Timing and market conditions could not be better for ratification of the short-wavelength standard.
Steve Stange
Transition Networks
The most significant development in recent months for desktop networking is the proposed 100Base-SX standard, which defines an affordable migration strategy for 10-megabit-per-second copper or fiber to 100-Mbit/sec fiber, using a cabling structure capable of Gigabit Ethernet speeds and beyond. For the first time, those interested in investing in a fiber infrastructure will have a road map that clearly defines the migration path from 10Base-FL to the future (see "The standards development process," page 50).
The proposed 100Base-SX standard defines a new physical-media-dependent (PMD) interface as well as a means for implementing auto-negotiation. It does not define anything new above the physical layer, but rather is intended to be used along with the IEEE 802.3 1998 edition standard to provide a complete definition. Above the physical layer, 100Base-SX is governed by the IEEE 802.3 standard, meaning that it will work with all of the Ethernet and Fast Ethernet equipment that currently makes up more than 80% of the premises local-area-network (LAN) market.
What is different about the physical layer, or PMD interface, as defined by 100Base-SX? The primary difference between 100Base-SX and 100Base-FX is the wavelength of the light-emitting diode (LED) used to launch the light signal into the fiber. 100Base-FX was written for the requirements of a 1300-nanometer LED. The 1300-nm wavelength was chosen so that the standard would support a full 2-kilometer distance, as did 10Base-FL. Unfortunately, although the distance was compatible with 10Base-FL, the optoelectronics were not compatible because 10Base-FL was written in accordance with the requirements of an 850-nm LED source.
Systems operating according to the proposed 100Base-SX standard would use an 850-nm LED source. This configuration has two benefits: Such systems are compatible with 10Base-FL, and they can use a lower-cost light source. One drawback of using an 850-nm LED is that channel distance is limited to 300 meters. Although this limitation may seem harsh, many estimates of the installed-fiber base show that most multimode fiber has a distance of 300 meters or less in the horizontal portion of the network. Surely, any fiber being installed in desktop applications will be within these limits.
When considering the PMD interface, people frequently ask about fiber type and connector support. The proposed standard supports both 62.5/125- and 50/125-micron fiber. Equipment developed around the standard will carry a "type rating" that will determine whether the product works on 62.5/125-micron fiber (Type I) or 50/125-micron fiber (Type II). Some equipment may be specified to work on both Types I and II, or it may contain a switch or jumper to configure it to one type or the other.
As usual, connectors continue to be a significant issue. The proposed 100Base-SX standard addresses this issue by specifying that the SC, ST, and any alternate connector design must meet the requirements of the corresponding connector intermateability standard and the optical-connector requirements of the proposed ansi/tia/eia-568b.3 standard. This statement allows the proposed standard to conform with the rest of the industry as it pertains to connector choice. All of the small-form-factor connectors currently on the market should readily meet these requirements.
The PMD section of the proposed standard also lists figures for details such as spectral width, optical-power requirements, rise and fall times, and jitter. While these parameters are important, they are not what makes the proposed 100Base-SX standard unique.
Auto-negotiation
Since the proposed 100Base-SX wavelength is the same as that used by 10Base-FL, there needs to be a method for migration and auto-negotiation between the two. The standard, therefore, defines auto-negotiation as the mechanism by which two Ethernet or Fast Ethernet ports can communicate with each other during the linkup phase of the connection. This capability has been around for some time in the 100Base-TX standard. In fact, the 10/100 auto-negotiation capability is now found in network interface cards (NICs), switches, and dual-speed hubs.
Auto-negotiation takes place when two devices are first connected. Each sends out a message (at the auto-negotiation level, not an Ethernet message), stating the functionality of which it is capable. Currently, the functionalities in question are speed (10 or 100 Mbits/sec) and mode of operation (full- or half-duplex). Each device then examines the message it received from the other device and compares it to its own capabilities. The devices automatically select the highest-performing combination for signal transmission. For example, if a dual-speed hub is connected to a 10/100 switch, the hub will send a message to the switch capable of both 10- and 100-Mbit/sec transmission speeds and a half-duplex mode of operation. Meanwhile, the switch sends a message to the hub that it is capable of 10- and 100-Mbit/sec speeds and a full-duplex as well as half-duplex mode of operation. Since both devices support 100 Mbits/sec, that will be the selected transmission speed, and since the hub supports only half-duplex, that will be the mode of operation.
The proposed 100Base-SX standard implements auto-negotiation by adopting the IEEE 802.3 1998 edition`s Clause 28 (the portion of the standard that defines auto-negotiation for 100Base-TX) and defining how it needs to be converted for use with fiber. As a result, the information passed and the state machine (the sequence of events and the logic that controls them) used are compatible with all of the existing 10/100 equipment and installations.
Auto-negotiation accomplishes the 10- to 100-Mbit/sec conversion by encoding messages over the fiber. For 100Base-SX devices to be compatible with existing 10Base-FL equipment, a 1-megahertz idle signal is interrupted periodically (every 16 microseconds) with a burst of data. These data bursts are encoded by skipping clock-pulse transitions in the 1-MHz idle signal. These skips--called fiber link negotiation pulses--are then repeated every 62.5 microseconds, alternating between clock and data pulses.
In the figure on the right, there are a total of 16 data pulses and 17 clock pulses. The first and last pulses are always clock pulses. The 16 data pulses (or bits) comprise what IEEE 802.3 1998 edition terms the "link code word," which carries information about speed and mode of operation.
What happens to auto-negotiation if the 100Base-SX device is connected to a legacy 10Base-FL device that doesn`t know about or understand auto-negotiation? In that case, parallel detection monitors the link, looking for legacy-type signaling such as would be received from a 10Base-FL device. If such signaling is detected, auto-negotiation is stopped and the interface links at the detected speed and always in half-duplex mode of operation.
Network solutions based on unshielded twisted-pair (UTP) cable have been successful for two reasons: They offer a smooth migration path and low cost. Fiber, on the other hand, has had no clear migration path and has carried a cost penalty.
Those who installed UTP--probably Category 5--several years ago also made an investment in 10Base-T electronics. These networks performed well at 10 Mbits/sec. As 10/100 NICs came to market, it was an easy decision to buy them because they would work with all existing infrastructure--cable and electronics--and promised a smooth upgrade in the future. Soon, 10/100 switches came to market and suddenly all of those devices that had 10/100 NICs were now running at 100 Mbits/sec.
To make the migration complete, dual-speed hubs soon were on the market, allowing workgroups to exist that had computers running at 10 and 100 Mbits/sec. Additionally, all of them could share a 100-Mbit/sec connection to the backbone. It was easy for companies to upgrade their networks from 10 to 100 Mbits/sec incrementally. As budgets became available or groups moved, they could be upgraded in part or in total, and all the parts worked together.
Fiber-optic-based electronics do not offer the same upgrade path for electronics. If you have devices operating at 10Base-FL speeds, there is no way to integrate them with devices running at 100Base-FX speeds. To upgrade a workgroup from 10 to 100 Mbits/sec requires changing every NIC (or media converter), every hub, and every switch that is used. It is an all-or-nothing proposition. The only part of the network that does not need to be changed is the cable; the multimode fiber that was installed was--and still is--more than adequate for the job.
Not only do the fiber electronics offer no migration path, but they also carry a cost premium because of the expensive optoelectronics they use. At 100 Mbits/sec, the 100Base-FX standard makes this problem even worse by specifying the more expensive, 1300-nm LED as the transmitter source. When all these factors are considered, it is not surprising that fiber-to-the-desk has not taken off.
However, several new factors in the marketplace will change this situation, including development of 100Base-SX, reduced cost of optoelectronics, and legacy Category 5 networks. The development of the 100Base-SX standard is timely because it coincides with the other two factors. Not only has the cost of optoelectronics continued to decline, but new technologies on the horizon promise to drive costs down even further. Perhaps more importantly, many companies will have to decide in the not-too-distant future what to do with their installed Category 5 cable. As they try to move beyond 100 Mbits/sec to gigabit-per-second speeds, their installed cable may need to be replaced. The IEEE 802.3 group, which is working on the 1000Base-T standard for Gigabit Ethernet over copper, says the standard will work on Category 5 installations, provided that the cabling also passes some additional tests to qualify each run. However, many in the industry feel that a significant number of the runs will not pass these additional tests and will need to be upgraded.
The UTP market is also promoting Category 5E, Category 6, and Category 7. Each of these proposals is targeted at providing Gigabit Ethernet in the horizontal segment of the network. These standards do not provide futureproofing; they merely catch up to what fiber has been doing for more than a year now. Furthermore, the Higher Speed Study Group of the Institute of Electrical and Electronics Engineers (New York City) is already discussing 10-Gigabit Ethernet.
The industry is now ready for a fiber solution that enables companies to operate their legacy equipment today and migrate to the future without having to reinstall the cable plant.
Companies are developing a wide variety of products for 100Base-SX. As the proposed standard gains acceptance, companies will begin rolling out products to support it (see "How to migrate your network with 100Base-SX," above). The proposed 100Base-SX standard will bring to fiber networks the features that have made UTP-based networks so popular. A migration path at an affordable price is the key to making fiber successful in today`s market.
Auto-negotiation makes use of a 1-MHz idle signal in a data stream to encode its message.
In the 100Base-SX auto-negotiation scheme, the link code word, consisting of 16 data pulses, carries information about transmission speed and mode of operation between dissimilar devices.
The standards development process
100Base-SX was not born in the usual manner. Instead of the proposed standard being developed by the Institute of Electrical and Electronics Engineers (IEEE--New York City), interest in 100Base-SX originated in the Fiber Optics LAN Section (FOLS) of the Telecommunications Industry Association (TIA--Arlington, VA). This group has a charter to "educate end users and network designers about the technical advantages and affordability that optical transmission brings to local area networks and specifically to horizontal cabling applications."
Several companies in the group independently recognized the need for this type of solution. At a FOLS meeting, members decided to make a proposal to the IEEE to see if the organization would consider developing such a standard. The IEEE declined the opportunity. Sensing that the need was real, FOLS formed the Short WaveLength Alliance to develop a draft of what could become the standard. Since FOLS is not a standards-making body, it asked the TIA if any of its groups with standards-development authority would develop such a standard. The TIA agreed, and the work was assigned to the FO 2.2 group.
There is an open liaison between TIA`s FO 2.2 group and IEEE 802.3. Several of the members of FO 2.2 are also active members of IEEE 802.3. Members of IEEE 802.3, who were not formerly members of FO 2.2, have joined and are actively providing input to the standard`s development.
The proposed standard has already gone out for ballot once, and like almost every standard, did not pass on the first round. All the comments submitted during that ballot have either been incorporated into the standard or have been addressed in another manner. At press time, the proposed standard had been reissued for balloting, which will be followed by a comment period. If necessary, it will be re-balloted again. Once the ballot has passed, it will then go to American National Standards Institute (ANSI--New York City) and eventually become an ANSI standard.
How to migrate your network with 100Base-SX
How could you use 100Base-SX products to migrate from a 10Base-T network today to a 100-Mbit/sec network that uses fiber and can handle all of the proposed desktop networking speeds?
Starting with a typical network configuration, imagine there is an existing Microsoft Windows NT server with a 10/100Base-TX NIC connected to a 10/100Base-TX switch, which in turn is connected to a workstation with a 10Base-T NIC. The link to the server operates at 100 megabits per second, while the workstation operates at 10 Mbits/sec.
The migration starts by installing the fiber and media converters. In the top figure, the NIC in the server has been replaced by a 100Base-SX NIC. At this point, all the major local-area-network electronics are the same as before and the performance has not changed. The next step in migrating the network to maximize performance would be to replace the NIC in the workstation. If this NIC is replaced, the stand-alone media converter is no longer needed, and the workstation is connected directly to the media converter chassis in the main wiring closet (bottom).
The 10/100Base-TX switch could eventually be replaced with a 100Base-SX switch. This final step would obviate the media-converter chassis but would not result in any performance increase. When migrating the network to Gigabit Ethernet, the fiber infrastructure that was put in place will be more than adequate.
Steve Stange is senior product manager for Transition Networks (Minneapolis, MN) and chairman of TIA`s Fiber Optics LAN Section. He can be reached at [email protected].