the datacenter and the endpoint may be a lot less clear than they are today. Where’s the boundary between the datacenter and the endpoint? It’s an amorphous blob of stuff and applications can move all around it.
That may be very futuristic, but if you look at past trends I think there very likely could be something there. Just like the PC supply chain ate into the mainframe, I believe that the mobile supply chain will eat into the datacenter.
If we were sitting here in 1980 and somebody suggested that the IBM PC and then the 8086 or 8080 architecture for processors would become the basis of datacenter computing, people would think you’re crazy. There were two floppy disks and it had no RAM and all that stuff. What has happened? The Intel architecture, which began on the PC, has become the standard in the datacenter.
What does that say about the future for the companies making processors, servers and other datacenter hardware today?
I think they evolve. A server could still be a server, you could have lots of CPUs on a motherboard, but they’re mobile-phone CPUs. It doesn’t mean that servers go away and we’re literally plugging in mobile phones in the datacenter. The architectures will be similar. Networking for mobile devices, memory for mobile devices will all become properties of the future IT infrastructure just because the supply chain, quantity, cost and economics of those components are so favorable that they will drive themselves into the datacenter over time.
For the big guys, they’ll have to go and figure out how to take advantage of some of these new trends and rely on different components. Certainly, it’s also going to lead the way for new companies to come into play. We see the seeds of that. I don’t mean to promote Mesosphere, but if a datacenter has a million CPUs and billions of processes, a Mesosphere-like architecture is exactly what’s going to be required to manage all that stuff.
We’re already starting to see this with microservices. Everything is becoming more granular, smaller and aggregated together. That transformation will continue to occur over the next 10 years, and smaller, lighter-weight software and hardware components aggregated together to provide a massive compute infrastructure is going to be the way things will go.
“My belief is that we’re going to have smaller machines that don’t need to be carved up into little pieces, so we need to think about aggregating compute resources.”
Can legacy IT keep up?
You spent a lot of time at XenSource, and then at Citrix after the acquisition. What’s your take on the state of server virtualization, a space where you blazed some trails but that seems on the decline?
My sense is server virtualization is counter to the trend that I just mentioned. Server virtualization was all about taking a big machine and carving it up into little pieces. In time, my belief is that we’re going to have smaller machines that don’t need to be carved up into little pieces, so we need to think about aggregating compute resources. That’s point one.
Point two is that server virtualization is very important when we need it to run both Windows and Linux workloads on the same system in a datacenter. As Linux has become more homogeneous in the datacenter and there’s less Windows as an operating system for datacenter computing, the importance of having the ability to run multiple operating systems has come way down. Server virtualization has been around for a long time and is always going to have a place, but I believe it’s less important in the future we’re talking about, with microservice architectures and the whole notion of containers being an important component of datacenter computing.
What does that say about companies like VMware, or even cloud computing providers, that have built their businesses around the virtual machine?
Part of this has to do with what enterprise customers will buy. There’s still work to be done on container security, for example, while virtualization is very secure.
But I believe that those older architectures will need to shift up over time as they become a lot less economically efficient to run. The next generation of cloud services that’s run on a container with a microservices architecture supporting it is going to provide a lot more density of compute resources than you get from a virtualized environment. The economics just start to look so much better. I can put more workloads on a server, move things around more gracefully, and that starts to shift the benefit from one cloud provider to another.
“Computing will fundamentally change once memory becomes flattened and we have a couple of tiers of memory instead of all the storage tiers that we have right now.”
When big data goes in-memory
You mentioned the big data space earlier. What’s your take on that space, and is Hadoop still a thing that we care about or is it just a generic term at this point?
I think Hadoop certainly became an onramp for the whole big data environment that we’re in right now. Hadoop was never designed to be a real-time system, it’s a batch-oriented system. And it’s a fine system if you’re going to ask a question and get a response back.
Where Hadoop is not particularly great is for new areas of analytics, like machine learning, where iterative real-time processing becomes really important to solve problems. That’s where Spark comes in, along with companies — like Databricks, a company in our portfolio — building on top of that. We’re investors in a company called Adatao that’s building a real-time analytics system on top of Spark that starts to promote the whole notion of machine learning and iterative learning as the next phase of big data and processing. I think that that’s really where things go from here.
One of the foundational components of Spark is its ability to process data in memory. Are we reaching a point with RAM, for example, where companies could begin their lives in RAM and grow with it, possibly not ever requiring a big disk-based file system?
Absolutely. We are investors in a very important technology called Tachyon, which is a file system abstraction. The basis of that entire innovation is that memory flattens, there’s no spinning disk anymore and everything is just in memory. I think that that’s a very, very real future area, and that computing will fundamentally change once memory becomes flattened and we have a couple of tiers of memory instead of all the storage tiers that we have right now.
It’s not only a function of how big data changes; everything changes at that point, including the way applications work. Imagine if I no longer have to store things out in files, because everything is just in memory.
If you look at the mobile supply chain and the cost of memory, I believe that over time we will start to flatten memory structures and that becomes a very, very interesting transformation. We have Spark as an in-memory database abstraction and then Tachyon as an in-memory file system abstraction. Those are the two pieces that start to unlock this next generation of real-time computing.
“Are all of these valuations justified? Probably not, but many of them certainly could be if you believe there’s a franchise company to be built.”
Bubble? What bubble?
What’s your take on the bubble that people worry about, at least as it relates to enterprise IT and enterprise startups?
First of all, in early-stage investing, A-round investing, I don’t think there’s a bubble at all. We’ve seen almost no change in early stage A-round investing. That’s where we spend most of our time in terms of making investments.
I think companies are staying private longer. Later-stage investors who typically have played in public markets now see opportunities to invest in private companies, when in the past those companies would have gone public. Some of these valuations are simply based on companies staying private longer.
In addition to that, I believe that this transformation from the older IT environment to this new world really is going to upset many many incumbents and change the landscape in enterprise computing in fundamental ways that we haven’t seen since the last transformation, when the Ciscos and Oracles and the EMCs of the world got started. A lot of innovation is occurring that I think starts to move the deck chairs around, and there’s a lot of money at stake. That certainly would warrant appreciable valuations for those companies if you believe they’re going to be one of the big franchises of the future.
Are all of these valuations justified? Probably not, but many of them certainly could be if you believe there’s a franchise company to be built.
“I think we’re shifting from open source being seen as paying for support — the Red Hat model — to some really interesting models that are quite exciting, quite innovative and quite profitable.”
So much of what we’ve talked about starts off in open source. Do you think public market investors, for example, have yet to fully grasp open source business models?
I think it’s starting to come around. The notion of open source has changed over the last five years to where folks look at new open source companies and say, “Wow, there’s actually a really interesting business model.”
Things have moved from just being a support and service orientation to something where companies can sell a license of some sort or provide software as a service, which I think people can get their heads around. We’re seeing open source companies being profitable and having business models that are sustainable, and that’s very exciting.
You don’t think they suffer from not producing big Oracle-like revenues?
It remains to be seen. The largest open source company right now is Facebook, and it certainly has Oracle-size revenues. I know that it’s not an IT company, but it’s just interesting to watch as business models shift to be able to leverage open source in new and interesting ways.
The way GitHub, for example, monetizes open source is very new and innovative. I think we’re shifting from open source being seen as paying for support — the Red Hat model — to some really interesting models that are quite exciting, quite innovative and quite profitable.
