Linda Boyer
During the past three years as an industry standard, Fast Ethernet has become the most popular high-speed networking technology for backbone connections. Recently, price wars between Fast Ethernet vendors have expanded Fast Ethernet's realm to include desktop connections as well. Not surprisingly, network administrators are anticipating the need for even more speed.
Gigabit Ethernet has rushed in to meet that need. As the name implies, Gigabit Ethernet transfers data at a rate of 1 Gbit/s--one hundred times as fast as Ethernet and ten times as fast as Fast Ethernet. Gigabit Ethernet supports the same frames, the same management objects, the same Carrier Sense Multiple Access/Collision Detection (CSMA/CD) protocol, and the same full-duplex mode of operation that the Ethernet and Fast Ethernet standards support. Consequently, Gigabit Ethernet promises a smooth path to higher speed. (See "Three Easy Paths to Gigabit Ethernet.")
In short, Gigabit Ethernet is fast, efficient, and easy to implement. So what's the proverbial catch? Because Gigabit Ethernet is a relatively new networking technology, Gigabit Ethernet products are a bit pricey at the moment and, for now, are intended for use only with fiber-optic cable or short-haul copper cable. In addition, if you want Gigabit Ethernet to be an approved standard before you implement Gigabit Ethernet products, you will have to wait a few more months.
This article provides an in-depth look at Gigabit Ethernet to help you decide whether or not Gigabit Ethernet is your pipe dream come true. Even if it is, when will you be prepared to pay the price for higher speed--and where will you use it?
If all goes well, the Institute for Electronics and Electrical Engineers (IEEE) will formally accept the 802.3z supplement to the Ethernet 802.3 standard in June. 802.3z is one of two Gigabit Ethernet specifications. The IEEE expects to finalize the other specification, 802.3ab, by the end of this year or the beginning of next year.
Both 802.3z and 802.3ab define the parameters for transferring data at a rate of 1 Gbit/s. However, these specifications define the parameters for different physical media. The 802.3z specification defines the 1000Base-X family of proposed standards for running Gigabit Ethernet over fiber-optic cable and short-haul copper cable. When completed, the 802.3ab specification will define the 1000Base-T standard for running Gigabit Ethernet over Category 5 unshielded twisted pair (UTP) cable.
1000Base-X is the term for all of the proposed Gigabit Ethernet standards that are based on the 8B/10B signal encoding scheme. Developed by IBM Corp., 8B/10B enables a high data transmission rate. The 802.3z Task Force adopted 8B/10B and then adapted it to ensure a data delivery rate of 1 Gbit/s.
When you encode data using 8B/10B, the data delivery rate is 8/10 of the data transmission rate. For example, Fibre Channel, which is a high-speed networking technology standardized by the American National Standards Institute (ANSI), is based on the 8B/10B signal encoding scheme. Fibre Channel transmits data at a rate of 1.0625 Gbit/s but delivers data at a rate of only 850 Mbit/s.
Since the 802.3z Task Force adapted 8B/10B, Gigabit Ethernet's data delivery rate is higher than Fibre Channel's data delivery rate. The 1000Base-X family of standards transmits data at a rate of 1.25 Gbit/s in order to deliver data at a rate of 1 Gbit/s. The 802.3z Task Force increased the 8B/10B data transmission rate to get a data delivery rate of 1 Gbit/s, which is necessary to earn the name Gigabit Ethernet.
The 1000Base-X family of standards includes the following Gigabit Ethernet standards:
The 1000Base-SX and 1000Base-LX standards define the parameters for Gigabit Ethernet transmissions over fiber-optic cable. 1000Base-SX provides specifications for short-wavelength (850 nanometers) laser transmitters operating on multimode fiber. 1000Base-SX supports distances of up to 525 meters, depending on whether you use 62.5-micron or 50-micron diameter fiber. (See "Multimode and Singlemode Fiber.") 1000Base-SX also supports the same serial card (SC) connectors that you use with 100Base-FX.
1000Base-LX provides specifications for long-wavelength (1,300 nanometers) laser transmitters operating on either multimode or singlemode fiber. 1000Base-LX supports distances of up to 550 meters on multimode fiber and up to 3 kilometers on singlemode fiber. (See "Multimode and Singlemode Fiber.") Like 1000Base-SX, 1000Base-LX requires SC connectors for terminating the fiber-optic cable.
The distances discussed in this article for 1000Base-SX and 1000Base-LX transmissions over multimode fiber are based on the specifications available at the time the article was written. These distances could change, depending on the outcome of current testing.
The 802.3z Task Force continues to test Gigabit Ethernet transmissions over multimode fiber to ensure that the differential mode delay (DMD) problem is resolved. DMD, which the task force discovered last fall, occurs with particular combinations of lasers and multimode fiber. (DMD doesn't occur over singlemode fiber.)
Basically, the DMD problem stems from the fact that multimode fiber was designed for light-emitting devices (LEDs) that spread light equally through all of the fiber's modes. In contrast, lasers send concentrated signals. In a DMD scenario, a laser might concentrate the signal in only one or a few modes of fiber. Either way, by the time the originally strong signal reaches the receiver, this signal has diminished into an unrecognizable blip.
The Modal Bandwidth Investigation (MBI) subgroup of the 802.3z Task Force has potentially resolved the DMD problem by requiring all 1000Base-SX and 1000Base-LX transceivers to create a conditioned launch. A transceiver that conditions the launch of its laser signal spreads the signal evenly across all of the fiber's modes. (For more information about DMD, see "The DMDelay".)
The 802.3z Task Force continues to test the effects of DMD and the conditioned launch solution, which is now included in the 802.3z specification. In fact, DMD (and the extensive testing it has given rise to) caused a delay in a schedule to which the task force had otherwise tightly adhered. The taskforce had originally planned to finalize the 802.3z specification last month and, until recently, appeared to be on track.
Early this year, the 802.3z Task Force announced its plans to continue testing Gigabit Ethernet transmissions over multimode fiber using the conditioned launch solution to ensure that the DMD problem had been resolved. At that time, the task force also announced that it would be unable to finalize the 802.3z specification until the next IEEE Standards Board meeting, which will be held in June.
If the 802.3z Task Force is unable to meet this deadline, the task force will have to wait until the IEEE Standards Board meeting in September. By then, the task force is certain that the 802.3z specification will be approved--finally.
The 1000Base-LX and 1000Base-SX standards are well-suited for server-to-switch and server-to-server connections spanning relatively long distances. However, you probably don't want to use fiber-optic cable to connect equipment that might be only a few feet apart. In this case, copper cable is a better choice.
Accordingly, the 802.3z Task Force has included one proposed standard for copper cable in the 802.3z specification: 1000Base-CX. 1000Base-CX defines parameters for Gigabit Ethernet transmissions for distances of up to 25 meters over a new type of shielded copper cable, which contains balanced wire pairs and is rated at 150 ohm. This cable, called twinax cable, may sound similar to the IBM Type 1 shielded twisted-pair (STP) cable, but STP does not meet the 150 ohm requirement.
You can terminate twinax cable at each end of the cable using a 9-pin D connector. Because Gigabit Ethernet products use receptacle-type, or female, connectors, you need plug-type, or male, connectors to terminate twinax cable. 1000Base-CX also supports a high-speed serial card (HSSC) connector, which is an 8-pin Fibre Channel Type 2 connector.
In choosing where to deploy 1000Base-CX, you should be aware of one key requirement: With 1000Base-CX, the grounding and electrical system must be the same at both ends of the link. That is, you must ensure that all connected computers have a common electrical feed.
The other proposed Gigabit Ethernet standard for copper cable is 1000Base-T, which the 802.3ab specification defines. When completed, the 802.3ab specification will define the parameters for Gigabit Ethernet transmissions over Category 5 UTP cable for links of up to 100 meters. The 802.3ab Task Force is hoping to develop the 1000Base-T standard to support the same RJ-45 connectors currently used to terminate copper cable on some Ethernet and Fast Ethernet networks.
The 802.3ab Task Force hopes to finalize the 802.3ab specification by November 1998. However, some people believe that this goal is overly optimistic. Developing the 802.3ab specification is inherently more difficult than developing the 802.3z specification, says Brian MacLeod, director of Marketing at Packet Engines Inc., a Gigabit Ethernet vendor. MacLeod explains that the 802.3z Task Force had an existing signal encoding scheme (8B/10B) and optical transceivers that used the scheme, functioning at a data-delivery rate near 1 Gbit/s. The task force simply adopted and adapted the signal encoding system for Gigabit Ethernet.
In contrast, no signal encoding schemes exist for Category 5 UTP cable functioning at gigabit speeds or even close to gigabit speeds. The IEEE 802.3ab Task Force is developing the 802.3ab specification from scratch. In light of this level of difficulty, MacLeod and other Gigabit Ethernet vendors do not expect the 802.3ab specification to be finalized until spring 1999. However, Gigabit Ethernet products based on the emerging 802.3ab specification should be available later this year.
In many respects, Gigabit Ethernet (as defined in the 802.3z and 802.3ab specifications) is Ethernet, only faster. If that sounds familiar, it should. About 18 months ago, we told you the same thing about Fast Ethernet. In fact, all Ethernet standards support the following:
Like Ethernet and Fast Ethernet, Gigabit Ethernet uses the variable-length 802.3 frame format, which can vary between 64 bytes and 1,518 bytes. (See Figure 1.) Because Gigabit Ethernet supports the traditional Ethernet frame format and size, you can connect existing lower-speed devices using switches or routers that simply adapt one line speed to another line speed.
Simple Network Management Protocol (SNMP) management information bases (MIBs) can track variables that measure performance and errors on Gigabit Ethernet systems, just as MIBs can do on Ethernet and Fast Ethernet systems. As a result, comparing network segments operating at different speeds is simple, and network support personnel will require little, if any, training to make these comparisons.
For half-duplex mode, Gigabit Ethernet supports essentially the same CSMA/CD protocol that both the Ethernet and Fast Ethernet standards support. However, the 802.3z Task Force enhanced this protocol to maintain a 200 meter network diameter at gigabit speeds.
Basically, the 802.3z Task Force enhanced the CSMA/CD protocol by changing the carrier event time (that is, the minimum time a transmitting station must occupy the wire) from the 512 bits that the Ethernet and Fast Ethernet standards specify to 512 bytes. This enhancement requires the use of a new feature called carrier extension. A carrier extension is a nondata signal that devices add to the data fields of each Gigabit Ethernet frame that is less than 512 bytes. (See Figure 2.)
Unfortunately, a carrier extension takes a toll on performance. To compensate for this performance hit, the 802.3z Task Force also developed an option called packet bursting. Packet bursting enables stations to transmit more than one packet of less than 512 bytes in one transmission event. (For more information about carrier extensions and packet bursting, see "CSMA/CD Gigabit Style.")
The irony is that despite the IEEE's efforts to preserve and support the CSMA/CD protocol, Gigabit Ethernet vendors do not believe that customers will use Gigabit Ethernet in half-duplex mode. As a result, all of the Gigabit Ethernet products available now support only full-duplex mode. None of the Gigabit Ethernet vendors I spoke with even have plans to develop products that support half-duplex mode.
"We don't believe that half-duplex products are a viable alternative," says Nathan Walker, a Gigabit Ethernet product manager at Cisco Systems Inc. Walker's comment represents the pervasive attitude among Gigabit Ethernet vendors. In fact, Bob Gohn, Gigabit Ethernet program manager at 3Com Corp., says he "knows of no one implementing half-duplex mode in their chips." Gohn adds, "Full duplex is the preferred mode of operation."
If no one is planning to use half-duplex mode, why did the 802.3z Task Force develop half-duplex specifications? "Good question," Walker responds, but he's only partly serious. The task force had several reasons to develop half-duplex specifications for Gigabit Ethernet.
One reason is that the timing wasn't right to develop a new Ethernet standard that would support only full-duplex mode. The Gigabit Ethernet Task Force (which is a combination of the 802.3z Task Force and the 802.3ab Task Force) began working on the Gigabit Ethernet standard in fall 1995, two years before the 802.3x specification for full-duplex mode was approved. At that time, suggesting a new Ethernet standard that operated exclusively in full-duplex mode would have been shot down immediately: The risk was too high without an approved full-duplex specification.
Furthermore, the Gigabit Ethernet Task Force had a responsibility to create a Gigabit Ethernet standard that was broad enough to accommodate every possible implementation of a new Ethernet technology while preserving Ethernet's basic characteristics. "The standards group wanted to ensure that if companies were interested in pursuing a shared alternative at some point in time, the specifications for doing so were established early on," Walker explains.
Not only that, but vendors--not standards groups--must decide which specifications best suit their customers. "Once the specification is done," Gohn says, "vendors look at the usefulness of different parts and say, 'Some of these things make sense, and some of them don't.' " In the case of Gigabit Ethernet, Gohn adds, "we think that [full-duplex products] are more commercially viable and make more sense for our customers."
On the other hand, Gohn suggests, if someone could develop half-duplex products for one-third the cost of full-duplex products, half-duplex products would certainly appear on the market. Instead, Gigabit Ethernet products that support half-duplex mode might actually be more costly to develop than products that support only full-duplex mode.
Developing a half-duplex chip requires support for the carrier extension feature and for the packet bursting option. Thus, developing a half-duplex chip is inherently more complex than developing a full-duplex chip. Additional complexity equates to additional expense, and spending more money to develop products that offer less performance than products that are already available simply doesn't make sense.
Finally, if the Gigabit Ethernet Task Force had omitted support for the CSMA/ CD protocol entirely, the Gigabit Ethernet standard might not have been considered an Ethernet standard at all. The task force maintained CSMA/CD support, in part, to ensure that Gigabit Ethernet remained safely under the Ethernet umbrella, which covers a sizable portion of the market.
In 1996, more than 80 percent of all existing network connections were Ethernet connections. (See Charles E. Spurgeon, Practical Networking With Ethernet, International Thomson Computer Press: Boston, 1997, p. 2.) In fact, recent statistics from Dataquest show that Ethernet technologies outship all other networking technologies by an 11-to-1 ratio, which is expected to rise through 2001. Because this huge installed base of Ethernet customers perceives Gigabit Ethernet as an enhanced version of Ethernet--rather than as an entirely new technology--chances are high that the Gigabit Ethernet standard, and the products based on that standard, will do well.
The proposed 1000Base-X and 1000Base-T standards support the 802.3x specification for full-duplex mode and flow-control methods. Other Ethernet and Fast Ethernet standards, including 10Base-T, 10Base-FL, 100Base-TX, 100Base-FX, and 100Base-T2, also support the 802.3x specification. However, the Ethernet and Fast Ethernet products based on these standards are typically half-duplex products that support a full-duplex option. In contrast, Gigabit Ethernet products are full-duplex products--period.
When both stations on a link are enabled for full-duplex mode, these stations can simultaneously send and receive frames, essentially doubling the link's capacity. In the case of Gigabit Ethernet, a full-duplex link has a capacity of approximately 2 Gbit/s.
Stations in a full-duplex system are interconnected via point-to-point links and do not share the wire. Hence, Gigabit Ethernet products (which, at the risk of being redundant, are all full-duplex products) do not use the CSMA/CD protocol or, therefore, the carrier extension feature or the packet bursting option because there is no risk of collisions on the link.
A Gigabit Ethernet full-duplex link uses the flow-control methods defined in the 802.3x specification. These flow-control methods help a full-duplex system accommodate extreme traffic conditions. Essentially, the flow-control methods enable a receiving station overloaded with traffic to send a command to the transmitting station, requesting that the station stop transmitting for a period of time.
Other high-speed protocols do not support flow-control methods. For example, Asynchronous Transfer Mode (ATM), as Gigabit Ethernet vendors like to point out, does not have any established flow-control methods to use.
As mentioned earlier, a full-duplex system experiences no collisions on the links. Without collisions and with flow-control methods, Gigabit Ethernet handles high traffic well. In fact, says MacLeod, with a full-duplex system "when you offer a load of 100 percent, the network could deliver a load of 100 percent." The bottom line is this: You can expect Gigabit Ethernet to handle heavy loads far better than shared Ethernet and Fast Ethernet segments. (For more information about Gigabit Ethernet, refer to the resources listed in "Interested in Learning More?".)
Although it's far too early to know just how well Gigabit Ethernet products will fare, one thing is certain: You can choose from a variety of products from a number of vendors.
Gigabit Ethernet vendors claim that all of these products are interoperable, and a recent demonstration supports this claim. Last year at NetWorld+Interop '97 in Atlanta, Georgia, the Gigabit Ethernet Alliance hosted the largest multivendor Gigabit Ethernet interoperability demonstration. This alliance of more than 120 vendors is dedicated to promoting the standardization of Gigabit Ethernet.
Members of the Gigabit Ethernet Alliance that participated in the demonstration include 3Com, Cisco Systems, Packet Engines, Bay Networks Inc., and Intel Corp. These vendors demonstrated the gamut of Gigabit Ethernet products that are available, such as network interface boards, router interfaces, switches, uplinks for Ethernet and Fast Ethernet switches, and one new product called a buffered distributor.
A buffered distributor is also called a full-duplex repeater, which may sound like a contradiction in terms. After all, an Ethernet repeater forwards all incoming traffic to all connected stations on a shared medium. But in a full-duplex system, the medium is not shared, so how can there be such a thing as a full-duplex repeater?
In fact, a buffered distributor is a little like a repeater (in terms of function) and a little like a switch (without the cost). Like an Ethernet repeater, a buffered distributor links two or more Gigabit Ethernet segments, forwarding all incoming traffic, without filtering addresses, to all connected links (except the originating link).
Unlike an Ethernet repeater, however, and more like a switch, the buffered distributor has memory on each port, and every port operates in full-duplex mode, allowing long full-duplex links to stations. Buffered distributors, Gohn says, "just run much more efficiently, much more effectively, and . . . are more elegant solutions than shared hubs." And because buffered distributors don't have to do any sophisticated address filtering, they are also less expensive than switches, the traditional, high-performance alternative to hubs.
How much is "less expensive"? Packet Engines' FDR12 Gigabit Ethernet Full-Duplex Repeater has a suggested retail price of approximately U.S. $1,000 per Gigabit Ethernet port. A switch, on the other hand, can range anywhere from U.S. $2,000 to U.S. $4,000 per Gigabit Ethernet port. For example, 3Com's 8-port Gigabit Ethernet switch, the SuperStack II Switch 9000 SX, costs nearly U.S. $2,500 per port (for a grand total of U.S. $19,995).
You can also buy Gigabit Ethernet modules for existing switches if you want a comparatively inexpensive, switched path to Gigabit Ethernet. For example, 3Com offers a Gigabit Ethernet module for its SuperStack II Switch 3000, 1000, or Desktop switches. The SuperStack II Switch Gigabit Ethernet SX Module from 3Com costs approximately U.S. $2,995. If you already have a couple of linked switches from 3Com, you could upgrade the 3Com switch-to-switch link from 100 Mbit/s to 1000 Mbit/s simply by installing this Gigabit Ethernet module on each switch.
Cisco plans to officially release Gigabit Ethernet modules for its Catalyst 5000 switches shortly after the 802.3z specification is finalized. However, Cisco had not announced the cost of these modules at the time this article was written.
If you are considering upgrading a switch-to-server connection, you should know that Gigabit Ethernet network interface boards are pretty costly right now, and not all of these network interface boards support NetWare. For example, 3Com's EtherLink Server network interface board costs U.S. $1,695 and supports only Windows NT. On the other hand, Packet Engines's G-NIC is available for only U.S. $995 and includes Novell-certified LAN drivers.
If the prices for Gigabit Ethernet products seem high, remember that when Fast Ethernet products first hit the market, prices for these products seemed high, too. The good news is the Gigabit Ethernet Alliance believes that prices for Gigabit Ethernet products will drop at about the same rate that prices for Fast Ethernet products dropped.
And what rate is that? According to one statistic from the Dell'Oro Group (as shown in the Gigabit Ethernet Alliance White Paper), a Fast Ethernet switch cost approximately U.S. $785 per port in 1996. This year, that price has dropped to an average of U.S. $500 per port--a 36 percent decrease in cost in only two years. If Gigabit Ethernet products follow a similar trend, a Gigabit Ethernet switch that costs U.S. $3,000 per port this year will cost less than U.S. $2,000 per port in two years time.
For now, however, prices are a bit steep. At what point should you consider upgrading to Gigabit Ethernet? The answer to that question depends on your company's network, of course, and on who you ask. Naturally, most Gigabit Ethernet vendors side with MacLeod, who says "when you need more than 100 Mbit/s, you definitely need Gigabit Ethernet."
And where might you need more than 100 Mbit/s? Most likely on links between switches, links to high-performance servers, and backbone links. Not surprisingly, the Gigabit Ethernet Alliance believes that initial implementations of Gigabit Ethernet will be on switch-to-switch, switch-to-server, and backbone links. (See "Three Easy Paths to Gigabit Ethernet".)
The Gigabit Ethernet Alliance also estimates that widespread implementation of Gigabit Ethernet to the desktop is still three or four years away. The reason is that the standard for Category 5 UTP, a more common desktop connectivity medium, has not yet been approved. In addition, a less pressing need for speed exists at the desktop.
On the other hand, you might decide that if you're going to upgrade your company's network, you should be consistent and upgrade the entire network infrastructure to Gigabit Ethernet. That's what Novell is doing. Novell thinks that the time to move to Gigabit Ethernet is now--and Novell's taking Gigabit Ethernet all the way to the desktop.
Novell is currently rewiring its Provo facility to upgrade the decade-old 10 Mbit/s Ethernet network to Gigabit Ethernet. Novell uses fiber optic cabling on the backbone and to individual labs but uses Category 5 UTP for desktop connections. Naturally, Novell will begin the Gigabit Ethernet upgrade process at the backbone and work gradually toward desktop connections.
But if you think "gradually" suggests "slowly," think again. In fact, according to Glenn Ricart, Novell's chief technology officer, Novell expects to have 15,000 Gigabit Ethernet connections by the end of this year. "Novell is one of the leaders in installing Gigabit Ethernet," Ricart explains, "but my experience is that we're probably not more than two to four years ahead of the mainstream industry."
Novell's Orem facility, where Ricart works, has been wired with Fast Ethernet to the desktop for the past two years. However, Novell is shifting its energies to the Provo facility, where a new building, Building G, is currently under construction. "In two years time," Ricart says, "I'm going to trade in my 100 Mbit/s connection here for a Gigabit Ethernet connection in Building G." (For more information about Glenn Ricart's view of Gigabit Ethernet, see the related article.)
More than 85 percent of Novell's 79 million users are connected to an intraNetWare or NetWare network via Ethernet. When will these users, like Glenn Ricart, trade in their 10 Mbit/s or 100 Mbit/s connection for a Gigabit Ethernet connection? Who knows. But what Ricart knows, or strongly suspects, is that you'll start trading in your company's backbone connection first. "To me, it's a no brainer," says Ricart. "If you're putting in a new backbone on a new network, you'll want to use the highest-speed networking technology available."
It seems both fitting and unlikely that after thirteen years as an industry standard, Ethernet continues not only to survive but to thrive. Who would have guessed the speeds that would evolve from the system Bob Metcalf first discovered and named "Ethernet" in 1973? Did anyone suspect that this technology could ever offer one-hundred times the speed of the 802.3 standard approved in 1985? And even in 1995, when the IEEE pushed the envelope on 10 Mbit/s networking technology by offering 100 Mbit/s, would many users have bet that the IEEE could successfully push the envelope yet again and increase the speed of Fast Ethernet tenfold?
Probably not. But Gigabit Ethernet is here--now what are you going to do about it?
Linda Boyer works for Niche Associates, which specializes in technical writing.
NetWare Connection, April 1998, pp. 10-23