SCIENCE
10 Gigabit ready to set up enterprise shop
By Phil Hochmuth, Network World -- Most corporate users are not yet clamoring for 10G Ethernet, but the technology is evolving from the concept phase to production customer trials and deployments in some cases. Falling prices and the proliferation of inexpensive Gigabit products are pushing demand for 10 Gigabit among some users. Others are holding out for the copper version of 10 Gigabit, in the hopes that this will drive down the per-port cost of the technology even further. Observers say this could come into play as 10G moves from switch-to-switch applications into high-end server and network-attached storage (NAS ) connectivity. While large government laboratories and research centers have been on the forefront of 10G adoption, a few recent installations of 10G have taken place in some more real-world enterprise environments, such as hospitals and in education. 
The North Bronx Healthcare Network (NBHN) recently completed an installation of 10 Gigabit switches from Extreme Networks to leapfrog its Fast Ethernet-based infrastructure one step ahead of the upgrade cycle. NBHN uses single-mode fiber runs to connect three hospitals with 10G bit/sec links. Driving the need for this kind of bandwidth was the growing use of applications such as medical-imaging technology, which can push digital magnetic resonance image (MRI) and X-ray files as large as 2G bytes across a LAN. While the 10G network was somewhat overkill, NBHN's CIO Dan Morreale says he anticipates that bandwidth needs will grow to utilize these pipes. Future plans for large-scale server consolidation also are driving some users to lay 10G foundations today. At Manchester Community College (MCC) in Connecticut, 10G is changing some of the most common blueprints in building LANs. Many organizations deploy 10/100M bit/sec edge switches, usually stackable boxes, which feed into a Gigabit-speed distribution layer and then the LAN core. MCC's approach is to eliminate the distribution layer and link desktop switches and servers directly to the core. To do this, 10G was required in the backbone, says Jason Blosser, IT director at the college. "This architecture will help us cut administration costs" and equipment expenses, because the aggregation switch layer - about a dozen boxes - will be eliminated, he says. 10G reality check The recent push of 10G products to corporate customers was spurred by a declining interest among carriers, which were the original 10G target market. "When [10G Ethernet] was being developed, there was an expectation that there'd be an uptake on the service provider side," says Bruce Tolley, senior manager of emerging technologies at Cisco , who was active in crafting the 10G standard. The promise of long-haul 10G Ethernet over copper as a SONET replacement was the main focus of the IEEE at that time. "But with the dot-com bust and telecom restructuring, that didn't happen," he says. The result of that action is 10G Ethernet is now an enterprise-focused technology. Even after this switch to an enterprise focus, however, several analyst firms say that demand for the products is not high. A recent In-Stat/MDR survey of 282 buyers of enterprise network equipment shows that 56% of respondents said they had "no plans" to deploy 10G Ethernet in their networks. This corporate position on 10G is reflected in product shipments. According to the Dell'Oro Group, only 1,000 10G switch ports were shipped last year, and only about 4,000 are estimated to be sold to customers this year. Comparatively, when Gigabit Ethernet was introduced to the masses in 1997, 11,000 ports were shipped. A year later, shipments reached 220,000. Observers add that the pent-up demand for 1000M bit/sec Ethernet was a pressing issue to businesses six years ago, creating a competitive market with many start-ups, led by firms such as Alteon, Extreme and Foundry Networks . "We're not seeing much demand for 10G among our clients," says Lawrence Orans, principal analyst with Gartner. But that's not to say 10G won't have its day. While not on the "to-do" list today for companies, In-Stat/MDR's survey showed that about 40% of users will have some 10G deployed within two years. Likewise, Gartner forecasts that shipments of 10G Ethernet adapters for servers and storage devices will jump from about 1,000 beta products shipped this year to server makers for trial, to more than 500,000 units shipping in production systems by 2007. The copper catalyst
While many analysts predict 10G shipments to rev up in the next several years, what could really make 10G a force in the market is the evolution of a copper-based version of 10G. Serious consideration of copper-based 10G began last November, when the IEEE formed two study groups: one to focus on studying how to run 10G Ethernet over Category 5 or 6 cabling; and another to figure out how to run 10G bit/sec Ethernet over four pairs of twin-axial copper cabling. Twin-axial cabling might be familiar to users of older IBM minicomputers or more recently for InfiniBand device interconnect cabling. The cables are used for runs of about 15 to 20 feet. The major driver for copper 10G development is cost. Fiber Ethernet ports cost more than copper at every level of connection speed. When 10GBase-LX debuted in 2001, ports cost as high as $80,000. This equates to 10 times the performance of Gigabit Ethernet, at about eight times the cost. Pricing has come down by about half since then,but analysts and IEEE engineers estimate that copper 10G will come in at about two or three times the cost of Gigabit Ethernet while offering 10 times the performance. For networks in which 10G is established, 10G over copper represents the next step for the technology. "10G on copper looks good because of the lower cost per interface and higher densities," says Richard Nelson, director of information processing at the University of Southern California's Information Sciences Institute in Marina del Rey, which also conducts supercomputing research. He says NAS disk arrays and high-powered servers are two areas where copper 10G could fit in now. The institute uses Foundry switches for its 10G Ethernet links, which extend several miles and connect multiple computer rooms across several campuses in the metropolitan area. But not all 10G links need to be long-reach, he says.
The North Bronx Healthcare Network (NBHN) recently completed an installation of 10 Gigabit switches from Extreme Networks to leapfrog its Fast Ethernet-based infrastructure one step ahead of the upgrade cycle. NBHN uses single-mode fiber runs to connect three hospitals with 10G bit/sec links. Driving the need for this kind of bandwidth was the growing use of applications such as medical-imaging technology, which can push digital magnetic resonance image (MRI) and X-ray files as large as 2G bytes across a LAN. While the 10G network was somewhat overkill, NBHN's CIO Dan Morreale says he anticipates that bandwidth needs will grow to utilize these pipes. Future plans for large-scale server consolidation also are driving some users to lay 10G foundations today. At Manchester Community College (MCC) in Connecticut, 10G is changing some of the most common blueprints in building LANs. Many organizations deploy 10/100M bit/sec edge switches, usually stackable boxes, which feed into a Gigabit-speed distribution layer and then the LAN core. MCC's approach is to eliminate the distribution layer and link desktop switches and servers directly to the core. To do this, 10G was required in the backbone, says Jason Blosser, IT director at the college. "This architecture will help us cut administration costs" and equipment expenses, because the aggregation switch layer - about a dozen boxes - will be eliminated, he says. 10G reality check The recent push of 10G products to corporate customers was spurred by a declining interest among carriers, which were the original 10G target market. "When [10G Ethernet] was being developed, there was an expectation that there'd be an uptake on the service provider side," says Bruce Tolley, senior manager of emerging technologies at Cisco , who was active in crafting the 10G standard. The promise of long-haul 10G Ethernet over copper as a SONET replacement was the main focus of the IEEE at that time. "But with the dot-com bust and telecom restructuring, that didn't happen," he says. The result of that action is 10G Ethernet is now an enterprise-focused technology. Even after this switch to an enterprise focus, however, several analyst firms say that demand for the products is not high. A recent In-Stat/MDR survey of 282 buyers of enterprise network equipment shows that 56% of respondents said they had "no plans" to deploy 10G Ethernet in their networks. This corporate position on 10G is reflected in product shipments. According to the Dell'Oro Group, only 1,000 10G switch ports were shipped last year, and only about 4,000 are estimated to be sold to customers this year. Comparatively, when Gigabit Ethernet was introduced to the masses in 1997, 11,000 ports were shipped. A year later, shipments reached 220,000. Observers add that the pent-up demand for 1000M bit/sec Ethernet was a pressing issue to businesses six years ago, creating a competitive market with many start-ups, led by firms such as Alteon, Extreme and Foundry Networks . "We're not seeing much demand for 10G among our clients," says Lawrence Orans, principal analyst with Gartner. But that's not to say 10G won't have its day. While not on the "to-do" list today for companies, In-Stat/MDR's survey showed that about 40% of users will have some 10G deployed within two years. Likewise, Gartner forecasts that shipments of 10G Ethernet adapters for servers and storage devices will jump from about 1,000 beta products shipped this year to server makers for trial, to more than 500,000 units shipping in production systems by 2007. The copper catalyst
While many analysts predict 10G shipments to rev up in the next several years, what could really make 10G a force in the market is the evolution of a copper-based version of 10G. Serious consideration of copper-based 10G began last November, when the IEEE formed two study groups: one to focus on studying how to run 10G Ethernet over Category 5 or 6 cabling; and another to figure out how to run 10G bit/sec Ethernet over four pairs of twin-axial copper cabling. Twin-axial cabling might be familiar to users of older IBM minicomputers or more recently for InfiniBand device interconnect cabling. The cables are used for runs of about 15 to 20 feet. The major driver for copper 10G development is cost. Fiber Ethernet ports cost more than copper at every level of connection speed. When 10GBase-LX debuted in 2001, ports cost as high as $80,000. This equates to 10 times the performance of Gigabit Ethernet, at about eight times the cost. Pricing has come down by about half since then,but analysts and IEEE engineers estimate that copper 10G will come in at about two or three times the cost of Gigabit Ethernet while offering 10 times the performance. For networks in which 10G is established, 10G over copper represents the next step for the technology. "10G on copper looks good because of the lower cost per interface and higher densities," says Richard Nelson, director of information processing at the University of Southern California's Information Sciences Institute in Marina del Rey, which also conducts supercomputing research. He says NAS disk arrays and high-powered servers are two areas where copper 10G could fit in now. The institute uses Foundry switches for its 10G Ethernet links, which extend several miles and connect multiple computer rooms across several campuses in the metropolitan area. But not all 10G links need to be long-reach, he says.