Although a 10-Mbps data rate seemed generous when Ethernet first appeared, today's network users increasingly view it as inadequate. As a result, IC and local-area-network (LAN) system vendors are rushing to define a faster Ethernet, one that runs at 100 Mbps. Two distinct IEEE standards for a fast Ethernet are emerging, but neither is fully developed. Some companies, however, have already released their own versions of 100-Mbps Ethernet.

The need for a faster Ethernet results from the combination of two factors: increasing numbers of network users and increasing user-bandwidth needs. The increasing number of users implies that the average bandwidth available to each user in a shared network is decreasing (fig 1). According to estimates from market-research firm International Data Corp (IDC), the average number of users per network segment will increase from 12 in 1991 to 21 in 1994. The average bandwidth available to each user thus drops more than 40%.

As available bandwidth is dropping, each user's bandwidth demand is increasing. The growth in processing power at each node, combined with the corresponding growth in data-transfer needs, boosts bandwidth demands of current applications. In addition, emerging data-intensive applications, such as video conferencing, contribute to the trend. These applications are time critical, which further complicates network needs. Audio and (to a lesser extent) video must be transmitted within a set period or they become impossible to reconstruct properly at the receiving end. If the network is crowded, conventional shared Ethernet cannot guarantee the timely network access such applications require.

Guaranteed access was one of several technical goals developers considered when choosing an approach to fast Ethernet. A second requirement was to use unshielded twisted-pair (UTP) wire for cabling—with as few pairs as possible. And a third goal was to limit the frequency content of signals along the UTP to 30 MHz to avoid FCC (Federal Communications Commission) Class A compliance testing.

Yet marketing—not technical—goals ultimately dictated the proposals vendors set before the IEEE 802 standards committee. Two markedly different proposals, representing distinct market views, have now reached draft-proposal stage: 100 Base-X and 100 Base-VG. Because these proposals serve different customer needs, the IEEE 802 committee decided (in November, 1993) that both would be refined into standards. The committee slated the 100 Base-X proposal to become standard 802.3U and the 100 Base-VG proposal as 802.12. Table 1 compares the two proposals.

A collection of networking product vendors known as the Fast Ethernet Alliance (see Table 2) spearheads the 802.3U, or 100 Base-X, proposal. Alliance members aim to leverage existing Ethernet system software and hardware to reduce the cost and risk of upgrading a system to fast Ethernet. Their proposal, therefore, calls for retention of Ethernet's present media-access-control (MAC) layer protocol.

The protocol, named CSMA/CD (carrier sense multiple access with collision detection), requires that nodes desiring network access first listen to the network to determine if it's in use. If the network is free, the node may begin transmitting but continues to listen to detect any collisions—simultaneous transmissions by two or more nodes. If a collision occurs, the CSMA/CD protocol calls for the contending nodes to cease transmission and try again following a randomly timed delay. If several collisions occur in succession, the nodes double their delay interval and keep trying.

Although the CSMA/CD protocol eventually allows one node to gain access, the delay between access request and acquisition is both variable and unbounded. Further, a heavily loaded network spends considerable time resolving collisions, which is known as thrashing. Thrashing causes available network bandwidth to drop well below the 10-Mbps data rate; by some estimates, as much as 40% of system bandwidth may be lost.

The 100 Base-VG proposal, which originates from industry giants Hewlett-Packard, IBM, and AT&T, seeks to avoid thrashing and to offer guaranteed network access by operating with a demand-priority protocol at the MAC layer. This deviation from CSMA/CD means that 100 Base-VG is not strictly an Ethernet variant; it is, however, a fast-Ethernet alternative, because it handles Ethernet data frames. 100 Base-VG also handles token-ring frames, thus providing an upgrade path for both Ethernet and token-ring users.

The demand-priority protocol of 100 Base-VG calls for the network hub to arbitrate network-access requests based on round-robin polling. The result is full utilization of network bandwidth. The hub polls for port-access requests, creates a requester list, then grants access to ports in the list order. Each requesting port gets a turn before the poll is taken again.

The demand-priority protocol also provides high-priority access, allowing a port to receive access ahead of its normal turn. Round-robin polling also determines high-priority request handling and requires that all high-priority users be serviced before normal-priority users gain access. Normal- priority users cannot be locked out, however. If a normal-priority access request has been pending for more than 250 msec, the hub automatically upgrades that request's priority status so that it is serviced in the high-priority sequence.

Along with their protocol differences at the MAC layer, the two fast-Ethernet proposals differ in their physical layer. The 100 Base-X proposal uses 3-level MLT-3 signaling employed by copper FDDI to reduce the signal frequency content on UTP cable. The scheme sacrifices the technical goal of less-than-30-MHz frequency content on the wires, however, by offering full-duplex transmission with only two wire pairs. The frequency content of 100 Base-X signals, therefore, mandates using data-grade (Category 5) UTP cable.

For new installations, using high-grade cabling is not a problem. Category 5 UTP cabling is now routinely installed in new network construction because its incremental cost is insignificant. Most networks, however, were built with voice-grade (Category 3) UTP cable.

To capitalize on that installed base, the 100 Base-VG proposal uses Category 3 cable in a scheme called quartet signaling. Quartet signaling divides the data stream among four wire pairs using 4B/5B block coding for error correction and NRZI bit coding for transmission. The result is half-duplex operation with each wire carrying a 30-MHz signal. If a better grade of wire or an optical fiber is available, the proposal allows the system to multiplex data streams together to take advantage of increased bandwidth.

The IEEE 802 committee decided to have the 802.3U group augment the 100 Base-X proposal with a scheme that would utilize Category 3 UTP as well. That scheme, called 4T-Plus, uses four pairs of Category 3 wire with 8B/6T block coding. The data run on three of the four pairs; the fourth pair handles collision detection. Because of the coding scheme, the data wires handle a maximum frequency of 25 MHz.

The advent of a second physical-layer specification under 802.3U prompted the IEEE committee to designate 100 Base-T as an umbrella term for fast Ethernet with 100 Base-X and 4T-Plus as subcategories for the two cabling types. The committee also elected to begin developing a media-independent interface for fast Ethernet similar to Ethernet's AUI (attachment-unit interface).

All of the fast-Ethernet proposals are still in committee, which implies at least a 6-month wait before the proposals attain draft-proposal status—and up to three years before they are full standards. In anticipation of the standards, products may begin to appear this year. AT&T and Hewlett-Packard have demonstrated working silicon of their quartet-signaling transceivers and anticipate releasing a complete 100 Base-VG chip set during the first quarter of this year. In addition, some nonstandard approaches have reached the market (see box, "Switched hubs unclog system bottlenecks").

Once both alternatives are represented in the marketplace, the success of either (or both) will ride on how well its (or their) developers predicted the market's response to MAC differences. Neither proposal provides an overwhelming technical advantage; hardware costs are likely to be comparable. The future of fast Ethernet boils down to a choice between speeding up what already exists and adopting a new protocol to guarantee timely access.

Looking Ahead
Until recently, debate over 100-Mbps Ethernet focused on which proposal, 100 Base-X or 100 Base-VG, would become the IEEE standard. The recent IEEE decision to pursue both has ended that debate; the conflict will now shift to the marketplace. The reuse of carrier sense multiple access with collision detection (CSMA/CD) gives the 100 Base-X approach a considerable market advantage. All of the system software and silicon developed for Ethernet will transfer virtually intact to a 100-Mbps network—with only a frequency change. Unfortunately, all of the problems also transfer. The alternative approach, 100 Base-VG, represents a risk. It requires new and unproven software and hardware and may introduce problems of its own. Advantages include its deterministic access scheme and applicability to both Ethernet and Token-Ring networks. In these lean economic times, expect reuse of existing resources as the approach favored by most of the market. The applications that require a deterministic access can be solved, at least in the near term, with switched hubs or carefully segmented network designs.