10-Gigabit Ethernet development for LAN cabling systems well underway

Dec. 1, 1999

Next-generation multimode-cabling solutions will allow network administrators to manage 10-Gbit/sec applications on the same fiber path as 10-Mbit/sec programs using familiar, economical technology.

In the 1990s, the data-networking industry developed numerous local-area network (LAN) protocols such as Token Ring, Fiber Distributed Data Interface (FDDI), Asynchronous Transfer Mode (ATM), and Ethernet. All share the common characteristic of pushing data rates higher to serve the increasing demands of users.

Of these protocols, Ethernet has become the most widely accepted networking application. Today, 10-Mbit/sec Ethernet is the prevalent desktop application, but the 100-Mbit/sec upgrade, known as Fast Ethernet, is quickly displacing its lower-speed predecessor. Because switches aggregate bandwidth from many users, the backbone network that interconnects these switches must support even higher rates (see Fig. 1).
Fig. 1. Today, 10-Mbit/sec Ethernet is still widely used for desktop applications, but 100-Mbit/sec Fast Ethernet is rapidly displacing its lower-speed predecessor. Enterprise switches aggregate bandwidth from many users, thus the backbone network that interconnects these switches must support even higher rates such as 1 Gbit/sec.

As Fast Ethernet is deployed to serve speed-hungry desktops, 1000-Mbit/sec speeds are needed in the backbone. Fortunately, the networking industry has responded to this need by developing standards and equipment that supports the latest Ethernet solution, known as Gigabit Ethernet (GbE, operating at 1 Gbit/sec).

But gigabit speeds are not just contained within the backbone. By 2005, we expect to see a significant deployment of gigabit applications to the desktop. These applications will be enabled by low-cost GbE network interface cards (NICs) running over high-grade unshielded twisted-pair (UTP) copper cables that are predominate in the horizontal portions of enterprise networks. NICs that meet the GbE specification for copper cable (1000 Base-T) were recently introduced and are priced significantly below their optical counterparts. These NICs most likely will drop in cost to a level that could drive significant adoption of GbE to the desktop within 5 years.

The Institute of Electrical and Electronics Engineers (IEEE) has recognized the need to address the impending backbone bandwidth bottleneck. Just as the dust begins to settle from the development activities that delivered the GbE standard, the IEEE 802.3 working group has begun the process of developing the next evolutionary step in the Ethernet family. The IEEE 802.3 Higher Speed Study Group has been in operation since March 1999 with participation from over 55 different companies. Given the historical progression of Ethernet speeds in tenfold increments, it should come as no surprise that the new target speed is 10 Gbits/sec, or 10-GbE.

To put 10 Gbits/sec in perspective, it is interesting to note that the evolution of the processor continues at exponential growth rates closely following Moore's Law, postulated by Intel cofounder Gordon Moore in 1965. Moore's Law states that the number of transistors on a chip doubles every 18 months (equivalent to 10 times every five years). This growth in transistor density has driven a commensurate increase in processor speeds (millions of instructions per second, or MIPS), to support faster processing and transmission of large files.

Based on these expected growth rates and historical increases in network capacity demands, seasoned network managers have learned that futureproofing bandwidth is essential when specifying a cabling system. They face a vexing question when installing cabling systems: What applications will this system have to support during its lifetime? The life of a typical cabling system is seven to 12 years, while LAN applications are upgraded every three to four years. Thus, cabling systems may need to support up to four generations of LAN applications.

Until recently, prevailing wisdom was that multimode fiber bandwidth would support any foreseeable application in distances useful in campus networks. This perception holds true for serial transmission up to 155 Mbits/sec. Above this rate, however, the bandwidth limitations of today's installed multimode fiber restrict its distance capability to less than 2 km.

Consequently, the IEEE limited the supported distance of conventional 50- and 62.5-micron multimode cables for GbE to distances of 220 to 550 m, far short of the 2-km distance such cables can support for Fast Ethernet and Ethernet applications. Due to these bandwidth limitations, the current installed multimode fiber base can only support 10-Gbit/sec speeds for about 60 m, unless complex optoelectronic techniques such as multichannel wavelength multiplexing are used in the equipment. Such short distances may be useful for equipment interconnections within a computer room, but fall far short of what is needed to serve building backbones.

In order to provide the most cost-effective solution, a new multimode fiber is needed that can provide the required performance necessary to support single-channel serial transmission, the simplest optoelectronic technique used in all fiber LAN solutions to date. In light of the limitations of today's multimode fiber, and customers' future needs, Bell Labs established the following key objectives for the development of a next-generation multimode fiber solution:

  • Meet the needs of in-building network applications with a single universal medium
  • Support lowest-cost and lowest-complexity system for LAN applications, which can support a minimum of 10 Gbits/sec
  • Provide a smooth migration path for applications from 10 Mbits/sec to 10 Gbits/sec
  • Remain "user-friendly" for cable manufacturer and installer.

A next-generation multimode fiber solution currently proposed by the cabling industry standards bodies of the Telecommunications Industry Association (TIA) and the International Organization for Standardization/International Electrotechnical Commission (ISO/ IEC) successfully meets all of these objectives. Here's how.

Surveys have consistently shown that the vast majority of building backbones are less than 300 m in length. Data from one such survey, gathered by the IEEE during the development of GbE, is shown in Figure 2. According to those surveyed, next-generation multimode fiber that supports 300-m link lengths for 10-Gbit/sec applications will likely meet the needs of most in-building LANs.
Fig. 2. Surveys have consistently shown that the vast majority of building backbones are less than 300 m in length. This data is from one such survey, gathered by the IEEE during the development of Gigabit Ethernet.

To that end, this next-generation multimode cable is designed specifically to support 300-m lengths for 10-Gbit/sec applications. Its industry standard 50-micron core size couples sufficient power from light-emitting-diode sources to support legacy applications such as Ethernet, Token Ring, FDDI, Fast Ethernet, and ATM for in-building network distances. The 50-micron core size is also directly compatible with laser-based applications such as GbE. Together, these capabilities satisfy the first objective of meeting the needs of in-building networks with a single universal medium.

Bell Labs examined the cost of various 10-Gbit/sec alternatives to determine the lowest-cost solution. The study considered the use of coarse wavelength-division multiplexing (CWDM) on the installed base of multimode fiber, upgrade paths using singlemode fiber and long-wavelength (1300-nm) lasers, and migration to an advanced next-generation multimode fiber using short-wavelength (850-nm) vertical-cavity surface-emitting lasers (VCSELs).

Figure 3 shows the relative estimated costs associated with each of these alternatives for a GbE backbone system plus the upgrade of a building riser backbone from GbE to 10-GbE. (Exact costs will differ in each case depending on the specific configuration and costs actually charged to the customer at the time of purchase.) In each of the cited cases, it is evident that the cost of the electronics far exceeds the cost of the cabling.
Fig. 3. This bar chart shows the relative estimated costs associated with different fiber solutions for the upgrade of a building riser backbone from Gigabit Ethernet (GbE) to 10-GbE. (Electronics includes LAN switch plus labor for installation. Cabling consists of cable, apparatus, and labor for installation.)

In Figure 3, the bar labeled "MM + WDM" shows the projected costs of using the installed base of multimode fiber with wavelength-division multiplexing (WDM). At 1 Gbit/sec, the electronics are cost-effective short-wavelength (850-nm) GbE devices, based on the standard known as 1000Base-SX. Migration to 10 Gbits/sec, however, requires the use of CWDM electronics, which have the highest projected cost due to the need for multiple lasers, combiners, splitters, filters, and detectors.

The bar labeled "SM" shows the projected cost of using singlemode fiber for both 1-Gbit/sec and 10-Gbit/sec applications. At 1 Gbit/sec, the electronics are long-wavelength (1300-nm) GbE devices, which are covered in the 1000Base-LX standard. Initially at 1 Gbit/sec, LX devices are more expensive than SX. But, the upgrade to 10 Gbits/sec would allow the use of single-channel, singlemode electronics (10-GbE LX), which are simpler than CWDM electronics. Overall, this solution is less expensive than the first, but higher in cost than necessary.

The "MM to SM" bar in Figure 3 shows the cost of a conventional GbE backbone system plus the cost of upgrading an existing multimode infrastructure to singlemode for 10 Gbits/sec. In this case, use of SX devices instead of LX electronics at 1 Gbit/sec more than offsets the cost of the additional singlemode cable installation, making this upgrade path less costly than the previous two.

The bar labeled "NEXT-GEN MM" shows the projected cost of using the next-generation multimode fiber-cabling system. Because this solution allows the use of single-channel short-wavelength electronics for both 1 Gbit/sec and 10 Gbits/sec, it has the lowest projected cost.

The next-generation multimode fiber solution also supports auto-negotiation between 1 Gbit/sec and 10 Gbits/sec on the same optical port. It is estimated that from 70% to 90% of installed GbE ports operate at 850 nm. By enabling 10-Gbit/sec operation at 850 nm, the next-generation multimode fiber solution will permit end users to upgrade backbone links incrementally, allowing for budget considerations, and minimizing disruptions to network operations.

Work is underway within the laboratories of several companies to develop and perfect low-cost 10-Gbit/sec VCSEL and detector technology operating at 850-nm wavelengths. Bell Labs worked with these devices as well as cutting-edge test and measurement equipment to develop a fiber-cabling system that supports these technologies in real-world applications.

The connectivity solution for the next-generation multimode fiber-cabling system supports existing applications as well as future 10-Gbit/sec applications without changing any of the cabling or connecting hardware. This allows users to have a ready upgrade path simply by plugging in new electronics. The emerging 10-Gbit/sec VCSEL technology, combined with this flexibility, will satisfy the third objective of providing a smooth migration path from 10 Mbits/ sec to 10 Gbits/sec.

Next-generation multimode fiber is also easy to cable and install. To satisfy this objective, the Bell Labs design team selected the core diameter of 50-micron fiber. This last objective is essential to industry acceptance. Today, 50-micron cable is manufactured and installed by many companies worldwide. Next-generation cable will use the same termination and test procedures already familiar to installers.

At the same time, it offers superior performance to conventional 50-micron multimode fiber. It is optimized for 10-Gbit/sec 850-nm VCSELs while retaining the performance required to fully support legacy applications. The critical performance parameter needed for 10-Gbit/sec transmission is laser bandwidth.
Bell Labs developed a test infrastructure that places extreme stresses on next-generation multimode fiber to prove the robustness of the system.

Bell Labs developed a test bed that places extreme stresses on the next-generation multimode fiber to prove the robustness of the system (see Fig. 4). The VCSEL launch condition can be varied to simulate transmitter production variations. The launch fiber includes mode scrambling to induce mode-field-diameter expansion, which could result from cabling induced microbending. Fiber shakers and adjustable insertion-loss connections allow testing of modal noise effects.

All of these impairments are introduced at the beginning of the link, prior to transmission through 300 m of the next-generation multimode fiber, to create a test environment exceeding worst-case operating conditions. A bit-error-rate-test set drives the VCSEL and compares the received signal, bit-for-bit, with the transmitted signal. An attenuator allows measurements of system operating margin. Under all of these simultaneous impairments, the system creates less than one bit error with every trillion transmitted bits.

This performance is attributed to careful attention to the light propagation properties of the fiber. In traveling from one end of a multimode fiber to the other, the light pulses can take many different paths, called modes. Like cars traveling on a multi-lane highway, each lane can travel at different speeds. In order for the fiber to support very high-speed digital transmission, the modes must all travel at nearly the same speed so that one pulse does not mix with the next.

The allowable difference in the propagation time between modes over the entire length of the link must be much less than the duration of one bit. As the data rate increases, the pulses become shorter and the propagation time between modes must be nearly equal. This effect is graphically depicted by plotting the difference in propagation times between modes in the multimode fiber. The conventional multimode fiber characteristics cause the light pulse to arrive at different times so that the signal energy is spread out over many bit periods causing interference and errors. The next-generation fiber characteristics keep the signal concentrated in a single wave that doesn't cause interference between bits allowing error-free transmission (see Fig. 5).
Fig. 5. The allowable difference in the propagation time between modes (light-pulse paths) over the entire link must be less than the duration of 1 bit. As the data rate increases, the pulses become shorter and the propagation time between modes must be nearly equal.

The cabling industry standards bodies of the TIA and the ISO/IEC recently agreed to develop specifications for a multimode fiber that supports 10-Gbits/sec transmission over 300 m for 850-nm systems. Revisions of the Commercial Building Telecommunications Cabling Standard, TIA/EIA 568a, and the Generic Customer Premises Cabling standard, ISO/IEC-11801, are now in progress. In light of these proposed cabling standards, the IEEE 802.3 Higher Speed Study Group decided to develop a 300-m, next-generation multimode fiber solution for 10 GbE.

These developments signal an exciting improvement in applications support for multimode fiber, breaking through the limitations of current technology. The next-generation multimode fiber solution will allow network administrators to manage in-building applications from 10 Mbits/sec through 10 Gbits/sec on one fiber path, using familiar and economical multimode technology.

John E. George is Systimax's fiber-offer development manager at Lucent Technologies (Atlanta), and Paul F. Kolesar is a distinguished member of the technical staff at Bell Labs (Murray Hill, NJ).