ray -- more from the NGN conference and Wirbel(!).
-----
Start-up Fever Follows Bandwidth Boom
11/05/99, 8:36 p.m. ET
By Loring Wirbel and Craig Matsumoto
Electronic Buyers' News
techweb.com
ARLINGTON, Va. -- Predicting a long boom in the communications industry, keynoters and attendees at this past week's Next Generation Networks (NGN) conference here pointed to a sector flush with venture capital that's fueling a new round of companies angling to innovate.
At least 10 significant start-ups tipped plans that range from specialized communication-processor silicon to hybrid optical systems for cross-connect and add-drop mux systems in metropolitan carrier networks. Meanwhile, PC giant Intel opened a venture fund aimed squarely at backing optical networks.
The driver for the manic activity is the belief that consumers are on the verge of demanding broadband access, creating a "virtuous circle" where backbone networks must be upgraded to terabit and even petabit performance to accommodate a rapidly expanding user base. That has become a driver for system-on-chip designs functioning at the edge of the public network, since small access platforms must use transceivers operating at Sonet rates and packet parsers capable of classifying packets in gigabit ranges.
In a keynote speech at the conference, James Q. Crowe, CEO of fiber-centric carrier Level 3 Communications, argued that low-cost broadband services based on standardized silicon will drive communications growth, just as Moore's Law and commodity pricing have helped proliferate the PC. Carriers, he said, should look to the Intel model of microprocessor production and make broadband access for unified packet and circuit services as cheap as possible, if they hope to achieve exponential increases in the user base.
Capitalists Jump Onboard
Venture capitalists are riding that promise of growth. Redpoint Ventures' John McQuillan, who chaired the conference, said $2.67 billion of venture funding was invested in communications start-ups in the first six months of 1999, compared with $2.63 billion in all of 1998. Matrix Partners, to cite but one example, reported it had made more money from start-up investments in the past year than in the previous 29 years of its existence put together.
"As we consider global expansion of services, global deregulation of carriers, and the implications of voice on the Net, it's obvious that the broadband revolution is firmly entrenched and can sustain unprecedented growth for many years to come," McQuillan said.
McQuillan said he's confident, given access growth trends, that share prices as high as $200 for such companies as Juniper Networks are unlikely to drop below $170 in the foreseeable future. When one considers that many of the start-ups realized $100 million in revenue from real equipment orders in their first year of existence, those prices do not seem inflated, McQuillan said.
New players hoping to grab a piece of that growth were found at every turn at NGN. Many would reveal only small details of their product plans, but all were anxious to position themselves to take advantage of the money flow.
Growth Networks, in Mountain View, Calif., founded by Ron Bernal and Daniel Lenoski of Silicon Graphics, is a switching fabric IC specialist that's competing with the likes of Power X and Abrizio.
Extreme Packet Devices, a Kanata, Ontario spin-off of Newbridge Networks, is focusing on chips that provide traffic management for packets and cells. Mark Janoska, chief technology officer at Extreme, said the performance of the devices planned by his company will require hard-wired silicon, rather than a programmable device.
Procket Networks, in Santa Clara, Calif., led by Sharad Mehrotra and Bill Lynch from Sun's Ultrasparc team and by Tony Li from Juniper Networks, is keeping strategy close to the vest, though sources said the company is looking at packet-classifier chip architectures appropriate for backbone networks.
Force 10 Networks, formed by Som Sikdar, the founder of Shomiti Systems, is focusing on the nascent 10-gigabit Ethernet market, where Sikdar said he hopes to enable "carrier-class Ethernet switching."
Corvia Networks, in Sunnyvale, Calif., is aiming at a system-level terabit switching solution, developing a system for backbones that appears to use electronic rather than optical switching. Corvia's competitor and soundalike, Corvis Networks, said it uses an all-optical routing switch.
Astral Point Communications, in Chelmsford, Mass., is looking at metropolitan-area transport for new optical transmission systems. The company, founded by Raj Shanmugaraj of PictureTel and Bruce Miller of Bay Networks, has pulled in $24 million in venture funding.
Mayan Networks, with a strong management team from 3Com and Raynet led by Daniel Gatti, has pulled in a total of $90 million in funding to develop the Unifier, an edge aggregrator that can route, switch, and combine any mix of time-division multiplexing and ATM service down to DS-0 granularity and can interface directly to carrier-class optical equipment.
Oresis Communications, in Beaverton, Ore., has been quietly pulled together by George Shenoda of ADC Telecommunications to develop service-mediation platforms for merging several disparate network traffic types.
Malibu Networks, in Calabasas, Calif., will use orthogonal frequency division multiplexing technology to offer low-cost broadband access options for consumers using the 2.5- and 5-GHz bands in non-line-of-sight access systems. The strategy is similar to that pursued by a Cisco Systems-led coalition. The company has patents touching on wireless IP networks it said it hopes to have in prototype form next year.
In the midst of the start-up explosion, Intel, in Santa Clara, Calif., has launched an optical technology group within its new-business-ventures program. Intel hired Atikem Haile-Mariam of Tyco Submarine Cables to assess strategies for investing in optical transmission startups. Haile-Mariam called the opportunities boundless given the scores of newcomers moving to product prototyping in early 2000, but added, "You have to be very careful in examining who is real in this market right now. There are a lot of great ideas, but not all of them may move past prototype stage."
Intel's new focus on optical networking is in line with what many see as an endgame in the broadband era, with years of innovation ahead.
Several speakers, including McQuillan and Gerald Butters, group president for optical networks at Lucent, said end-to-end optical switching is still a few years away from true implementation. In the meantime, the expansion of first-generation optical WDM equipment with electronically switched optical add-drop muxes and optical cross-connects will allow carrier backbones at metro, national, and global levels to move to unprecedented levels of capacity.
Butters said optical capacity growth is only in the early stages of expansion. He said it's feasible that single-fiber WDM capacity will exceed 1,000 channels in the near future, with total throughput of 3.2 terabits per second. That "puts us in the holy grail space," he said.
Advances In Optics
Bell Labs already has shown a 1,022-channel WDM system with 10-GHz spacing using standard 80-nm optical amps, Butters said. Murray Hill, N.J.-based Lucent also is experimenting with free-air optics, in which four channels of 2.5-gigabit-per-second capacity can be carried in free space out to 4.4 kilometers; that makes Lucent's OpticAir "a formidable alternative to LMDS," a broadband wireless technology.
For both fiber and free-space optic systems, Lucent is working on a Sonet alternative to packet processing, earlier called "Digital Wrapper," with new semiconductor concepts called "WaveWrapper."
Butters said true photonic switching based on MEMS mirrors is not far behind. Lucent's MicroStar is a 4-by-4 array of two-axis micromirrors; eventually, the design could be reduced down to each mirror's being fabricated in a single crystal. That would allow multiple petabits to be switched in a space of less than 6 square inches, Butters said he predicted.
"We are reaching the point where silicon and electronic switching will become more expensive than photons," he said.
For Lucent, the decision to tackle optical technologies was economic. Despite all the progress, Butters said, "We can't see sufficient technological breakthroughs to satisfy the growing need for bandwidth." In fact, as many as 3,300 companies have some kind of optical products in development, Butters said, and those companies are not concentrated in Silicon Valley but spread globally.
He argued that all that bandwidth will find usage because of price elasticity. Only certain products are price-elastic and many people, including Butters, think bandwidth is one of them.
Beyond optical, the role of software also took center stage in several NGN sessions on Internet scaling. New ASICs alone aren't enough, said John W. Stewart, marketing engineer for Juniper Networks; "to actually take [an ASIC] and do packet forwarding, you'd have to write a lot of software." Expansion of Internet bandwidth is "pushing the envelope of physics and computer architecture," he said.
"Building software that scales is a mystery, in terms of scaling to the tens, twenties, and hundreds of terabits [per second] people are talking about," said Derek Oppen, vice president of carrier router product management at Nortel Networks.
Juniper's Stewart added, "Even if you can put enough gates on a chip to perform certain functions, you have to get enough information in and out of the chip to operate on. That's going to be a limit before the number of gates is a limit."
Reliability is another issue, as data networks improve their up-times to match the quality of the telephone system.
"We have to look at every single packet at line rate," said Jeff Wabik, chief architect of IP switching products at Lucent. "If you miss just one, it could be the packet that's initiating a 911 call."
But at the 10-Gbps rates likely to be used in the network core next year, that means examining one packet every 4 millionths of a second, Wabik said. ASICs might be able to handle that speed, but software can't. |
