Interview: Stefan Dyckerhoff, Juniper EVP of Infrastructure

Stefan Dyckerhoff is old-school Juniper Networks Inc. (NYSE: JNPR), having started at the company in 1997. He's now the face of the company's hardware, having become the executive vice president of the Infrastructure Products Group (IPG).

He hasn't had an uninterrupted run at Juniper, though, as he defected in 2004 to Cisco Systems Inc. (Nasdaq: CSCO), where he ran several businesses, including edge routing.

But he was reunited with Juniper recently, and took over IPG in February, replacing Kim Perdikou in what looks like a pretty major shift for Juniper's upper ranks. (See Juniper Moves Emphasize Carriers & Software.)

Juniper is known for grand proclamations about "The Future," as exemplified by some of the speeches at its analyst conference in February. Dyckerhoff and other Juniper folks refer back to those talks repeatedly.

And in an interview session with Light Reading, Dyckerhoff got his chance to paint big pictures.

Juniper sees a chance for the network to follow the path laid down by computers, where the PC (or the iPhone) became a platform that draws its real power from untold numbers of applications. So, too, should the network become an application spawning ground -- and that thinking has led Juniper to opening up its software, allowing others to write for Junos, in hopes of bringing the network up to platform status.

It's a grand vision that could take years to unfold, and it's not all necessarily going to go Juniper's way.



Table of Contents:

— Craig Matsumoto, West Coast Editor, Light Reading

Light Reading: I want to ask you about Juniper's current mission, as opposed to 1997. I was talking to someone about the early days of Juniper, how the company targeted one problem and just put a ton of engineering on that one problem.

Dyckerhoff: That's right.

Light Reading: Now you're bigger, so what would the focus be?

Dyckerhoff: One of the reasons that we rebranded the company, and why we're much more vocal about what our long-term vision is for the industry, is that it's a new phase for the industry. It's also a new phase for us.

I mean, we delivered some fundamental innovations in the early years. How do you use silicon to drive up performance and drive down cost per bit? How do you address a software problem to address higher feature velocity and high reliability? Those are still part of what we need to deliver to be successful, but now it's about: How do you actually use this technology to change how our service provider customers and enterprise customers use the network to drive their businesses?

Most of our competitors will say: "What you need to do there is go end-to-end, not just on the network, but also up the stack. So you need to buy all my applications and so forth." That's what I call my "IBM in the '80s" pitch. From our point of view, that slows down innovation on top of the network infrastructure, which is at the heart of all our customers' businesses, not just the service providers. And it doesn't address the economics.

Light Reading: It gets harder, then, doesn't it? You're a big company. You have to worry about more things at once.

Dyckerhoff: Exactly. One thing will not cut it. So, we've been placing bets on how security will evolve, how the data center will evolve, and then, fundamentally, on how software will evolve.

You have to be able to provide a technology platform that goes across those areas. You don't have to own every application that goes on top, or even every single piece.

We've licensed Junos to one partner. There'll be only a few of them, because by definition, it's a very tight integration. But as we go to the application space there'll be a lot of partners on top. That's what I mean by providing a technology platform as opposed to providing a fully integrated solution by one vendor.

The customer still asks us how to put that technology in their network, and we still need to, with our partners, provide some professional services or some integration services. But the basis for us is that we provide an open technology platform on top of which ride all the applications that you might want to run on your network. I think we've only seen the tip of the iceberg. Every application will use the network in some way going forward.

Light Reading: That's a different view from Cisco's, but you still draw inevitable comparisons to Cisco. When you bought Ankeena, the headlines said, "Juniper Follows Cisco," because Cisco is so keen on video.

Dyckerhoff: That's right. But the Cisco strategy is: "I will own all the assets, and I will create hooks between my infrastructure and telepresence and my set-top box, and I'll system-integrate that." That's a valid strategy. Our strategy is a little bit different, because again, we think there will be thousands of applications on the network that will drive some revenue. So, we're saying we're going to provide more general-purpose, open solutions to this problem.

The reason we think this will win is because customers will want to do many, many, many things on the network, much more than any one vendor, regardless of size, can develop.

Light Reading: It's interesting, the way software is becoming more important. I've heard people talk about that for 10 years. Now you're even opening up your software. (See Juniper Opens Up to Apps Developers.)

Dyckerhoff: Even in 2001 and 2002, we were starting to think about opening up Junos and having SDKs [software development kits] and so forth.

The reason this now resonates extremely well with our customers and partners all goes back to the economics. It was a smart idea eight years ago, but nobody put a gun to your head in terms of the economics of what you were doing. Now, customers see the end of the runway in terms of how they do their business.

Light Reading: Is that what you mean when you say the industry is moving to a new phase?

Dyckerhoff: The way that we've been building the networks just doesn't work. It's one of those issues where everybody sees it, but nobody knows exactly what to do about it.

Sometimes when we put up the chart of building these sorts of siloed vertical networks, people think of ATM or the old Frame Relay networks. But in essence, the industry built the first IPTV networks the exact same way and built the mobile networks the exact same way. We need to fundamentally reformat how the network is built, and then how we put applications on top of the networks.

We can be smart about how we build the network and drive lower cost. That addresses the capex problem. Then you have to address the opex problem, which we think you can only address through automation. There's a lot of room for automation, through the Junos Space platform, or through the various open APIs [application programming interfaces] and SDKs, to actually reduce the total cost of running the network.

You also need to look at what the top line looks like for the service provider. There's another chart that everybody likes to put up, but that there aren't any good answers for today, that says we're going to go from three or four killer applications that generate 90 percent of the revenue for a service provider to 20, 30, or 50 applications. That's a big opportunity for the service provider, but today, the incremental cost of every service they roll out is extremely high. When you have 50 applications that make up the addressable market as a service provider, the incremental revenue per service is lower, and so the cost of introducing a new service has to be lower. Otherwise, you can get some growth on the top line, but in terms of profitability, you haven't solved the problem.

So, for us, then, it is the new applications on top of the network, whether they are ours, Nakina Systems Inc. 's, or partner builds -- partnership with Feeva Inc. on advertising, partnership with other folks like IBM Corp. (NYSE: IBM) on automation and integrating with the back-office software. As the service provider business model changes, they need to be more agile and have a lower cost of introducing a new service.

Light Reading: Software seems like it's a harder sell than hardware. To investor types it's easier to say, "We're gonna do 100 Gbit/s."

Dyckerhoff: That's right. That's a transformation, and I would say with our largest service provider customers, at the CxO level, this resonates extremely well. In the SP space, there'll always be the war over the newest fastest interface, the newest system -- that's baseline. A company like us, we have to be at the leading edge there, no matter what. This [the software part] talks about how we grow our business model and how we make the connection for our customers between what they buy in their infrastructure and how to fix their business.

And we are serious about creating a new business here. You probably saw the org announcement. We made this a business group [called Junos Ready Software], and Manoj [Leelanivas], who sits next door, runs it, to start thinking not just about how we monetize it, but how we build a business on top of software. But I would distinguish between what is the biggest part of our business today, and what are the strategic pieces that we put in place to grow in the medium and longer term.

Light Reading: A lot of people expected you to have bought something optical by now. [Ed. note: This interview took place before Juniper bought a $3 million stake in ADVA Optical Networking .] You're partnering with Nokia Networks instead -- why do it that way instead of building your own packet-optical box?

Dyckerhoff: We definitely see the need to have integration and orchestration between the Layer 1 transmission and that MPLS layer. The fundamental thing that carriers will think through is: Either you keep it all optical -- which you can do, but the units are very big: 10 Gbit/s today, and it'll go to 100 Gbit/s very quickly -- or, you make use of the dynamic control plane and the statistical multiplexing with MPLS. That will be the next step.

As we model the cost, we think that definitely wins. [The MPLS model, that is.] We've done it with customers, and we've done some studies with academia -- Carnegie Mellon University -- to model the different ways to architect the network. What you find is that even at a higher per-port cost, you get much better economics when you go to MPLS. Statistical multiplexing and dynamic provisioning are the two basic reasons. That's the architectural outlook of how we think this will happen.

From there, we definitely think that tighter integration with the optical layer is something that we need to do, both from a software provisioning point of view as well as from a hardware point of view -- so, integrating WDM components into our systems. So, we have a set of suppliers that we work through, like Opnext Inc. (Nasdaq: OPXT) and Finisar Corp. (Nasdaq: FNSR), and a set of partners. There's a lot of fundamental technology that goes into these optical components, not all of which we want to take on.

It's been interesting with Verizon Communications Inc. (NYSE: VZ) and how they've approached the discussion.

Light Reading: You're talking about Verizon being interested in using packet-optical transport systems (P-OTS) to do some core-router bypass. (See Verizon Rethinks Long Haul.) They don't seem interested in the MPLS approach.

Dyckerhoff: That's one of the reasons that we did the financial analyst discussion with Pradeep [Sindhu, Juniper's founder] and myself -- I think that's available on the Website, yes? -- because a lot of the financial analysts said, "I don't think Verizon believes you."

Then, at CTIA, Dick Lynch [Verizon's CTO] gave a keynote and said, "We think the future is around MPLS plus optical."

Light Reading: That's interesting.

Dyckerhoff: I remember Nik Theodosopoulos [a UBS AG analyst] was very excited. He came back to our booth and said, "This is fantastic! This is what you said!"

It's not a simple transition, but eventually, I think there'll be broad alignment across the industry. Now, which path we take and how those competitive dynamics play out -- that's a separate discussion.

We still see innovation in the optical space, particularly focused around the sweet spot of medium-length transmission, or metro -- sort of 600 to 800 km. Lots of customers are trying to tackle that. And at the same time, we see that the need for capital for every development really goes up, so there's some consolidation in terms of suppliers. Definitely an exciting space to work on, and lots of moving pieces.

Light Reading: An exciting space to work on, if you're not the one spending the R&D money to do it.

Dyckerhoff: We spend a lot of time, really, breaking down what's in a transceiver, as to how the economics of those different pieces will work out and what lends itself to integration and what's best left to a set of well-funded physics Ph.Ds out there. We have a detailed view of how we think this works out. It's a dynamic space. We're very engaged with a number of companies both big and small to make sure we have a healthy supply of technology and a healthy ecosystem.