Ethernet fiber media-conversion technology holds key to future LANs
Yoseph L. Linde
LANart Corp.
Local area networks with mixed media infrastructures may become the norm in the next few years, thereby focusing more attention on the next generation of media converters.
Optical-fiber deployment in local area network (LAN) backbones has increased significantly in the last two years. The consensus is that fiber will eventually replace copper as the medium of choice for LANs, and even reach the desktop. There is some debate, however, about the time frame of the copper-to-fiber transition.
Today, the percentage share of fiber-optic links in LANs is in the low single digits. Even if the share of optical links in new plants doubles each year, it will take more than seven years before fiber is the dominant wiring medium in the installed base.
While fiber has many advantages, its relatively high cost will slow its adoption, creating a period of several years in which mixed-media network infrastructures are common. As a result, media-conversion technology is central to the deployment of cost-effective, reliable networks.
Interfacing with LAN hardware
Most installed LAN equipment provides only the twisted-pair interface. And it doesn`t make sense for users to replace hubs, switches, or network interface cards (NICs) just because the media interface is different. Many vendors offer fiber media interfaces as an option. If fiber is the dominant plant wiring, then fiber interfaces are probably the right approach. If not, consider the following issues: From the vendor`s point of view, fiber interfaces are a specialty item. The volumes are relatively small, so the cost is higher. And, in general, the market is not as competitive as the twisted-pair market. It`s not unusual for a switch or hub vendor to charge 10 times as much for a fiber interface relative to the twisted-pair version. With very few exceptions, hubs and switches do not support different media interfaces on a port-by-port basis. On a standalone or stackable fiber switch or hub, typically all eight to 16 ports are fiber. The choice for a non-backbone fiber interface on a modular switch or router is a fiber module with all fiber ports (typically four to 12). Therefore, if users need only a few fiber connections, they end up paying a lot for unused ports. The problem is exacerbated when a user needs multiple speeds (10 and 100 Mbits/sec), a mix of singlemode and multimode fibers, or short and long optical wavelengths.
Fiber media conversion
External media converters are a popular solution to the interface problem. A media converter is a simple, inexpensive device with both a twisted-pair port and fiber port. It converts the twisted-pair signals into optical signals and vice versa (see Figure). This conversion can be done selectively on a port-by-port basis. In addition, because of the low cost of twisted-pair interfaces, the combined cost of the media converter and a twisted-pair port on a hub, switch, or router is usually less than the cost of a fiber port on the same device--and ports don`t go to waste.
A converter is a full-duplex device; the conversion from twisted-pair to fiber is independent of the fiber-to-twisted pair conversion. Ideally, the converter is completely transparent. If a fiber port is connected to a twisted-pair port via the converter, each respective port operates as if it is connected to the same media interface--fiber to fiber and twisted-pair to twisted-pair.
A media converter does not alter the data rate. Devices that do are called rate converters and, in effect, are 2-port switches. The most popular media converters today operate at 10 and 100 Mbits/sec (see Table). Gigabit Ethernet converters are not widely used because most of these interfaces are anyway.
The existing IEEE 802.3 standards are 10Base-T, 100Base-TX, and 100Base-FX. A proposed standard, TIA/EIA-785 for 100Base-SX, uses a short 850-nm wavelength at 100 Mbits/sec. The existing 100Base-FX standard uses 1300-nm long-wavelength optics. Short-wavelength optical components are less expensive than long-wavelength components, but these devices support a more limited range. The 100Base-SX standard, like 100Base-TX, defines an optional auto-negotiation mechanism called N-way, which allows 100-Mbit/sec ports to communicate with legacy 10-Mbit/sec ports at the lower speed.
In addition to the converters listed in the table, many vendors offer a singlemode fiber interface at 10 and 100 Mbits/sec. Although the 10- and 100-Mbit/sec standards do not define such an interface, product interoperability among vendors is usually not an issue because the optical signaling is the same as that used in the standards.
Packaging options
There are four popular packaging options for media converters:
Single: The individually packaged single media converter is probably the most popular version (see Photo 1). The vast majority of such devices use a wall-mount power supply, although more-expensive versions with internal power supplies are also available.
Chassis: A chassis is a rack-mount device that includes a power supply. Some vendors offer power redundancy as well. Depending on the vendor, a chassis allows eight to 16 single converters to fit neatly into a rack space without the problem of individual mounting and a multitude of external power supplies (see Photo 2). Different types of converters can be mixed and matched inside the same chassis.
Multichannel: The most cost-effective solution in cases where many converters of the same type are needed, the multichannel or "rack" converter is a single rack-mount box with an internal power supply. It contains a fixed number of converters of the same type, typically 6 to 12. It is also the most efficient in the usage of rack space.
Modular: A modular unit is a rack-mountable, powered card-cage that accepts typically eight to 16 plug-in cards, each with a single or multiple converter. The concept is similar to that of the chassis and, theoretically, because the modules are not individually packaged, the cost should be lower. Although the chassis modules can be used as standalone units, the modular cards are useless without the card-cage.
Additional features
In addition to media conversion, vendors have added useful features to these devices to enhance functionality and differentiate their products.
Full transparency: An ideal converter is totally transparent. It responds to not only data and idle signals but link faults as well. If the converter`s twisted-pair link input is down, for example, its optical-fiber output should also be down and vice versa. Full transparency is useful because it allows a managed port to detect and report faults in whichever segment (fiber or twisted-pair) they occurred. Without full transparency, some faults are masked.
Steady state delay: A good converter will have a very small delay or, ideally, no delay. The delay is especially critical at 100 Mbits/sec because the delay budgets are very tight. Most 100-Mbit/sec converters use standard transceiver chips that perform serial-to-parallel and parallel-to-serial conversion. These devices also use first in, first out operation, or a repeater, to connect the two transceivers, resulting in significant delays. Some converters use pure serial conversion, which produces a shorter delay by a factor of three to six.
Twisted-pair cross-over: Twisted-pair ports come in two flavors: crossed and straight (uncrossed). A crossed port can be connected to a straight port with a straight cable. But a crossed-over cable is needed to connect two crossed or two straight ports to each other. Terminal-type ports (for example, a PC NIC) and backbone ports are straight. Downward-looking hub and switch ports are crossed over, allowing straight cables--the preferred type--to be used almost universally. A converter may be connected to either type. Thus, a converter with a fixed interface configuration may require a crossed-over cable in some cases. Vendors have taken two approaches to solving this problem: a mechanical switch that allows the user to select a crossed or uncrossed configuration or a "smart" interface that automatically adapts to the configuration at the other end, independently of the cable used.
Mass termination: Instead of individual RJ-45-style connectors, a single telephone connector provides connectivity for up to 12 twisted-pair ports, simplifying the connection to switches and hubs with the same multipair component. This feature is only available in multichannel and modular converters.
Redundancy: Some users migrate to fiber because of higher reliability and noise immunity. For users concerned with downtime, redundancy is important. Some "rack" converters, chassis, and modular converters offer power supply redundancy. Converters also can provide fiber link redundancy.
Full-duplex support: Full-duplex Ethernet, theoretically, allows an unlimited link distance (2-node half-duplex Ethernet is limited to about 4 km at 10 Mbits/sec and 400 m at 100 Mbits/sec). Virtually all twisted pair-to-fiber converters are full-duplex. But problems can arise when these devices are connected to auto-negotiating 10/100 twisted-pair ports.
If an auto-negotiating port is connected to a simple converter, it cannot detect whether the converter is capable of full-duplex communication (because the converter does not implement N-way) and thus will default to half-duplex. The way to get around this problem is to force the port into full-duplex mode via management. As an alternative, some vendors have added a subset of N-way to the converter`s twisted-pair port (at 100 Mbits/sec), which allows it to advertise its full-duplex capability. This feature can be enabled only if the fiber port is full-duplex as well.
Because no converters on the market today implement N-way on fiber--100Base-SX will add this feature--there is no way for the converter to know if the fiber interface is full-duplex, so the configuration must be manual. Therefore, the choice is between a manual switch on the converter versus a manual software switch on the remote twisted-pair device. Still, this is a useful feature in the case of low-end 10/100 switches and hubs that do not have any management or configuration capabilities.
Problems solved by the next generation
Existing 100-Mbit/sec twisted pair-to-fiber converters have several drawbacks. For starters, these devices are not backward compatible with the 10-Mbit/sec 10Base-T and 10Base-FL standards. 10Base-FL uses a short light wavelength while 100Base-FX uses a long wavelength, creating a basic incompatibility at the lowest physical level. Secondly, today`s converters are not fully transparent. Therefore, these devices do not pass auto-negotiation signals from twisted-pair to fiber and vice versa; full-duplex auto-negotiation support is manually configured and only at 100 Mbits/sec. The problem is the lack of an auto-negotiation (N-way) standard for fiber. Thirdly, cost is still relatively high. Long-wavelength 100-Mbit/sec optical transceivers are approximately three times more expensive than 100-Mbit/sec short-wavelength transceivers.
The proposed 100Base-SX TIA standard, which is likely to be approved by mid-year, will solve these three issues. It uses short-wavelength optics and specifies optional N-way auto-negotiation on fiber. The standard will allow the construction of converters that are truly transparent (including auto-negotiation), provide backward compatibility with 10-Mbit/sec fiber links, and bring the cost of 100-Mbit/sec conversion close to the cost of 10-Mbit/sec conversion.
As usual, however, there is a price. The 100Base-SX standard, which supports a distance of 300 m (about 1000 ft), is not compatible with 100Base-FX, which is capable of 2000 m (approximately 1.25 mi). The 300-m distance for 100Base-SX is very conservative, and in all likelihood, 500 m is achievable with better control of the optical drivers and receivers. The latter distance is more than sufficient for desktop-wiring closet connections and for most home-runs (desktop to data center via a wiring closet). However, it is not sufficient for many backbone applications, so there is still a need for 100Base-FX.
Clearly, the use of optical fiber is becoming more common in LANs. In all likelihood, networks deployed during the next five to 10 years will make use of twisted-pair copper and multimode and singlemode fiber at both 850 and 1300 nm, because each of these media has advantages in specific circumstances.
Current and future generations of media converters will simplify the joining of these various media, both in terms of flexibility of design and ease of installation. Media converters have evolved quickly, and emerging standards and technologies promise more improvements in the immediate future.
These devices will also progressively reduce costs, especially the cost of fiber-to-the-desktop. As costs come down, fiber deployment will escalate and with it will come increased network reliability and robustness as well as the potential for higher network performance. u
Yoseph L. Linde is chairman and president of LANart Corp. (Needham, MA).