Will high-speed copper blunt fiber`s Gigabit Ethernet edge?
An upcoming Category 5 UTP cabling standard will enable copper to bring high-speed local area network technology to the premises, raising questions about fiber?s role.
Most prognosticators have granted the Gigabit Ethernet backbone to fiber, as equipment based on the new standard is finally ready for implementation. But given users` existing bias toward copper and the imminent arrival of a copper-based Gigabit Ethernet (Gigabit Ethernet) standard, will users overlook copper`s limitations and shun fiber?
At 10 times the speed of 100-Mbit/sec Fast Ethernet, Gigabit Ethernet (1000 Mbit/sec) promises to deliver more data to the local area network (LAN) in record time. Perhaps even more importantly, as an extension of the 10Base Ethernet and 100Base Fast Ethernet technologies, 1000Base Gigabit Ethernet allows network operators to incrementally upgrade existing Ethernet networks without disruption. Like its predecessors, Gigabit Ethernet uses the 802.3 Ethernet frame format and the CSMA/CD (carrier sense multiple access with collision detection) access protocol when communicating in half-duplex mode. These characteristics allow network operators to install Gigabit Ethernet as a backbone interconnect for 10/100Base-T switches and as a link to high-performance servers.
No one questions the demand for greater bandwidth in data networks. There is considerable debate, however, about how much bandwidth is really needed and the most cost-effective way to reach higher performance while at the same time futureproofing the network.
Last June, fiber got a boost when the Institute of Electrical and Electronics Engineers (IEEE) ratified the Gigabit Ethernet 802.3 standard, which supports 1-Gbit/sec data transmission over fiber and short-haul copper. The 802.3z Gigabit Ethernet task force defined specifications for media access control, an optional Gigabit media-independent interface, and three physical-layer (PHY) standards (1000Base-X). The Gigabit Ethernet PHY standards were based in part on the X3.230 Fibre Channel PHY technology developed by ANSI, allowing many manufacturers to re-use existing PHY components for Gigabit Ethernet. The 1000Base-CX specification defines a copper-based link of up to 25 m for machine room or switching closet connections. The 1000Base-LX long wavelength PHY standardizes Gigabit Ethernet data transmission over 62.5-micron multimode fiber to 220 m, 50-micron multimode fiber to 550 m, and singlemode fiber to 5 km. The 1000Base-SX short-wavelength standard specifies Gigabit Ethernet over multimode fiber (see Table 1).
Fiber-based Gigabit Ethernet faced some technical hurdles, however. Light-emitting diodes (LEDs), which are sufficient for Fast Ethernet, are too bulky and slow for high-speed Gigabit Ethernet, which requires laser technology. But when a laser is launched at a high bit rate into multimode fibers with a less than ideal index-of-refraction profile, the light may start to disperse faster than it should and jitter can result. This phenomenon is called differential mode delay (DMD). As a result, the IEEE 802.3 task force used conservative attenuation (signal loss-over-distance) estimates for multimode fiber and short-wavelength transceivers. Singlemode fiber and copper solutions are not affected by DMD. (For more information on DMD, see "Conditioners overcome DMD" on page 43.)
"For a long wavelength, one of those trouble areas is in the middle of the fiber," says Bob Mayer, vice president of sales and marketing, Cielo Communications (Boulder, CO). "But the standard says you have to launch your light in the middle so that you can go directly into the very small core of singlemode fiber. So when you launch it now, into multimode fiber, you are focused on that danger area. The solution for long wavelength is to use an external offset patch cord, which takes that little spot of light and moves it over to an outer ring of the core of the fiber."
Another way to control the light source is by using the relatively new vertical-cavity surface-emitting lasers (VCSELs). "It has a narrower spectral width, which is what a laser has, versus an LED," says Mayer. "A vertical-cavity laser is similar in manufacturing to an LED but it performs like a laser with improved reliability, so it is clearly the technology for gigabit and above fiber-optic links."
At this point VCSELs are only available at the lower wavelengths--850 and 980 nm. The Gigabit Ethernet short-wavelength standard supports 850 nm. "VCSELs don`t yet function at the 1300-nm wavelength in any mass produceable way," says Mayer. "People are looking at having products in the next two to three years, but there is still a lot of work that needs to be done before that is real."
The use of lasers also has implications beyond DMD. For example, re-testing existing premises networks and horizontal infrastructures to determine whether the cable plant complies with Gigabit Ethernet parameters is a major issue, according to Dennis Mazaris, president of PerfectSite, a Sterling, VA-based consulting firm. "Does the end- user want to go back to retest his existing cable plant to see if it is going to comply with the parameters of Gigabit Ethernet?" asks Mazaris. "If you want to do it properly, you would have to use a laser source because odds are that everything that has been tested inside your premises wiring system has been tested using an LED source."
If an LED-based light source is used in field testing--important for networks supporting both 100-Mbit Fast Ethernet and 1000-Mbit Gigabit Ethernet communications--the IEEE`s conservative attenuation limit of 2.38 dB for 1000Base-SX can be easily violated. The TIA FO2.2 working group is currently trying to determine a way to standardize a single test method.
Gigabit Ethernet over copper
With these technological questions as a backdrop, the advent of a copper-based Gigabit Ethernet standard that would extend transmission distances beyond those covered in 1000Base-CX appears potentially troubling. The IEEE created a task force in March 1997 to work on an 802.3ab standard that would support Gigabit Ethernet transmission over Category 5 unshielded twisted-pair (UTP) copper. The objective of the standard is to allow people to take advantage of existing horizontal Category 5 cabling, which is the predominant infrastructure worldwide. This second copper standard is for distances of up to 100 m (based on cabling regulations) and networks up to 200 m. The standard is on schedule to be ratified in March.
However, copper-based Gigabit Ethernet must surmount technological hurdles of its own. Unlike Gigabit Ethernet over fiber, which allows manufacturers to use existing Fibre Channel technology, copper requires new technology and coding schemes. "The copper standard is taking quite a bit longer," admits Colin Mick, editor of the 1000Base-T Gigabit Ethernet copper standard and principal of the Mick Group (Palo Alto, CA). According to Mick, multiple silicon vendors are developing transceivers and several device companies are working on designs for adapter cards and switches. He expects many companies to have prototype products by the end of the first quarter of this year.
"Part of the issue is that we`ve learned a lot about copper cabling over the last four to five years," says Mick. "Initially, the Category 5 cabling standard wasn`t refined until 1995. So there still exists some cabling that was installed prior to that time. The concern the industry has is that some of that cabling may not support 1000Base-T operation. In fact, we don`t know." The 802.3ab Gigabit Ethernet standard is designed for operation over cabling systems that meet ANSI/TIA/ EIA-568A 1995 installation requirements.
"In developing 1000Base-T, we used an extremely conservative model to characterize worst-case cable plants--and I mean really conservative," continues Mick. "And we also use in our design models extremely conservative assumptions about performance parameters. Our assessment is that the product when it comes out should be fine and should work over almost all installed Category 5. As a good rule of thumb, any Category 5 that would support 100Base-X [Fast Ethernet] should have no problem with 1000Base-T."
In contrast to 100Base-T Fast Ethernet, which uses a 3-level binary symbol rate of 125 Mbits across only two pairs of the 4-pair cable (one pair is used to send the signal, the other pair is used to receive the signal), Gigabit Ethernet requires a full-duplex transmission protocol. It uses a five-level coding scheme at the same symbol rate of 125 Mbits. Each pair of the 4-pair cable sends and receives data simultaneously.
The Category 5 1995 installation standard did not provide measures for two important characteristics of the cabling: return loss and far-end crosstalk. "The reason there is a little bit of industry confusion is because all we can do is say, `Look, you should test Category 5 links for return loss and far-end crosstalk to qualify them for 1000Base-T,`" says Mick. "We think this won`t be an issue, but we won`t know it until we`ve got hard products out in the field operating."
How quickly will such offerings reach the field? Products should start rolling out by mid-year, say sources. Many will support 10/100/1000-Mbit operation using a switching algorithm called auto-negotiation.
Meanwhile, the cabling industry has developed installation practices for Enhanced Category 5 (5E), which include tests for return loss and far-end crosstalk (see Table 2). "There is enough concern out there about whether Category 5 is robust enough that people are looking at Enhanced Category 5 systems and Category 6," says Dan Kennefick, copper business manager, Berk-Tek (Holland, PA). "The IEEE group will come out and say eventually--once they get all the electronics worked out--yes, Category 5 can support Gigabit Ethernet.
"What will probably happen in practice, if we look back at history with the development of 100Base-T Fast Ethernet, is [users] won`t use the minimum essential standard. Most users and consultants won`t want to risk using the minimum essential. They will design-in headroom to ensure that they reach a level of comfort that their system is going to work," he concludes.
Looking to the future, it appears that copper and fiber will both find niches in Gigabit Ethernet applications. Overall, copper will continue to be the predominant medium for the next five years, according to cable market researcher World Information Technologies (North Port, NY). The fiber-optic market is, however, growing much faster than the cable industry as a whole. Fiber revenue in the United States is expected to increase more than 20% annually for the next five years. During the same period, UTP copper revenue is estimated to grow less than 14%.
"A lot of companies that are now in the Category 5 and UTP area are pushing fiber from a strategic point of view," says Amadee Bender, president of World Information Technologies. "The main reason is that price competition is getting very fierce in the UTP area and profit margins are narrowing. Fiber is basically a less competitive environment and a lot of companies are probably going to migrate to fiber as a result. The more optimistic view for the fiber cable is really a result of its increased bandwidth compared to copper. And that the standards for Category 6 are not expected until 2000. The standards for Category 7 are not anticipated until well after 2000--maybe 2003--so it has become a more attractive option."
Many leading optoelectronics manufacturers view the 802.3ab Gigabit Ethernet standard as a market driver for the fiber industry. "We are actually not a manufacturer of copper-based Gigabit Ethernet products--and we view this favorably," says Roselyne Genin, vice president of optical networking, Ericsson (Richardson, TX). "All those deployments that take place on the copper side closer to the subscriber are helping bring more data to the network and are helping bring data from LANs to WANs, which is then carried over fiber."
"As Gigabit Ethernet actually goes closer to the desktop, and the easiest way to get there is over twisted-pair because it is already in the walls, the more of a need you`re going to have for Gigabit Ethernet in the workgroup and backbone," states Mayer.
Fiber-based Gigabit Ethernet thus becomes an effective medium for backbone applications (see Tables 3 and 4). The advantages of Gigabit Ethernet over competing technologies should ensure it establishes a preeminent place in this part of the network. For example, a Gigabit Ethernet switch is more cost-effective than an Asynchronous Transfer Mode (ATM) switch, according to one vendor. "At the present time an ATM switch running at 622 Mbits/sec can cost anywhere from $5000 to $10,000," observes John Phillips, senior solutions marketing engineer at Ericsson. "But a Gigabit Ethernet switch can cost $1000 to $3500.
"I think as the standard matures, you will see a lot more network providers being able to provide Gigabit Ethernet from the desktop all the way to the backbone, which will increase the deployment of Gigabit Ethernet. ATM switches right now are running at 155 or 622 Mbits/sec, so Gigabit Ethernet basically doubles the amount of bandwidth that you have at your access points," he says.
Yet, while the fiber-based Gigabit Ethernet standard promotes fiber as the backbone candidate for Ethernet LANs, fiber-to-the-desk is still a long way off. "There isn`t an application to the desktop yet that can`t be supported by UTP copper, although there probably will be down the road," says Berk-Tek`s Kennefick. "Certainly with Gigabit Ethernet, we`re taxing the capability of UTP copper, but the [IEEE Working Group] just about has all that worked out."
Yoseph Linde, chairman of media converter manufacturer LANart Corp. (Needham, MA), shares a similar viewpoint: "There is really no compelling reason to go to fiber, because twisted-pair is giving people the performance that they need. Eventually, there is no question that fiber is going to be at the desk because a gigabit seems to be really pushing it for twisted-pair--there`s a big discussion whether it is going to run on Category 5 or not."
Prohibitive cost is still a major impediment for fiber-to-the-desk as well. While the cost of the actual cable is roughly the same when considering fiber and Category 5 UTP, the optoelectronics for fiber are significantly more expensive.
"It all comes back to the higher cost of the electronics with fiber, that`s as simple as it gets," says Mazaris.
"You can pay for it now or you can pay for it later," counters Steve Montgomery, president of market research firm ElectroniCast Corp. (San Mateo, CA). Montgomery thinks new builds likely to remain in the same locations for more than five years should opt for fiber-optic cable. One solution is to put in fiber cable and use copper switches or media converters, which are both considerably less expensive than optoelectronics.
"The bottom line is if somebody is pulling new cabling, they`re going to have to make the decision on whether they are going to go copper or fiber," says Mick. "That isn`t a decision we can really effect. All we can say is that if they want to use existing horizontal copper cabling and it is installed according to 1995 copper cabling installation practices, we think Category 5 cabling should work fine."
Like industry manufacturers, Mick does not expect the copper-based Gigabit Ethernet standard to hinder fiber-based Gigabit Ethernet. "The first place we`ll probably see [copper-based Gigabit Ethernet] is as a cheaper way to connect servers in machine rooms," says Mick. "The second place will be in high-performance workgroups."
If the ratification of the 802.3ab copper-based Gigabit Ethernet standard occurs on schedule in March, the market is one standard closer to mass adoption of this high-speed networking technology. Many manufacturers will offer end-to-end fiber and copper Gigabit Ethernet solutions--fiber on the backbone and copper cable as the horizontal infrastructure. Some users, especially new builds, may consider fiber-to-the-desk, but the higher expense and current bandwidth demands still make it hard to justify. u
Editor`s Note: The source documents used to compile this report can be accessed via the World Wide Web at www.computer-crafts.com.
Conditioners overcome DMD
The question of whether fiber-based Gigabit Ethernet can withstand the potential onslaught of the upcoming copper-based standard would have been rendered moot if the Modal Bandwidth Investigation group of the Gigabit Ethernet Task Force hadn`t overcome the problems associated with differential mode delay (DMD). DMD became a point of concern when investigators discovered that the use of long-wavelength singlemode lasers severely reduced the usable bandwidth of 62.5-micron multimode fiber. This phenomenon occurs because lasers optimized for singlemode fiber preferentially excite modes near the core of multimode fiber. The propagation delay between paraxial and skew rays results in DMD-induced jitter.
Several factors complicate the suppression of DMD. First, DMD itself is a complex amalgamation of the modal weighting of the source, launch conditions, the fiber`s mode delay, mode group separation within the fiber, and mode-specific attenuation. Second, typical bandwidth measurements based on the over-filled launch method (OFL) do not correlate with DMD. Thus, a new bandwidth measurement, called radial over-fill launch (ROFL), had to be created. The challenge for the Modal Bandwidth Investigation group was to get the ROFL figures closer to OFL numbers.
The first step involved accepting the obvious: The group members couldn`t change the fiber, so they had to concentrate on the laser launch. An offset launch that would move the center of the light projection away from the center of the core appeared a logical solution to the problem; such a launch would be conditioned to be similar to that achieved under OFL conditions. An offset launch cable offered a good mechanism to achieve this condition, and the group investigated two cable designs. The first would accomplish the singlemode-to-multimode offset via a ferrule placed between the launch cable and the plant cable or with a secondary jumper. But a tolerance analysis revealed that variability in the connector and coupler components might result in poor alignment in some applications.
In the second design, two offset ferrules were used between a singlemode and multimode fiber and actively tuned. The actively tuned ferrules would introduce a controlled amount of lateral offset between the fiber cores. Simulation and experimentation proved that this technique would overcome the effects of DMD in at least 99% of installed multimode fibers. When using 62.5-micron fiber, a lateral offset in the range of 17 to 23 microns did the trick. Members of the group also demonstrated that a 10- to 16-micron shift worked well for 50-micron fiber.
Thus, specialty launch cables using mode-conditioning units are the recommended method for overcoming DMD in Gigabit Ethernet applications where long-wavelength singlemode lasers must be used with multimode fiber. Such conditioners are available from Computer Crafts Inc. (Hawthorne, NJ); other companies are expected to follow suit in the near future. Vendors have already begun to incorporate these units into their products. For example, IBM Corp. will use conditioners for its HiPerLinks adapters, which will enable singlemode-to-multimode conversion within the context of the company`s Parallel Sysplex architecture. In the future, IBM plans to use these conditioners for both Fibre Channel and Gigabit Ethernet offerings, as well as other future applications.
-- Stephen Hardy, editor in chief