A new standard for Fast Ethernet applications using short-wavelength optics has neared completion. Now the question is, who will adopt it for their product lines?
The cost/benefit tradeoff for fiber optics versus copper has always been clear: Fiber costs more for customers to install, but they get better performance in terms of bandwidth capacity and transmission distance. Unfortunately for companies attempting to push fiber into the Ethernet space, this balance doesn't hold in the normal way. Both Ethernet and its Fast Ethernet sibling cap their bandwidth requirements at levels well within the reach of copper. Thus, fiber's only advantage for these applications lies in transmission distance--and the comparative lack of users who have been willing to pay a premium for long-distance 10-Mbit/sec transmission indicates how appealing this advantage has proven.
With the advent of Gigabit Ethernet, most fiber proponents believe that bandwidth has finally reached a point where the usual cost/benefit analyses come into play once again. Thus, as the cost of fiber-based applications continues to decrease because of smaller transceiver sizes and other innovations, the door may finally be open for significant entry into the enterprise space, particularly for backbone applications. Many feel this light at the end of the tunnel has illuminated other opportunities at lower speeds and shorter runs, as well--provided costs can be reduced even further.
The new short-wavelength Fast Ethernet specification--which should be ratified as TIA-785 by the end of this quarter--could meet this goal by replacing the 1300-nm optics currently required in the IEEE's Fast Ethernet standard with less-expensive 850-nm devices. At the same time, it will enable network managers to use their currently installed copper-configured 10/100 network interface cards (NICs) if they cleverly configure their setup.
The 100Base-SX standard, which is expected to be cleared for final balloting at the end of this month, addresses a pair of hurdles that the manufacturing community believes have stood in the path of fiber's migration into the horizontal premises space: the comparatively high cost of fiber components and the lack of a clear fiber migration path from 10-Mbit/sec Ethernet through 100-Mbit/sec Fast Ethernet to Gigabit Ethernet. Two provisions address these concerns: One allows the use of the less-expensive 850-nm optics (at the price of lowering transmission distance from 2 km to 300 m); the other establishes the parameters necessary for auto-negotiation between 10-Mbit/sec and 100-Mbit/sec devices.
The change in the physical media-dependent interface to allow the use of the 850-nm devices represents the most significant change from the 100Base-FX specifications; nothing above the physical layer varies from the 1998 edition of IEEE 802.3. The standard will support 62.5/125- and 50/125-micron fiber; equipment conforming to the standard will carry a type rating to alert the user to which fiber the equipment supports. The standard also incorporates a variety of connectors, specifically the SC, ST, and "any alternate connector designs meeting the requirements of the corresponding intermateability standard and the optical-connector requirements of proposed ANSI/TIA/EIA 568-B.3." This last group roughly conforms to the new generation of small-form-factor connectors.
The switch to 850-nm optical devices means that users stepping up from Ethernet to Fast Ethernet on fiber won't have to change all their electronics to devices with 1300-nm optics, as is required to conform to 100Base-FX. The wavelength hiccup arose because when the IEEE established Fast Ethernet, the transmission-distance advantage of fiber was considered important enough to retain. Boosting transmission speed by 10 while retaining the 2-km distance attainable with the fiber-based 10Base-FX Ethernet standard was beyond the capability of 850-nm optics. Thus, the new 100Base-FX Fast Ethernet standard included a requirement for 1300-nm lasers. Unfortunately for vendors, while the distance stayed the same, the cost went up. Naturally, this didn't boost fiber's popularity with the end user.
Thus the groups behind 100Base-SX--principally the TIA Fiber Optic LAN Section (FOLS) and the Short Wavelength Fast Ethernet Alliance (see Table)--view the 850-nm solution as an important step toward taking advantage of what they perceive as an increased willingness on the part of forward-looking end users to consider fiber for local-area-network (LAN) applications. Light-emitting diodes operating at 850-nm inherently cost less than1300-nm lasers. Additional cost savings accrue from the fact that users operating at 10 Mbits/sec can continue to use many of their existing NICs and other LAN electronic devices when they migrate to 100 Mbits/sec.
The auto-negotiation feature aids this smooth migration. The feature allows Fast Ethernet hubs to support workgroups operating at regular 10-Mbit/sec Ethernet speeds, just as in the copper-based 100Base-TX standard. "Auto-negotiation" occurs between two devices when they first appear on the network; each sends a signal that defines its top transmission speed and whether it operates in full- or half-duplex mode. The two devices compare the operating parameters they have received with their own parameters. The highest-performing combination is then used whenever the devices need to communicate.
The architects of the 100Base-SX standard adapted the auto-negotiation technique spelled out by the 1998 edition of IEEE 802.3 for 100Base-TX for use on fiber. This means that the information exchanged between devices and the state machine employed are compatible with existing 10/100 devices, and therefore 100Base-SX networks will support legacy devices.
The adaptation of 100Base-TX auto-negotiation into a fiber environment requires a change in the way the parameter messages are encoded. The 100Base-SX standard uses a 1-MHz idle signal to ensure compatibility with legacy devices (see Fig. 1). This signal is interrupted every 16 microsec with a burst of data. The bursts are encoded by skipping clock pulse transitions in the idle signal. The skips, called Fiber Link Negotiation Pulses (FLNPs), are repeated every 62.5 microsec and alternate between clock and data (see Fig. 2). There are a total of 16 data pulses and 17 clock pulses; the first and last pulses are always of the clock variety. The 16 data pulses make up what IEEE 802.3 Edition 1998 calls the "Link Code Word," which is the carrier of the speed and operational mode information.
In instances where a legacy device is not equipped for auto-negotiation, "parallel detection" offers a fallback solution. In this scheme, a 100Base-SX device monitors the link in search of signals from legacy devices such as those running 10Base-FL. When such signals are detected, auto-negotiation ceases and the link is established at the legacy speed in half-duplex mode.
On the face of things, the advent of 100Base-SX might elicit a yawn from the average vendor of LAN hubs, switches, or NICs. Even with the switch to 850-nm optics, a fiber-based Fast Ethernet solution generally remains more expensive than running Ethernet or Fast Ethernet over Category 5 copper, particularly if the devices attached to the network already have 10/100 interfaces. And with the comparatively few applications currently running 10-Mbit Ethernet over fiber, why would auto-negotiation even be an issue in more than a handful of instances?
The companies in the FOLS believe that the advent of Gigabit Ethernet will tip the scales toward fiber in an increasing number of applications. While 1000Base-T will treat gigabit speeds over Category 5 wiring, fiber equipment vendors are betting that a significant number of current installations won't measure up to the performance requirements the new standard will demand. Thus, users who see the need to support gigabit speeds in their current planning cycles--or in the next one--will be forced to rewire their networks. If you have to lay new cable anyway (a scenario that obviously applies to greenfield applications as well), why not choose a medium that will ensure you only have to go through this nightmare once, particularly with 10 Gigabit Ethernet on the horizon? With the cost reduction offered by 850-nm devices-which could reach 40% of comparable 100Base-FX equipment-the price differential between fiber and Category 6 or 7 networks may be close enough to make a "pay a little more now or pay a lot more later" sales pitch effective.
Meanwhile, the auto-negotiation feature allows network managers to use existing 10/100 copper-based NICs to implement a fiber-based architecture through the addition of media converters. "We're really focusing on people who would like to put in fiber-to-the-desk but can't afford to go with a pure fiber-to-the-desk system that would include a fiber-optic network interface card," says Allen Dixon of Siecor (Hickory, NC).
Dixon chairs the Telecommunications Industry Association's FO2.2 committee, which is shaping the 100Base-SX standard. "This allows them the benefit of reusing the 10/100 card that probably shipped with the computer that they have," he says.
While fiber proponents have hung their hopes on the hook of increasing bandwidth demands within premises environments, the emphasis on the auto-negotiation capability reveals a realization that lower speeds remain popular. "There are still a tremendous amount of people out there running 10 Mbits, even though the use of 10/100 NIC cards has outstripped the use of 10-Mbit NIC cards," says Dixon. "So [with 100Base-SX] even if you've got an existing user who just decides that they're moving into a new area or they're rewiring and they're going to go to fiber, they can still utilize the 10 Mbits/sec--and probably will."
"Fast Ethernet to the desktop is still not a common orientation," agrees Cheri Podzimek, vice president of marketing at Transition Networks Inc. (Minneapolis, MN). Transition has announced both a media converter and 10/100 NIC that support 100Base-SX auto-negotiation. "We really believe, and we're seeing the market for it today, that people are going to put in 10-Mbit fiber rather than putting in 100-Mbit fiber when they install it," observes Podzimek.
Prior to releasing its current generation of products, Transition had mar keted a 100 Base-SX product without the auto-negotiation feature. "We've actually been relatively surprised at the strength of the sales, especially outside of the U.S. market, predominantly in the Scandinavian regions," Podzimek reports. "The 300-m limitation of SX hasn't been an issue in that, there, it's really being deployed as it was designed--as a replacement for the copper. And in most situations, that's under 100 m, so it's actually providing a lot more distance than is required. So it truly is being put in place like a 10/100 NIC card in copper has been done for the last five years."
While the members of the FOLS are hopeful that 100Base-SX will open the door to new customers, they appear to have no illusions about the size of the room they'll find on the other side. "SX is still not a mass-market solution," Podzimek comments. "Fiber is still a specialty kind of setup, and fiber is still a premium, even at the SX cost structures. It's still something that the cable installer is not comfortable with--it's still fiber. So we're not advocating 10/100 SX as a mass-market solution. What we're saying is that 10/100 SX removes some of the migration barriers that people had in the past, and so it becomes a wider market but not a mass market."
Even a few percentage points growth in market share in the lucrative horizontal LAN space would spell victory for fiber. "I think you could talk to anyone from the FOLS organization; they would be thrilled with 10 or 20% market share of all desktops [becoming cabled] with fiber," says Podzimek. "We're not anywhere near that today."
The relatively small size of the potential market will likely limit the number and size of the players that will pursue it. Transition is counting on this remaining the case. "The reason that we're doing the NIC rather than a standard NIC offering is that there aren't a lot of people providing them, and so there is a market for a smaller company such as ourselves," Podzimek explains. "Fiber hubs are still a small part of the market, so it's not going to be something you're necessarily going to see from the 3Coms of the world. It's not a mass-market solution."
The relatively small market size may even limit the areas that FOLS members choose to pursue. For example, 3M (Minneapolis, MN) is currently debating whether to support auto-negotiation in the 100Base-SX products currently on its drawing board. "First of all, is there enough of a market to interject the initial cost of the circuitry to handle that?" questions 3M's Dan Silver, who chairs the FOLS. "Because if there's not that much 10-Mbit on fiber today, and maybe the reason they put in fiber was to go with 100, is there really a need to go with a 10/100? The other thing is that you can still use the card with 10 or 100; it's just manually switchable on some of the cards. So you have to look at [whether] there is enough of a volume to get into the business."
Dixon believes that several products will support auto-negotiation. However, he sees a time limit on how long customers might require such a feature. "I think that maybe the window for 10/100 auto-negotiation may have passed, simply because the swing is predominantly now toward 100 Mbits," he explains. "[Auto-negotiation] was one of those products that, if we had been able to nail it early, early on, it would have been a great thing. But now as networks are migrating faster and faster, the attraction of being able to auto-negotiate between 10 and 100 may not be as great, simply because there's not that much 10 out there that might be upgraded."
Podzimek says that auto-negotiation may be most suited for NIC and media-converter products. "The advantage of auto-negotiation in the NIC and the media converter is the fact that you don't have to change or take anything out, whereas people are going to be replacing [their hubs anyway]," she says. "It's a wiring-closet setup; they can replace their hubs, they can replace their switches as necessary without any huge investment because they're not having to open up a system, they're not having to go to the desktop. So I do think that's a media-conversion opportunity." However, Podzimek does anticipate that someone will offer hubs with auto-negotiation.
In fact, Silver anticipates that several companies should enter the 100Base-SX market in the near future. "Most of us have hedged our bets and have produced an 850-nm short-wavelength 100-Mbit product, whether it's a media converter or a NIC, with the anticipation that the standard is going to be ratified," he says. "As far as the technical part of [the standard], that's been pretty well agreed to for over a year. So many of us have products on the shelf ready to go, and we're comfortable that they're going to work."
As proof of his point, the Short Wavelength Fast Ethernet Alliance held an interoperability demo in October 1998. Companies such as 3M, Allied Telesyn International, AMP (now part of Tyco Electronics), Corning, IMC Networks, JET Networks, LANart (now part of Transition Networks), LANCast, Lucent Technologies, Ortronics, Siecor, and Transition Networks interconnected a variety of active and passive equipment to a network consisting of 300 m of multimode cable.
Companies wishing to join this list will find it relatively easy to transform an existing 100Base-FX product into 100Base-SX. "Implementing 100Base-SX doesn't need to be any different than 100Base-FX, in that you're still probably going to use the same transceiver suppliers," says Podzimek. "The biggest difference for us is that we made the choice, although it was not a requirement, to not use standard off-the-shelf ASIC chips but to do our own to provide a higher level of features within it."
Chips for 100Base-SX have started to become available. For example, Micro Linear Inc. (San Jose, CA) began sampling its ML6696 100Base-X fiber physical-layer chip in 1998. According to the company's Edgardo Laber, adding a 100Base-SX product to one's roster requires little more than "changing the sugar cubes"--the optical components--from 1300 nm to 850 nm and swapping some chips. Micro Linear is probably the leading supplier of chips in this emerging area, although JET Networks (Tucson, AZ) has offered a 100Base-SX media-converter development kit based on its 10/100 Ethernet media-converter controller.
Firms that continue to shun fiber in their product lines should at least keep the 100Base-SX movement on their radar screens. "If there wasn't a need, we wouldn't have had so many of the various companies come together in a short period of time and work very hard with weekly conference calls and face-to-face meetings in putting all this together," concludes Silver. "It was really a great step in the right direction with the recognition that fiber is getting closer to the workstation. It's real, there's a need, and let's address it."
Acknowledgement: Information on auto-negotiation was derived from "10/100 Base-SX: Lowering the Cost of Fiber Migration" by Transition Networks Inc.