8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`The n -Dimensional Superswitch
`Larry Roberts has a next-gen router he says will kick Cisco's ass – oh
`yeah, and reinvent the Internet. Lawrence Roberts is not a patient man.
`But he's trying. He's leaning over a table in the San Jose headquarters of
`his new company, Caspian Networks. The building's exterior is standard
`fare – one more box […]
`
`L ARRY ROBERTS HAS
`
`https://www.wired.com/2001/05/caspian/
`
`1/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 1
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`JOSH MCHUGH
`05.01.2001 12:00 PM
`a next-gen router he says will kick Cisco's ass - oh yeah, and
`reinvent the Internet.
`Lawrence Roberts is not a patient man. But he's trying. He's leaning over a table in the San
`Jose headquarters of his new company, Caspian Networks. The building's exterior is standard
`fare - one more box in a Silicon Valley office park - but this spartan corner nook is at the end
`of a curving yellow hallway lined with eerie purple lights. The shades are drawn, and the
`leathery, 63-year-old engineering legend is pointing out the shortcomings of the packet-
`switching architectures employed by his competitors - dullards like Cisco Systems, Juniper
`Networks, and Avici Systems. "They're using hypercube or hypertoroid topology, so they're
`limited to six dimensions, really - three up and three down," he explains. "I've been able to
`take more steps, to go into n-dimensional space."
`His eyes bore into me from beneath a pair of bristling eyebrows, and soon he realizes that
`these terms are not ringing a bell, that n-dimensional space is not my natural habitat. So he
`switches gears. "Look. It's like you've got a bunch of dumb gorillas working on something," he
`says. "You can do it a lot faster and better if you get some smart people." It sounds like a crude
`and condescending remark, but for Roberts, it's a multidimensional metaphor. In a technical
`dimension, the gorillas symbolize the IP routers at the core of the world's long-haul optical
`data networks, while the smart people are Roberts' new creation, an intelligent "superswitch"
`that he insists will replace them. In a competitive dimension, his "gorillas" are also the core-
`routing heavyweights - Cisco and Juniper - while the smart people represent (surprise!) the
`engineers at Caspian.
`h
`h
`d
`d
`h l
`h
`b
`b
`h
`h
`h
`k
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`2/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 2
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`More than three decades ago, while a researcher at MIT, Roberts began hatching the network
`designs that would evolve into the Internet. For the last three years, he's been holed up with
`what has become a cadre of 200-plus engineers nourished on a $140 million diet - provided
`by Paul Allen's Vulcan Ventures, WorldCom, Lucent, Merrill Lynch, and others - working on
`what he promises will be one of the biggest things since the birth of the Net. Now Roberts is
`just eight weeks from launch, when he'll announce a major backbone operator as a flagship
`customer. And while neither that company nor any other carrier - including WorldCom -
`would comment on Caspian's innovation prior to launch, Roberts has agreed to give Wired a
`sneak preview.
`To the untrained eye, Caspian's product, the Apeiro, is a new kind of router. But Roberts says
`it's not a router at all, because where traditional routers are "dumb" - Roberts' shorthand for
`the fact that they don't differentiate between the kinds of bits running over a network - his
`"optical IP superswitch," as he calls it, is smart. It can identify packet types (voice, text, video,
`et cetera) and priorities, allowing it to determine one packet's relation to others, and expedite
`traffic in a way that's impossible today. For example, the Apeiro will recognize all portions of a
`videostream and label them as part of a greater whole so they can be more efficiently slotted
`and moved to their ultimate destination.
`This may seem like a minor improvement, but Roberts says recognizing and prioritizing
`packets of data - coupled with the fact that racks of his devices can be stacked together to
`increase performance - will make the Apeiro as much as 1,000 times faster than typical core
`routers in use today. If those speed estimates prove out, Roberts says it will mean supercheap
`worldwide IP telephone calls clearer than those on the voice network. For people with high-
`speed Web access, it'll mean on-demand full-screen streaming of high-definition video. Even
`the dialup crowd will be able to pull up Web pages at least twice as fast.
`Ultimately, Roberts says, the Apeiro will deliver an Internet that finally lives up to all the
`broadband hype - a platform that will dislodge every other entertainment and
`communications medium. "All entertainment is going to shift to the Web," he says. "You'll be
`able to get the programming you want, the way you want it." On the way, he promises,
`Caspian will reshape the routing landscape and put an end to the notion that the Net works
`best as a dumb network.
`Of course, the fact that our Internet is dumb at all is mainly Roberts' doing. In 1967, while
`translating Leonard Kleinrock's theoretical work on packet switching into the Arpanet, he
`designed the network so that the complex routing decisions were made on the fringes. The
`core machines were to be as dumb as possible. For most of the 34 years since, a lot of people
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`3/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 3
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`core machines were to be as dumb as possible. For most of the 34 years since, a lot of people
`have seen this dumb-core, smart-edge setup as the very essence of the Internet. But Roberts
`says that's not the case. "Cisco thinks the network has to be dumb because we used to say,
`'Keep it dumb,'" he explains. "They're doing it in a clunky, old, expensive way."
`As it turns out, Roberts and his cronies (Kleinrock, Vint Cerf, and Robert Kahn) simply wanted
`to make sure that they, not the telecommunications carriers, had control over the fledgling
`network. They designed it so the complex computing happened in the university computers
`they had access to. "We were afraid a smart switch would get in the way of our experiment,"
`Roberts says.
`With telecom analyst firm RHK estimating that the core-routing market will grow from $2
`billion in 2000 to $12.5 billion in 2003, the sort of architecture shift Roberts is suggesting
`could mean a windfall for Caspian. But of course Cisco is not about to hand the game over to
`an entrepreneur - no matter how respected he may be. In fact, Cisco and company have
`suspected for some time that the surging volume and complexity of Net traffic call for more
`routing intelligence, and they've been designing switching fabrics of their own. The trade
`press has been eagerly anticipating the Caspian launch, but Cisco doesn't seem concerned.
`"Every 18 months or so, somebody comes along and says 'We're going to reinvent routing,'"
`says Rob Redford, the company's senior director of marketing. "Ipsilon Networks came in with
`an IP switching plan," he says, referring to a networking startup touted as a Cisco-killer before
`it was quietly acquired by Nokia in 1997. "That fizzled."
`But if you listen to Roberts, the incumbents are doing little more than tweaking a
`fundamentally dumb design. He singles out Internetwork Operating System, the software
`running Cisco's routers, as an example. IOS, he charges, has evolved from a time when core-
`routing functions were much simpler, bulking up with layer upon layer of software to adapt to
`new demands. "You can't have a monolith like IOS running a 10-terabit switch," he says. "The
`only way to get where we're going is to start from scratch. And Cisco doesn't want to redo
`IOS."
`The select few who have been party to the Apeiro's prelaunch blueprints say Caspian is about
`to redefine the way the Internet works. "What they're suggesting is revolutionary," gushes
`Matthew Carpenter, a VC with Salomon Smith Barney, which ponied up part of the $87 million
`financing round Caspian raised in December. "They take the issue of routing to a different
`level."
`A different level, indeed: Cisco says its fastest new core routers, the 12400 series, can be
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`4/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 4
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`hooked together to process up to 5 terabits per second. Pluris, another newcomer, has a
`multirack system that will churn as much as 19 terabits per second. By contrast, Roberts says
`an Apeiro system - made up of scores of interconnected 7-foot racks filled with switching
`cards, each 23 inches deep, three-quarters of an inch wide, and 15 inches tall - can handle 160
`terabits per second, with petabit capacity on the way. And the analyst community is backing
`him up. Says Muayyad Al-Chalabi, a director at research firm RHK, a rack of Apeiro switching
`systems "can get bigger throughput than anyone else - by an order of magnitude."
`To make a fundamentally smart switching machine, Roberts had to scrap the routing
`blueprints. Beginning in the spring of 1998, he and a handful of engineers cranked out new
`designs for wiring, software, and chipsets. But between the Apeiro prototype and market
`success are any number of hurdles. For starters, Roberts' leap in functionality mandates a leap
`in complexity. Caspian has contracted IBM to manufacture the ASIC chips that will run the
`Apeiro, but that will be no easy task - even for IBM. The most advanced routing chips in use
`today can handle only three dimensions; Caspian's chips must maintain awareness of two
`dozen routing dimensions at once. Meanwhile, the software, which will run in a distributed
`manner - controlling each switching card separately to isolate failures even while functioning
`as a single entity - has several months of testing and debugging ahead.
`__Dumb core, smart edge was the essence of the clunky, old, expensive Net. With Apeiro
`prioritizing packets, Roberts says, it will finally get smart. __
`The reward for clearing these technological hurdles: Roberts gets to bring his baby into a
`crowded market to compete against the likes of Cisco, Juniper, Avici, and Pluris. And he'll have
`to do it without Grahame Rance, the 18-year Nortel veteran Roberts recruited as Caspian's
`CEO. In mid-March, Rance abruptly left Caspian to run electronics component maker SBS
`Technologies, causing some observers to ask, "Is there a problem they're not telling us about?"
`Together it all seems daunting, but the good news for Roberts is that the exponential growth
`of Internet traffic over the last few years has made carriers like WorldCom, Sprint, and AT&T
`receptive to anyone who can help them handle the deluge. It's what allowed Juniper, which
`crashed the party in 1999 with a faster router, to promptly grab a third of Cisco's market share.
`And with Cisco's recent announcement that it will sack as many as 8,000 workers, the router
`monolith appears less intimidating than it once was. "In the days of single-digit growth,
`incumbents had a leg up," says RHK's Al-Chalabi. "Now, the phenomenal growth encourages
`carriers to try new stuff. If you're the top dog today, there's no guarantee you won't be out
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`5/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 5
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`y
`y
`p
`g
`y,
`g
`y
`tomorrow."
`Wall Street's recent infatuation with optical networking has ensured that any company whose
`products come within a light-year of a fiber-optic cable touts itself as optical. But in reality,
`there are no true optical packet switches or routers on the market yet - at least not in the
`sense that engineers mean when they use the term. What many of the so-called optical
`networking companies actually do is optical circuit switching - connecting big pipes to other
`big pipes.
`Caspian also identifies its wares as optical networking equipment, but - as with Cisco and
`Juniper - the company is optical only in the sense that its product sits adjacent to the optical
`backbone trunks, converts optical data into electronic form, decides where the packets should
`go, and converts them back into photons before moving them along. So, the real promise of
`Roberts' innovation has less to do with photonic optical switching - routing data as light
`without electrical intervention, which Roberts says is still many years away - than with
`advancing traditional routing to the next level. In short, Roberts wants to make packet routing
`as dependable as the switching that occurs in a voice network, until pure optical switching
`arrives.
`"Our software is designed so it will run nonstop. That's never been done in a router," says
`Roberts. "When you're making a phone call and there's an outage, the system switches the call
`and you can't even tell anything happened. On the Internet, you get a 'server not available'
`message."
`The idea of switching will make the carriers more comfortable, since the current system of
`routing is costing them a lot of money. The IP traffic explosion should have been a bonanza
`for the carriers, but it hasn't. Because the backbone routers can't differentiate between
`packets, they move all traffic as quickly as possible. As a result, carriers are unable to charge a
`premium for high-priority data like voice and video. What's more, today's core routers aren't
`smart enough to handle a port failure without shutting down, so carriers have to buy two core
`routers for every one they use - paying for twice as much equipment and capacity as they
`need.
`They're also buying more often. Until a few years ago, a carrier would typically replace the
`core routers every 18 to 24 months - with a system from Cisco or Lucent full of OC-192 ports
`costing around $2 million. The IP boom has only accelerated that cycle. As a result, the chief
`financial officers at the major carriers are reeling. Caspian's top selling point with these CFOs,
`according to Roberts, is the Apeiro's ability to scale up without requiring a whole new system.
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`6/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 6
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`according to Roberts, is the Apeiro s ability to scale up without requiring a whole new system.
`Roberts says the Apeiro will also create new revenue streams for the carriers by solving the
`"voice and video problem." IP voice and video, unlike email and static Web pages, breaks
`down dramatically if there's a delay - as little as a few milliseconds - in getting packets from
`host to recipient. To guard against delays, Roberts and his team designed and programmed
`the Apeiro to examine incoming packets (about 4 million at any given instant) and decide
`when to put them through. As a result, once the lead packet in a videostream has been
`identified, examined, and assigned a priority, the rest of the packets in the session need only a
`cursory identification to be assigned to the "flow" that Caspian's switch has designated.
`While Caspian competitors attempt to similarly separate packets by priority, with a method
`called DiffServ, Roberts says DiffServ works only if a router has lots of excess capacity. The
`Apeiro technique is based on a standard called multi-protocol label switching, which Roberts
`has tweaked and renamed D/MPLS - the D is for dynamic. It's one of seven Caspian-related
`patents that Roberts has applied for. The result will be that carriers can assign higher priorities
`not just to types of traffic, but to traffic coming from a specific customer - allowing them to
`offer varying levels of service at a range of prices. Such differentiated service could be a boon
`for the carriers. "This is a revolution in routing," says David Yedwab, a former Bell Labs
`engineer who works as a consultant to carriers. "It's a technical breakthrough."
`Inside the Caspian switch, Roberts has created a mini Internet. Between the main switching
`cards lies a switching fabric, a local fiber-optic network containing a group of processors that
`act like tiny routers, finding the optimal path for each packet. Avici and Pluris also feature
`switching fabrics, but Caspian's design sets Apeiro apart. Roberts says the competition's use of
`hypercube and hypertoroid designs - internal wiring schemes for moving bits through the
`router - limits them to three dimensions. "They probably saw that in some book and figured
`that was as far as they could go," he scoffs.
`Roberts realized that, in theory, there's no limit to the number of dimensions along which
`packets can be routed, assuming the chips can keep track. For now, the n in his n-dimensional
`design is 24 - but only because that's as much sophistication as the first round of processors
`and software can handle.
`__Little surprise carriers call Apeiro "a revolution": Prioritized packets means preferential
`treatment - and differentiated service at a range of prices. __
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`7/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 7
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`It's this move toward unlimited switching dimensions that has the carriers, the investment
`banks, and investors salivating. The upshot: If Caspian's chips come back from IBM's foundry
`just right - itself no mean feat - and if the software is completely bug-free, Roberts will have
`designed a switch that won't have to be tossed within 18 months. Instead, it'll grow to
`intelligently handle as much traffic as the Internet can throw at it for years to come.
`Lawrence Roberts grew up with a professional-grade laboratory in the basement of his home
`in Westport, Connecticut. His parents, Elliott and Elizabeth Roberts, both earned PhDs in
`chemistry, but young Larry preferred electrical engineering. While other 8-year-olds were
`building tree houses, Roberts put together a Tesla coil. A few years later, in the early '50s, he
`told his parents they should get a TV. They made a deal: They'd pay for the components if he'd
`put them together. In high school, while on summer vacation, Roberts created a transistor-
`based telephone network for the Girl Scout camp his parents ran in nearby Darien.
`He went on to attend MIT, where his projects grew even more interesting. One winter evening,
`he and some friends decided to demonstrate the properties of sodium in a spectacular fashion
`- in the form of an ice bomb. "We took a large piece of sodium, wrapped it in paper, and
`dropped it on the ice on the Charles River," he says. "Once the water got through the paper, it
`blew up. Then each of the chunks that blew off it hit the ice and blew up in turn. The
`explosions spread across about 2 miles of ice." He also got involved in communications
`networks, figuring out how to use MIT's network connection to IBM to get inside New York
`City's phone system and place free phone calls.
`Roberts got three degrees in electrical engineering - a bachelor's in 1959, a master's in 1960,
`and a PhD in 1963. "I was an A student without any trouble," he shrugs. Luckily for the world,
`a more constructive pursuit than phonephreaking or explosives deployment managed to grab
`Roberts' attention before he left MIT: getting computers to talk to one another.
`After earning his PhD, Roberts stayed on at MIT's Lincoln Laboratory, which was originally
`funded by the government for air-defense projects. Inspired by Leonard Kleinrock and J. C. R.
`Licklider's 1961 paper on the concept of a "Galactic Network," Roberts began working on a
`packet-switching protocol. In 1965, he pulled off the first computer-to-computer
`communication using data packets. The next year he was strong-armed by the head of the
`Advanced Research Projects Agency and the director of Lincoln Lab into taking a job as chief
`scientist at Arpa's Information Processing Techniques Office. Roberts went to work on
`Arpanet. By March 1970, computer messages were being zapped across the US. In recognition
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`8/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 8
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`Arpanet. By March 1970, computer messages were being zapped across the US. In recognition
`of their work in founding the Internet and developing its core technology, Roberts, Cerf, Kahn,
`and Kleinrock recently received engineering's highest honor: the Charles Stark Draper Prize
`for engineering achievement.
`Roberts took his first crack at running a business in 1973, as CEO of Telenet, where he
`developed the X.25 data standard. GTE bought the company in 1979. Until 1998, Roberts
`worked to develop and refine the ATM protocol - which allowed data to move faster than
`TCP/IP, but in a less versatile manner - first as CEO of NetExpress, then as president of ATM
`Systems. The two major lessons he cites from those experiences: First, it's hard to get people
`to change to a new technology if an existing one works pretty well. Second, he's an engineer,
`not a businessman. "After ATM, I knew being a CEO wasn't my strength," Roberts says. "I was
`spending all my time building the company, and not what I do best - this facet of technology."
`In April 1998, ATM Systems was shuttered by its corporate parent, AMP. That was fine with
`Roberts. By then he was finally coming to terms with the fact that ATM was on its last legs.
`More important, he'd been thinking about a few things he should do with IP switching, and
`the transition gave him the opportunity to grab some of ATM's brightest people and head back
`into the computer lab.
`Within a month, Roberts had $3 million in funding from Vulcan Ventures. It was enough to get
`the team - a dozen engineers working under the name Packetcom LLC - to the starting line.
`By February 1999, they had burned through about $2 million and had a rough design and a
`business plan - which would get them $12 million more from Vulcan, Lucent Venture
`Partners, New Enterprise Associates, and Applied Technologies.
`In August 1999, Roberts pitched the CEO job to Grahame Rance. Rance took the job and
`instituted the stealth rules that have kept Caspian an enigma until now: No talking to the
`media, or even family members, about anything more detailed than what Caspian served up
`on its Web site; no whiteboards left unerased or paper trash unshredded; and no Caspian
`dumpsters left unguarded. In November, Roberts landed a team of Lucent programmers who
`had mastered the routing protocol known as BGP, a notoriously buggy platform originally
`developed by Cisco.
`March 2000 brought a new investor, WorldCom, which chipped in a portion of the $40
`million second round. Meanwhile, Juniper was shocking the industry with its winning forays
`against Cisco, and Avici and Pluris were busy hauling in war chests of $168 million and $162
`million, respectively.
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`9/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 9
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`__Roberts isn't the only one pushing dynamic switching. "The difference is," says a fan, "here's
`a guy who understands the Internet from its roots." __
`Each of these firms was on a hiring spree, but star power allowed Caspian executives to use
`some hardball tactics. The company would tell the people it offered jobs to that they had to
`give an answer before they left the building. Caspian says 97 percent of its offers were
`accepted.
`That hiring record is largely due to Roberts. "You know how people get starry-eyed when, say,
`Tom Cruise walks into the room?" says Tim Bajarin, president of research firm Creative
`Strategies. "For us techie types, it's the same thing with guys like Larry. Like everyone else,
`he's showing you impressive charts. The difference is, here's a guy who understands the
`Internet from its roots."
`Of course, for all of Roberts' brilliance, totemic status, and crafty designs, there's no guarantee
`Caspian will even be a factor in - much less walk off with - a market that Cisco now owns.
`Cisco's Redford implies that Caspian's innovations are just more in a long line of technological
`developments without a demonstrated demand in the marketplace. The switching fabrics
`designed with the aim of lashing many server racks together (like the ones made by Pluris,
`Avici, and Caspian), Redford insists, are overkill. "Most of them are science experiments," he
`says. "Carriers are just starting to deploy OC-192. We're not even close to needing that yet."
`Cisco didn't reach its position in the industry by being gracious toward its competitors,
`however small or obscure. And the company didn't get to the top by standing still. The latest
`additions to Cisco's core-router line boast a distributed architecture and MPLS forwarding.
`Caspian's finished ASIC chips aren't due back from IBM until late summer, and the company
`hopes to have the Apeiro in production early next year. Much will depend, however, on the
`outcome of testing that one major carrier is doing on an early version of the Apeiro, using
`field-programmable gate arrays to simulate the Caspian ASICs. Caspian's coming-out party,
`scheduled for April 10, was timed not to announce big-name customers or even to declare the
`Apeiro ready, but to announce the capabilities and technical specifications of the IP
`superswitch in time to be considered for carriers' 2002 buying cycles. At press time, the April
`launch remained in question, due to Rance's unexpected exit.
`The fact that no carriers would agree to be associated with Caspian in print suggests the
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`https://www.wired.com/2001/05/caspian/
`
`10/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 10
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`backbone operators view the Apeiro as highly experimental. Even the analysts are quick to
`point out that execution is everything. "The proof is in the pudding," says RHK's Al-Chalabi. He
`grants that the Apeiro architecture would indeed let carriers string an unprecedented amount
`of switching power together before they need to forklift out Caspian's racks, but cautions that
`nothing is certain until the company's machines are churning out terabits for paying
`customers. On top of that, up-front costs could also be a factor. Caspian has not released
`pricing for the Apeiro, but getting the throughput Roberts is talking about would require a
`significant outlay for the carriers. As for the two-year head start Roberts claims, Al-Chalabi
`says, "It's safe to assume that Cisco, Juniper, and the rest aren't going to sit there and say, 'Oh,
`game over.'"
`Then there's the awkward matter of Rance's departure a month before Caspian's launch date.
`David Liddle, a board member who was chief executive of Interval Research until January
`2000, is serving as Caspian's CEO until the company finds a replacement for Rance. As for
`why Rance left to run a relatively small components company, Roberts professes to be
`nonplussed. Speculation in the trade press points to the darkening prospects of taking a
`company to IPO, internal disagreement over Caspian's burn rate, and the undeniable strain of
`playing second fiddle to Larry Roberts.
`Not surprisingly, Roberts can spin any potential pitfall as an advantage. Caspian may be new
`to the market and will have to build its customer list from scratch, but Roberts says he's been
`working with the carriers ever since there was data traffic to carry; getting his foot in the door
`is not a problem. Besides, carriers are dying for new vendors to play against Cisco and Juniper
`- competition always drives down costs. Regarding Rance's exit: "It certainly came as a
`surprise," Roberts says, "but I'm confident we can find an even stronger person to take us the
`rest of the way."
`Asked a month before his departure whether Roberts is the kind of guy who will wait for the
`market to come to Caspian's supposedly revolutionary technology, Rance laughed: "You're
`talking about a queuing-theory expert, a guy who can be sitting in a long line of traffic waiting
`to turn left at a stoplight and can tell you down to the second when you'll actually make it
`through the light." (In fact, Roberts has a detailed table of hundreds of miles of roadways in
`Silicon Valley on which he's marked the travel time for each significant stretch.) Like its
`founder, Caspian leaves very little to chance. If you thought the name might be a reference to
`the prince in C. S. Lewis' Chronicles of Narnia, think again: It came from a naming
`consultancy, and, yes, it tested well.
`If his new machine is as superior as it looks on paper, the only thing that can get between
`

`

`8/16/2021
`
`The n -Dimensional Superswitch | WIRED
`
`1 Year of WIRED for
`$10.
`Plus a free tote!
`
`https://www.wired.com/2001/05/caspian/
`
`11/11
`
`Sable Networks, Inc., Exhibit 2001
`Page 2001 - 11
`IPR2021-00909, Cloudflare, Inc. et al. v. Sable Networks, Inc.
`
`If his new machine is as superior as it looks on paper, the only thing that can get between
`Roberts and his vision of a smart-core Internet, with Caspian as its titan, is an unforeseeable
`malfunction, perhaps in the software or in the chips IBM is working on. Or perhaps another
`high-level defection like Rance's. Who knows? Roberts radiates the dead-confident aura of
`someone who has the future locked up. But you can't account for everything. As Roberts raises
`an interview-ending wrist toward his eye, signaling that he's got places to go, an unexpected
`thing happens: He asks what time it is. His watch has stopped.
`TOPICS
`MAGAZINE-9.05
`S U B S C R I B E
`