Cumulus Networks CEO JR Rivers on what’s hot for Open Computing in 2015

Over the past few years of blogging, many of my most popular posts have been interviews with industry thought leaders. To that end, I have started a new interview series I call ‘the catch-up,’ where I will catch up with interesting thought leaders in the industry.

This is the first interview in the series, featuring Cumulus Networks CEO JR Rivers. Over the next few weeks, I will post interviews with Martin Casado, David Meyer, Rob Hirschfeld, Dan Pitt, and many more. In the past I have only posted the text from interviews; now I have also started recording the interviews and will post them on my YouTube channel, OpenNetworking.TV.

For Today’s interview, JR shared some great stories about how Cumulus got started and what they are seeing in the market today. Read on for the full transcript.

Please excuse any spelling or grammatical errors in this transcript.

Art Fewell: Welcome to OpenNetworking.TV this is the catch up, and today we’ve got on the show for you JR Rivers of Cumulus networks, the makers of Linux for networking and really brought networking into the Linux age. From a market dynamics standpoint or from a Wall Street standpoint Cumulus has really been one of the startups watch and a huge disruptive growth vehicle and you’ve been sitting very visibly at the helm of that.

JR Rivers: Thanks Art and thanks everyone for tuning in. The Cumulus story begins from a pretty simple perspective in that I’ve been a networking for sixteen years, worked for Cisco, helped Google as they started building their network equipment. Inevitably from when I started off in networking and in every big new transition there’s always been one fundamental goal, which is to help customers get better interconnect than they had before.

When I talk to customers they often talk about the super fancy things they might do with networking and I always try to break it down for them and say ‘what do you really really want?’ In the end it always turns into ‘I want a big fat free bus’. So I want really high-capacity network I want it to not get in my way.

James Hamilton said a great in some of his Amazon pitches. In recent times networking has been in the way of customers and a lot of times people want to take James’s statements to mean, ‘hey I need to have some new fancy network feature the network didn’t do yesterday’, but really what James was trying to say is that from a cost performance standpoint, networking as of a couple years ago was nowhere near the evolution phase that compute or even storage are at.

And so from a Cumulus perspective we are really focused on moving the networking industry into that same evolved barrier of super high capacity interconnect, easy to deploy, easy to deploy at scale, works easily & is easy to acquire. Its kind of affordable capacity that’s easy to get out there and that’s what drives us in pretty much everything that we do.

Art Fewell: If you take a step back and talk about the changes that we’re trying to make in the networking industry and if you look at the macro-economic standpoint, we are saying that it’s nothing that’s surprising or unique about the networking industry, any industry that has lopsided market share without multiple really strong and competitive suppliers they all end up bad, they all suffer loss of innovation and all pass along a lot of extra cost to the customer. I think having open source as the center for your ecosystem is is really a critical component for building the industrial commons a healthy industry needs.

JR Rivers: Yeah you’ve raised a really good point, Cumulus Linux from a networking perspective – a lot of what we do leverages and focuses and integrates open source and a lot of networking technology that customers consume today. We were recently going through a master source agreement with a customer and they had us give the open source list and were worried about libraries that we might have and licensing requirements. We took a look at a bunch a competitive products found that all those products had the same license exposure to the customer that they were worried about with our platform, but we were transparent about it, the customer could see everything under the covers.

It’s that transparency that the open source community or that open source perspective gives to customer’s. It’s not so much the customer wants to go build something themselves it’s the fact they can see inside if they choose to. If they want to look at a piece a code and understand how some things work they’re allowed to but more often than not they want somebody to take responsibility for making sure that everything works all the time. As a customer you don’t want to go hire a bunch of developers to integrate stuff. Any of this code whether its networking or storage or your compute code you want somebody else to be responsible for the integration at scale, you just want to use it to further your business cause.

Art Fewell: We often get focused on the technical side of networking, but as the internet grows we see a lot of potential to help solve a lot of really big human problems. We’ve seen people like Peter Diamandis or the Clinton Global Foundation out there showing that as we bring the rest the world’s seven billion people onto the internet, it’s not just about ‘this product is better than that product’, it’s about how we’re going to

to change everything we know about health, we are going to take all the people that starving and without water in the world and change that, and when we can come together in these open movements and put that behind us its fundamentally important in realizing that we all are doing really important work for the future of the world.

JR Rivers: I agree with you about that what we’re doing is quintessential and meaningful; we’re doing it in our small little piece in the industry in the industrial complex that gets technology applied usefully to the common good.

You know when I look at some of our funding rounds, Andreesen Horowitz happens to have lead our Series A funding and when we were talking to the partners, we started explaining what our vision was and why this matter to us, why were we interested in this whole networking thing. Mark Andreesen latched on to both Nolan and my core visceral perspective very early in the presentation he stood up and looked at the rest of the partners and said, ‘look guys you know its clear that the Cumulus team isn’t focused on reducing the cost of what somebody bought yesterday, the Cumulus team is focused on giving people way more networking than they had yesterday so they can write applications that they couldn’t write yesterday and do things that we don’t even know about’ and that’s really in the end what drives us.

Art Fewell: If you look at the Internet of Things from Fitbits to everything else is going to be connected to network, in there is pretty much everything that anyone will be making in the future. Now a lot of those are going to be products where the people launching them will want to create competitive differentiate with right? And in some cases you may be able to say ‘I can take a standard TCP/IP stack and will differentiate somewhere else application layer.’ But in a lot of these cases, to be able to find your differentiation you’re going to need to go in and customize various components of the stack, and regardless of which off-the-shelf software features you’re using, one thing is for certain once a feature is available in any off the shelf OS or app it will not be a differentiated piece of your product offering.

As people look forward and especially as more and more companies get into the Internet of Things, into making more and more of their own products network-centric, to me it seems Cumulus is really important in terms of, if I want to be able to innovate there I need to have that flexibility right? 

JR Rivers: The Internet of Things (IoT) is a really interesting place to start. You can take that broad picture and broad perspective and start to narrow it down. When I look at IoT from a commercial perspective I look out and I see companies like Cisco and others stand up and try and be the leaders of the Internet of Things, from my perspective its kind of a crock of poo, because the IoT by definition is a federation, a loosely coupled ecosystem. I have bits and pieces that can connect to each other in ways that are recently well understood and they interact with other pieces of the IoT ecosystem as ships in the night, not knowing the other exists.

My classic example I give to people all time, consider your cell phone, you connect to at least 10 services through the network whether it’s the data network from your your smart phone provider or through your WiFi, it’s all happening – you don’t know who’s transporting any of this data anywhere but you connect to the services and you get the behavior that you’re expecting out of it. That federation is what makes it work and the definition of federation means that nobody owns it you don’t get to decide that the Internet of Things is X because I am company Y and I’m big and I have a 40 billion dollar annual revenue stream. Its gonna happen without you. Nobody owns this, but you can help enable it and from a Cumulus perspective our goal in enabling it is high-capacity interconnects in data centers and in carrier networks and to just focus on that. How do you move bits around really fast and really easily with low OPEX and low CAPEX. 

Art Fewell: It is fascinating when you think about, you see all these guys out there saying ‘hey we have the platform for the internet of things and we’re going to be the center’. Whenever I hear that I think, ‘wow you really don’t understand how big how diverse that the internet of things is going to be’

JR Rivers: Right, it already is, the Internet of Things is just kinda the continuation of where we’re at today. Refrigerators and TV’s, Nest, your humidifiers, your heating controllers, your lighting systems all these things are already on the internet. This isn’t a new initiative.  It’s not a set a standards that exist around it, its about a very loosely coupled ecosystem.

Boiling it down to the world that we live in, the small piece that Cumulus rules – we chose Linux as the foundation for the networking operating system because it allows federation in this small part of the world.

We have customers that put together well known configuration management tools or some homegrown monitoring mechanisms, they use the set of tools that makes sense for them in their environment. And it’s very federated, all these pieces can work together or you know or in exclusion of each other depending on what a customer might want and we just provide that platform and a foundation that an operating system is supposed to provide that allows customers to make those choices. And that’s why we’ve been reasonably successful along the way.

When we started the company we recognized that we would be very applicable to Internet facing businesses whether it’s a web services company or a service provider, a big internet retail presence or others like that, we knew we would make sense to them. We were unclear as to how this was going to work in the context of enterprises and what we’ve noticed over the the second half of last year is, we’re seeing enterprises look at their internal IT and they say ‘look, we have this legacy IT that’s is not built on a federated model it doesn’t look like the Internet of Things, I need this crazy little networking technology to make my application run and it cost me a lot of money in People time & OPEX dollars going to vendors and in CAPEX dollar’s going to vendors to run these applications and these networks this way.’

And then when they look at the rest world they see people like Zynga that are able to run a very distributed federated business, they have some data centers where they run an application half inside/half outside the public cloud, they move applications around, its a very loosely coupled highly federated business that’s run and these enterprises are looking at that and thinking ‘I need to have that same flexibility on my side’, and yes I can look at something like VCE and get a vBlock and plug it in but when I look at the metrics around that it doesn’t align with what I expect from my business metrics.

When I look at my IT department I want to measure those people as someone like Amazon or Google measure themselves. When Amazon sits down and looks at CAPEX and availability and OPEX and head count and all those things, they come out with a certain capacity per capita, like how many network switches do I have per networking engineer how many servers do have per sysadmin.

Actually when you look at some of the really big guys they dont even break it down that way, they say ‘how many switches and servers do I have per cloud admin’ They are starting to wrap their heads around a cloud admin or a cloud admin team that may have people that know little bit more about networking or a little bit more about storage but they’re still on the same team expected to have that same administrative breadth and the enterprises are seeing that unfold and are saying, ‘I need to have that same perspective around my IT, and I’m not going to get there by tweeking my current IT department, I’m gonna get there by starting a new IT department that is focused in this new way.’

Art Fewell: I think its been interesting to watch over the past, I guess its been over a decade now as cloud computing has marched through, at first it was like ‘wow that cloud stuff is cool’ and it grew to ‘ok, IT directors and CIO’s start to get fired we don’t have something similar’ and I think now businesses have gotten much more to the realization that if they are not able to operate their infrastructure in a manner that’s near the efficiency levels of some of these cloud deployments then we might as well pack it up and go home, compute has always been important but the extent to which it’s becoming even that much more important top core business operations is really profound. I think it’s becoming one of those things you just gotta do, you have to hit those efficiency targets.

JR Rivers: I think when you sit down and seriously study the industry what you recognize is that there are many existing businesses that do a great job leveraging hybrid private and public cloud,  I used Zynga as an earlier example and will continue to use them because they’ve been pretty public about how they do this. When they’ve decided that an application has matured they can run their own private data center much more cost-effectively at a much higher pace than they can by using public cloud. They grow new games in the public cloud because the ability to expand is very high. Once the game becomes mature they always move it back into private cloud and you’re seeing roughly the same things starting occur in the enterprise where in some cases they use public cloud, some cases they use private cloud and they want the flexibility to do both. They want to be able to look at public cloud as an expansion, not as a cost reduction but as an expansion, if you’re gonna have private cloud, it has to measure up to public cloud.

People use public cloud for a set of characteristics, expansion, maybe reliability, global reach etc. There is a whole series of reasons why public cloud might make sense to you. You don’t want it to be because it’s the most cost-effective way to do compute.

Art Fewell: If you had to pick one or two thing in the Cumulus community or on the roadmap, what do you think are some of the most exciting things for the Cumulus community in 2015?

JR Rivers: That’s hard because we have so much happening. There’s a new networking feature we are introducing that a lot of customers use especially enterprise customers, multi chassis link aggregation, some people call it MLAG some call it virtual port channel. It’s the ability to dual home a server in a layer 2 type domain and we’ve been working on that over the last four or five months. It’s being released, we have a bunch of customers that need it right now and it’s being used at scale in a couple interesting places. For us rolling this feature out gives a slightly bigger addressable market makes it easier for some customers to consume.

What was super exciting about that is you know I was involved both directly and peripherally with that and similar features being rolled out at Cisco a couple of times and it was usually large teams of engineers and a huge amount of time and we just got done doing a push of our rollout of that feature and it was about fifty man months, right and thats 50 man months, and thats from start to finish.

Art Fewell: It shows the type of agility you can achieve with some of these new platforms right?

JR Rivers: Part of it is the platforms, part of its the maturity of the underlying silicon, part if it’s the fact that we are working with the Linux networking stack which is a pretty rich networking stack. So you’re able to you make minor modifications that move the whole industry forward in that context. So its super powerful and great for customers and its pragmatic for us to develop and deploy on time.

Another cool one is we are working with a couple service providers around MPLS and carrier networking technologies. I’d say that’s early stage for us so that wouldn’t be rolling out until late this year in terms of general availability, but it’s super exciting to see what we’re able to do in the context of industry standard silicon with large carriers & service providers around MPLS.

The last piece that is kinda near and dear to us collectively is, we have a lot of people here that are very fixated and focused on automation, and one of the observations that we have collectively made, not just ourselves but people in the industry,  is that there’s not a great REST API around Linux.  So what we are starting to do is put together some programs where we facilitate a REST API. Its not the Cumulus Linux API, but it’s more about an API framework for Linux, you can use pretty much any Linux package to plug into it, you get a REST termination.

We have a great vision for it, we’ll see where it goes, we are starting to engage not only our resources but a couple of our customers and other friendly organizations around the globe in contributing into this. It is going to be really powerful to see that play out over the first half of next year.

Art Fewell: If you were to take another swing at that question but with a little bigger picture, let’s talk about the industry as a whole, whether it’s open computing open networking or wherever you want to go with it, what are some of the big trends that are catching your interest and excitement this year?

JR Rivers: The acceleration is what’s catching my interest and it’s got me really excited right now. When you look at what’s going on when we started Cumulus, and not just us, there’s a lot of really smart people in the world but, when we started Cumulus networks we talked a lot about what we were doing, we ran into people all the time that said ‘this is a great idea I love it, why didn’t this happen ten years ago’,  and while that feels good as someone agrees with you, this is just someone that’s talking right, they are not buying, not making a commitment, they’re just talking.

What we’ve seen over the last six months and we’re continuing to see this year is more and more customers are coming along and saying not just ‘I think this is a good idea’ but, ‘I want this, this is my thing.’ When we look at these customers they are not just looking at it from a networking perspective as you pointed out, they are looking at it from a full data center perspective. ‘I want an open ecosystem around my data center, I want cloud admins, I want to leverage all of these technologies to build infrastructure that allows me to build great business applications’ so that’s what really excites me, people are coming not just to drive costs down, its not ‘hey make me the same products you made me yesterday at lower costs’, but they are saying, ‘I need to build an infrastructure that allows me to build better business applications’

Art Fewell: If you look at VMware and Microsoft or enterprise private cloud specifically, it’s like the whole industry has been hyping people up about this for like 3,4 plus years now and we’re just finally getting to the point in 2015 where we are giving people that trigger to pull and I think from a Cumulus perspective, a lot of that is more open mindedness towards Linux and more interest in DevOps as a whole and I think as we get into the network side we are seeing more of that Linux interest carry over into ‘now I want to run my network that way’. Its going to be an interesting year to watch so many new things hit the fan.

JR Rivers: It’s kind of a coincidental collision of technology maturity and perspective, meaning perspective on the customer side right? I keep harping on this cloud admin perspective but I can’t tell you how easy it is of a filter when I talk to a customer when they talk in the context of having a cloud admin I recognize what they mean, what they want out of their system and if they don’t understand what I mean when I say cloud admin, then there’s a pretty big disconnect.

We’re seeing more and more companies – you know, we’ve been around for a while we’ve talked enterprises for two years or so and what we are seeing, when we talked to customers a couple years ago they didn’t understand what a cloud admin was. All of the sudden they are coming back and one of the things they are actually saying is ‘hey JR, I finally understand what you meant, I get it, we’re ready to go’. That’s what’s exciting

Art Fewell: And soon I think the next wave of craziness is very close to us, i’m talking to David Meyer in a little while and he always bring this up the a fact that inside of these Google-scale data centers, we have billions of transactions happening every second and so there is a lot of fear of the cloud efficiencies I think now that autonomic agents and the growth of autonomic agents is probably going to be a whole new inflection point really for the coming years.

JR Rivers: Part of its that, but part of its maybe slightly simpler from a technology perspective you know I’m when I grew up as an engineer I used to write a lot of assembly language programs. You’d be pretty hard pressed to find people to write in assembly language anymore because you can generate a much more efficient time spent on a Java app or Python or some super high level language like Ruby on Rails allowing you to go up and up the stack you can be way more efficient by using this high-level language and leveraging overcapacity of the infrastructure.

If you look back to when people started thinking about cloud computing it was a time realistically when networks were low capacity, servers were low capacity storage was slow and fragile and what’s happened is the technology has matured, the servers are blazing fast right now, storage is quick and low latency networks are becoming higher and higher capacity which allows you to, you know all these fears the people had around what does it mean to be cloud, why do I have to build some super-brittle infrastructure, those all disappear and you kind of end up with that computer, for those of you that are Star Trek fans remember the computer that was everywhere you could ask it anything? Well that’s we’re headed right? It’s this super high capacity computer that’s everywhere and can do anything for anybody.

Art Fewell: I’m really glad and grateful to have you on the show, you’re a big tech idol to the Linux community and really are THE guy when it comes to you Linux and networking and Cumulus has done such tremendous progress in helping to facilitate more open ecosystem and driving good things.