"Service providers have to make IP profitable within a year or they are going to go under"

January 25, 2001

14 Min Read
Dr. Lawrence Roberts

3390.gifThe Light Reading Interview

In The Spotlight: Dr. Lawrence Roberts

This was a big week for Light Reading. Dr Lawrence Roberts, one of the founding fathers of the Internet, was “in the house” at our Tribeca trophy office in Manhattan.

We did the right thing, of course. Lined up the staff. Scattered the rose petals. Chanted “Larry! Larry!” And, naturally, made sure to snap a photo with the Great One in front of our new salt-water aquarium. (We’ve just named our newest fish “Dr.Huber,” incidentally, on account of the fact that it doesn’t say much).

3390_1.gif Larry hasn’t been saying much recently, either -- an unusual state of affairs, for him. But at least he has an excuse: Caspian Networks (née “Packetcom”), his latest venture, is still officially in stealth mode.

Fortunately, we found Larry in ebullient form, and he was generous enough to drop some new and exclusive details about Caspian’s product. (His decision to make said revelations appeared to come as a surprise to his marketing handlers, who are about to discover that retroactively saying “that was off the record!” after the company founder has just said something really interesting doesn’t actually stop journalists from writing it down, attributing it, and publishing it.)

Caspian is building a box targeted right at the meat of the core Internet router market and, by extension, at the current router duopoly of Cisco Systems Inc. (Nasdaq: CSCO) and Juniper Networks Inc. (Nasdaq: JNPR).

Clearly then, a nervy move on Larry’s part. And with terabit router companies like Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7), Charlotte’s Web Networks, and Pluris Inc. already vying for the number three spot in the router game, why is it even worth giving serious consideration to Caspian (let alone give it the number five spot on our list of the Top Ten Private Companies? -- see Caspian Networks).

A few reasons: First, Larry doesn’t want to be number three in routing -- he wants to be number one. Second, Caspian isn’t building an IP router –- it’s building a mondo switch with some snazzy signaling smarts that could really set it apart from the hoi polloi.

Third, Larry’s interviews are always of value. For journalists, interviewing Roberts is like a refreshing day at the spa: Warnings are issued. ("Service providers have to make IP profitable within a year or they are going to go under.") Entire swathes of competitors can be dismissed with a wise smile and a two-word epithet ("too edge"). And, of course, answers -- definitive ones -- are delivered with an air of absolute authority. ("There’s got to be a shakeout.")

It’s all choice stuff. And Roberts's pitch has already gone over big with a number of venture capital firms and banks, which have bankrolled Caspian to the tune of $140 million in three rounds of funding (see Caspian Prepares for an IPO and Internet Pioneer Plots IP Revolution).

Not everyone’s a fan, of course. Witness this missive from a partner at a venture capital firm in California, who cravenly requested anonymity:

“Larry Roberts may be an Internet pioneer, but he has a long history of cooking up 'technology religious' ideas that are not commercially viable. As a result, he shopped Caspian (then Packetcom) in the valley for 12 months or so before he got someone to bite. As times have changed he's added some optical themes to the punch for effect. Larry Roberts is no Dave Huber and Caspian is no Corvis.” (Begging the question: Is this a bad thing?)

Make up your own mind about Doc Roberts’s new gig by reading the following interview, in which he also talks about:

Edge vs. Core
The Business Case
The Competition
OOO versus OEO
The Future of Caspian
3390_2.gif Light Reading: Larry, how are you doing?

Larry Roberts: Doing well. Very well.

Light Reading: When we last talked Caspian was called PacketCom.

Roberts: Yes, it was called PacketCom, but then they decided they needed something a little more broad. Something that they could trademark.

Light Reading: Who’s “they”?

Roberts: I brought in a whole team: Graham [Graham Rance, Caspian’s CEO] and marketing people. Marketing started working on how we could get a real name we could handle. With PacketCom, I knew I could get it [as a URL] on the Internet, and that’s the only check we made.

Light Reading: Did somebody else have a trademark for PacketCom?

Roberts: No, it’s just that “packet” is generic. There was a 50/50 chance that we would have a problem with it because of its generic nature.

Light Reading: What does Caspian mean?

Roberts: Nothing. It’s a sea.

Light Reading: Yeah?

Roberts: Yes. We just wanted something like Exxon that’s totally neutral.

Light Reading: Obviously, you’re not yet talking about your products. But we’ve poked around and written a couple articles about what sort of product we think you’re working on. How close to the mark have we been?

Roberts: You’ve actually been pretty good.

Light Reading: So what are you up to, eh? Making switches with some sort of very fast signaling for ATM [asynchronous transfer mode] and IP [Internet protocol]?

Roberts: We’re not going to talk about what we are doing with IP calls yet. And we’re not using ATM signaling. It’s more like we’re changing the IP switch to do something new, and different. What we’re really trying to do is change the switch, and not the protocol.

The challenge is to take all this traffic that is coming into the switch -- which could be made up of different protocols, or just IP -- and send it out over the network with QOS [quality of service] and load-balancing. And to do this reliably. So the switch just has to be more intelligent. It has to have more knowledge of what’s going on [on the network]. Thirty years ago, we told everybody to make the switch dumb. We tried to get it so the core was dumb and the edge was smart. But now that’s not the way to go -- we can make the switch more intelligent.

Light Reading: So people who say the core should be dumb with an intelligent edge are wrong?

Roberts: I think in the short term they’re wrong, yes. The core is going to have to be more intelligent [to do all the things it needs to do]. In the long run things may shift again. Once the edge is doing exactly the right things it may be possible to make the core less sophisticated.

Remember that IP is the network protocol in the core. We can’t change it overnight. That means all of the nodes in the network have to be intelligent. If you can start changing the protocol later, then you can start thinking about making [those nodes] less smart. But not right away.

Light Reading: OK, so you’re talking about a very intelligent switch, and one that handles IP in the core and multiple protocols at the edge of the network. Isn’t there a danger that you are trying to do too much, old chap? Is there a risk that you’ll fall into the trap of putting too much functionality in one device?

Roberts: I don’t see that. Multiprotocol isn’t very painful to install. The simple way to handle it is to translate everything into IP. You don’t want to fool around with multiple protocols inside the switch. The point is, if you make IP capable of handling all the QOS and so on, now the network can do what you need to do and the translation becomes trivial, because it’s only happening at the edge of the network.

Light Reading: You’ve got some edge switches translating multiple protocols into IP.

Roberts: I’m not sure we’d call our product an edge switch. When it’s installed at the edge of the network, there are multiprotocol interfaces. But the switch certainly serves in the core, as well. It’s definitely a core switch.

Light Reading: So it sounds like you’re not making two different hardware platforms [edge and core], you’re making two different layers of functionality you can enable in the same product.

Roberts: That’s a better way of talking about it.

Light Reading: What else can you tell us about your product?

Larry Roberts: Reliability is another thing we are working on. Service providers need equipment with carrier-grade reliability. Equipment that can stand up to upgrades -- so that they can perform software upgrades and fix failed boards and offer uninterrupted service for five or ten years, or until they see a return on their investment.

And right now they don’t have that. The feedback that we’ve gotten from all the service providers is that products should have hitless software upgrades and hitless cutovers for BGP [border gateway protocol] when a failure occurs. And it’s the hardest thing to do – especially for BGP. None of today’s router vendors have succeeded in doing it.

What does that mean? It means that in any network today carriers have to have complete duplication of the [routing] hardware. They have two have two duplicate pieces of equipment sitting next to each other [for redundancy]. They’re both live; they’re both working, but at half their capacity.

And that’s not the only problem. The lines are working below capacity because of QOS inefficiencies. If you look at networks today, some of them are running at 15 percent to 30 percent utilization.

And they don’t have any way to manage load balancing on the network, so they have to do that manually. They’re having to scale [the number of engineers they hire] linearly with the traffic. This is a major cost for them.

So to sum up, the first network cost is to light the fiber. Then the cost of the switching equipment. And then the operational costs -- of the people that run it. And right now these guys are sitting there with data accounting for 80 percent of their traffic, but only 20 percent of revenue. And data will be 90 percent of their traffic next year, so they better make IP profitable within a year or they are going to go under.

Light Reading: And Caspian will help them do that?

Roberts: Yes. Only a slight change in the cost of the switch can bring them to profitability. And then the IP SLAs [service level agreements] can make them very profitable.

Light Reading: Let’s talk about your competitors. In terms of routers, they’re people like Avici, Cisco, Juniper, and Pluris. Would you include Foundry Networks Inc. on that list?

Larry Roberts: No. It’s Juniper, Avici, Cisco, IronBridge, and any others in that market. But the technology is nothing like any of those people.

Light Reading: How does this compare to optical switches from Sycamore Networks Inc., Ciena Corp., Corvis Corp., and Tellium Inc.. Does it compete with them, or is this a complementary technology?

Roberts: All of the optical products fit in a layer below ours, where they are used to manage the structure. Optical switches don’t do anything [to help] switch traffic. The lambda is too big an entity — it’s not granular enough. So you have to do packet switching. Optical is very cheap, but it’s not going to be a big part of the revenue picture. Optical ports will be 50 times cheaper than packet-switching ports. The real challenge is not how do you install optical end to end; it’s how do you use those optical channels? The challenge is: "How do I load those pipes?"

Light Reading: So to be clear, those companies will continue to build those products, and you will work over those products.

Roberts: [The optical layer] will just be underneath.

Light Reading: You’ve called your product an optical superswitch. What is optical about it?

Roberts: All the [external] interfaces are optical. Optical interfaces, with ASICs.

Light Reading: Are you doing the ASICs [application-specific integrated circuits] yourself?

Roberts: Well, we’re going with IBM.

Dallas Kachan [Caspian’s director of communications]: That hasn’t been formally announced yet.

Light Reading [laughing]: Until this afternoon. But you’re just talking about the manufacturing, not the design.

Roberts: The design is ours.

Light Reading: What about inside? We’ve heard negative feedback that this is PacketCom with an optical spin. Is that unfair? I mean, is there an optical crossconnect or an optical backplane in here?

Roberts: Yes, all of our interconnects between the external components are optical.

Light Reading: How do you set yourself apart from companies like Amber Networks Inc., which talks about carrier-class switch routers, or Village Networks Inc., which has an IP/optical product that also uses optical interconnects?

Roberts: The only people to compare us with are the petabit switches, because the other people are sitting in a different market. They’re too edge.

There’s another issue that’s subtler. One of our potential customers decided they’re just not going to buy any switches that aren’t 19 inches wide any more. They’ve made a formal decision to not buy 23-inch boxes, because they can’t fit them in their sites. Weight is another requirement. They don’t want to have to reinforce floors. These are the kinds of things that people should have been checking on before now.

Light Reading: It’s back to the fundamentals, isn’t it? How much does it weigh, how much does it cost, and will it burst into flames?

Roberts: [Carriers] are really looking to scale their networks over a reasonable period of time. Three years, for example. It really is a very serious problem for the service providers, because they can’t keep replacing [their equipment] all the time.

Light Reading: Where do you stand on the debate over the merits of OOO [all optical, or photonic] switches and OEO [optical-electrical-optical] switches?

Roberts: I don’t see any benefit to doing OEO if you can do photonic. Photonic is cheaper.

Light Reading: OEO does allow you to see, or manage, the traffic though, right?

Roberts: If you look at the photonic switches that are out there, whether it’s Corvis or anybody else, they claim that they’re all photonic -- but in reality there are OEO components too. The grooming of traffic and the ability to add an optical stream is something that is fundamentally important, and a pure photonic play will not be able to address that.

Light Reading: What about these really big photonic switches that have 1,000 ports?

Roberts: There’s no use for them yet.

Light Reading: Interesting that you should say that. I heard that there are only maybe 70 sites where you could install them.

Roberts: There’s no use for them anywhere today. The biggest site we can see at the moment has got maybe 10 OC192 [connections].

Light Reading: So the whole OOO market is not really big? And yet, there are about 100 startups trying for a piece of it.

Roberts: I have a similar problem with it. And I have another problem. Once you successfully do photonic switching, it’s so cheap, where will [photonic switch manufacturers] get the revenue? Let’s just count the ports and how many there are going to be and what the market size is going to be. It’s actually in the hundreds of millions at the best.

Light Reading: What about the ultralong-haul market? Some people also think that is a niche market, but there are a lot of players there also.

Roberts: Oh, there’s going to be a shakeout.

Light Reading: What are your plans for equity events, or an IPO? How manyemployees do you have?

Larry Roberts: Three hundred and twenty-three. We've done three privateplacements, we'll probably do one more, and then an IPO some time afterthat.

Light Reading: Have you shown anybody the product?

Roberts: We're working on the box with potential customers. We've notdelivered equipment to anybody.

Light Reading: These days, service provider customers sometimes ask forequity from equipment vendors. Sometimes individuals at those companies evenask for equity. Have you experience any of that?

Roberts: We've seen enough of it that we're going to be very careful. That'scome up. We're not going to do that.

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like