Editor's note: To view a PDF version of this article, click here.
This month's column will describe one of the industry's fastest-growth areas: Gigabit Ethernet. The pieces of the evolving Ethernet puzzle cover way too much ground for one column, so I will focus on some of the differentiating aspects of the Gigabit technology. Later, I will devote some words to the various efforts under way in "next-generation" Ethernet.
The very term "Gigabit Ethernet" itself has gone beyond buzzword status. There are two reliable indicators of this. First, talking about Gigabit Ethernet development is almost passe; knowledge of 10-Gbit Ethernet is what you want to put on a resume these days.
While reviewing papers for an IEEE conference, I noticed the second indicator: One paper described a wireless Gigabit Ethernet link operating at 60 GHz. These technology crossover ideas always happen when the two technologies in question have independently progressed beyond "questionable" status.
While the paper under review was well-written and intriguing, one of the driving forces behind Gigabit Ethernet development was to create a technology that could scale bandwidth without disrupting or replacing existing infrastructure, and at low cost. Getting the most out of the existing capital investment is a common theme in today's economy and in fact is fundamental to the approach that the telecom and cable industries took in the first place for DSL and cable modems.
The official reference for the Gigabit Ethernet standard is IEEE 802.3z. IEEE 802.3, a.k.a (take a deep breath) the Carrier Sense Multiple Access with Collision Detection (CSMA/CD) Access Scheme and Physical-Layer Specification, is the original Ethernet standard. The spec's verbose title harks back to a time when the primary innovation in Ethernet was the media-access technique, whereby a station would listen before speaking (carrier sense) and subsequently stop and warn other stations when transmission collisions between stations occurred (collision detection). The method led to more efficient use of a shared access medium, providing more throughput over simpler protocols, such as Aloha.
Nowadays, when we talk and read about Ethernet technology developments, the ideas are often more relevant to switched Ethernet segments and networks, rendering the key originating innovation of an efficient shared multiple-access scheme unimportant to the discussion. Nonetheless, 802.3 standards-based implementation details, including CSMA/CD, remain part of the technology as it increases in speed.
Specifically, the 802.3z committee's objectives were to evolve a 1,000-Mbit/ second system that could:
- perform half- and full-duplex;
- remain compliant with the 802.3 frame format;
- support the CSMA/CD access method;
- support backward compatibility with 10Base-T and 100Base-T systems; and
- support multiple predefined link lengths over multiple predefined media types.
The standard, whose development took a little over a year from start to finish, was completed in 1998 (hard to believe it's been five years).
Conveniently enough, and in keeping with the need for backward compatibility, the Gigabit Ethernet task force determined that from Open Systems Interconnect (OSI) Layer 2 (the data link layer) on up, Gigabit Ethernet and Ethernet would be identical. The physical layer—OSI Layer 1—would have to undergo some change, of course, to increase to 1,000 Mbits/s. Rather than reinvent the wheel (or, in this case, the photon transport mechanism), a successful physical-layer transmission approach was "borrowed" from ANSI X3T11, also known as the Fibre Channel standard.
The integration of these two already mature and proven technologies (Figure 1) accelerated the standards process as well as the hardware development itself, since each problem had already been solved, albeit in different arenas. Most important, there was no technology barrier between the two that would prevent their integration.
PHY Fun
Gigabit Ethernet achieves a tenfold data rate increase over Fast Ethernet systems (100 Mbits/s) while leveraging current infrastructure. There are some key differences, of course. The most notable is that the optical physical layer is front and center, whereas in 100-Mbit/s systems electrical and optical interconnects more or less share significance. In fact, Gigabit Ethernet is often referred to as optical Ethernet because of the primary role of optical transmission in its physical layer.
This optical focus resulted in two optical PHY and two electrical PHY standards (one still in the works). The optical physical layers are 1,000Base-SX and 1,000Base-LX, signifying "short wave" fiber-optic links and "long wave" links. In this case, short wave refers to low-cost lasers operating in the 850-nanometer window, covering link lengths up to 550 meters, per the standard. The long-wave lasers operate at 1,300 nm and, using single-mode fiber, cover at least 5 km. Both numbers are generally conservative minimum requirements. Obviously, this version is aimed at longer, backbone interconnects.
The third PHY that falls within the 802.3z standard is 1000Base-CX (the C stands for copper, meaning electrical Gigabit Ethernet). The cable of choice here is 150-ohm twin coaxial cable, and the distance guarantee is a mere 25 meters, indicative of what gigabit rates mean to your once-proud physical infrastructure capabilities. The application in mind is the wiring closet or rack-to-rack interconnect, where simplification becomes important for low maintenance.
Finally, there is a separate standards effort—IEEE 802.3ab (802.3 is running out of suffix letters)—that defines gigabit transmission over category five (CAT-5) cabling. Clearly, the goal here is to take advantage of even more already-deployed physical infrastructure, in this case the CAT-5 cabling that is already prevalent. Accomplishing this feat is not easy, and relies heavily on state-of-the-art digital communications technology, including multilevel signaling at a high symbol rate, as well as a real-time digital tx/rx pair with sophisticated equalization.
The use of Fibre Channel specifications deserves some mention here (particularly in light of Fig.1, where the functional step of "encoding" is mentioned). Fibre Channel uses a well-understood 8B/10B encoding/mapping scheme. The scheme is useful for serial transmissions, whereby an 8-bit word is translated to a unique 10-bit word. The process adds robustness and randomization characteristics to aid in the transmission, in particular helping synchronization and error-rate performance.
Clearly, there's no free ride to 10-bit transmission from an 8-bit payload: The approach carries a 25 percent overhead penalty, as any error control or correction scheme would. Thus the gigabit transmission on the fiber is, in fact, a 1.25-Gbit/s transmission.
MAC attack
Though GbE's frame format makes it compatible with its predecessors, the 1,000-Mbit/s rate creates some challenges for implementing the CSMA/CD shared-access protocol, although in practice this detail is not widely significant.
One constraint with this access technique is that the shared line results in some ambiguity as to whether the lines are busy, based on the finite transit time of information on the bus and the speed at which the bus operates. Specifically, it is possible for a station to sense the line is clear only because another station's transmission has not yet reached the listening source. The use of gigabit rates on the bus aggravates that problem, shrinking the network diameter to unreasonably small sizes. The slot time corresponding to the duration that a station uses to transmit a frame has thus been increased to offset the problem of the incredible shrinking network.
Another minor modification adjusts the operation of this new slot time under the condition of small frames. The minimum Ethernet frame size of 64 bytes would account for an unfortunately small percentage of the larger 512-byte slot time on a Gigabit Ethernet link, and the remainder of the bytes would simply become padding. The inefficiency in this case would be an unacceptable degradation of the Gigabit Ethernet capacity for small frames aggregated onto the link. Instead, a feature called burst mode allows a continuous transmission of multiple smaller frames.
Upgrade Examples
An obvious example of adding network speed in a place that would result in capacity enhancements would be through upgrading a backbone switch from a Fast Ethernet (100Base-T) device, shown in Figure 2, to Gigabit Ethernet equipment. In the most straightforward scenario, the upgraded network now can support more segments and associated bandwidth. In addition, or as another example, servers that are outfitted with Gigabit Ethernet capability can communicate with the Gigabit Ethernet switch, opening up the opportunity to support premium bandwidth services (Figure 3).
Thus, the upgrade can be considered an avenue for increasing the quantity of the same type of traffic and services or improving the service offerings themselves. A focus around increasing existing traffic as the theme can use Gigabit Ethernet connectivity as a natural upgrade for switch-to-switch links. That lets the Gigabit Ethernet switches support more segments, more average bandwidth between segments and increased traffic-handling capability.
The impact of Gigabit Ethernet has barely been felt. Some observers have it as a necessary desktop technology by the year. . . well, take your pick. My guess says it will at least get there before ATM.
Related Articles
- "Manning Up For 10-Gigabit Ethernet"; htttp://www.commsdesign.com/story/OEG20010323S0081
- "Uncovering the Challenges of Delivering GigE over CAT-5"; www.commsdesign.com/design_corner/OEG20020822S0002
References
- Breyer, Robert and Sean Riley, Switched, Fast, and Gigabit Ethernet, New Riders Publishing, 1999.
- "Gigabit Ethernet: Accelerating the Standard for Speed," Gigabit Ethernet Alliance white paper, 1998.
Rob Howald (rhowald@gi.com) is the director of systems engineering in the transmission network systems group of Motorola's Broadband Communications Sector in Horsham, Pa. He has a BSEE and an MSEE from Villanova University and a PhD from Drexel University.
