The rise of Gigabit Ethernet
How did it all begin -- and where are we now?
Gigabit Ethernet is making headlines even before it has become a blessed standard. This emerging LAN networking system provides a good solution for future LAN servers. This month we take a look at how Gigabit Ethernet is evolving and the vendors that are bringing it to market. (2,400 words)
ou have undoubtedly heard that Gigabit Ethernet is coming alive. At one time it was thought that 100 Mbps was impossible over the standard Ethernet Category 5 twisted-pair wiring. Now we not only have 100-Mbps LAN switches with each port capable of supporting an independent 100-Mbps segment and up to 96 ports in some cases, we have gigabit switches going beyond the traditional hub and simple switch arena into the land of ATM switches.
We've certainly come a long way from the original project by Bob Metcalfe, et al, to develop a simple communications media for desktop computers in 1982. Originally 10Base5 coax-based 10-Mbps Ethernet ruled the campus backbones of companies and universities. The word "Base," as you may know, indicates that this is a baseband or narrowband signaling system using a specific frequency or small frequency range for all communications as opposed to other broadband systems like cable TV which help multiple ranges of frequencies. Soon LANs were reaching that speed; and they created a simpler and cheaper coax-based system out of RG-58 cable in 10Base2.
Finally someone hit on the idea of using the same kind of wiring that was traditionally used for telephones all over the world instead of special coax cabling. This made Ethernet cheaper, more flexible, and available in the LAN environment. The concept of a central hub and a star pattern for the distribution of the network also made more sense because the number of points of failure was radically reduced; rather than the possible cut in the coax of 10Base5 and 10Base2 which could disable the entire network, the problem was now localized to either the hub or a specific segment to the station.
In 1986, 10BaseT (T for twisted pair) became the new name for this technology, and it spanned almost the same distance as 10Base2 while providing new benefits. Then the availability of fiber cabling brought another new change to the infrastructure. The distance limitations of all the copper-based Ethernet systems was becoming apparent as corporations became larger and the number of computing systems increased. 10BaseF brought the benefits of increased reliability and distance permutation. It never caught on too well but nevertheless has been implemented successfully worldwide.
The next step was to take the whole concept to a new level with the development of 100BaseT. Initially, competition resulted in several different standards (100Base-VG, 100Base-TX, etc.). Eventually it settled back down to just an improved version of 10BaseT. The new standard had stricter requirements for the cabling and the electronics that drove the communications. Category 5 wire became the recommended standard -- although Category 3, a lower-quality wire, worked at shorter distances. The problem lies in the fact that at higher data-transmission rates, the signal can deteriorate in shorter distances.
100BaseT was to take over the almost decade-old 10BaseT system, but to this day the number of hubs sold of the predecessor still far exceeds the newer technology. The cost of 100BaseT systems has fallen significantly since introduction, especially compared to what 10BaseT cost five years ago. The transition continues as vendors combine their current Ethernet network interface card products to support both standards. Now a new development is coming into focus.
A thousand little bytes
In its simplest form, the concept isn't too hard to imagine. Take traditional 10-Mbps or 100-Mbps Ethernet. Shorten the distance that it has to go, and you may just be able to attain a much higher speed. Gigabit Ethernet started out in that direction; it actually isn't gigabit since it ranges around 1000 Mbps rather than 1024 Mbps -- but who's counting. It did not go through the same growing pains that 100BaseT did (with the attempted divergence to 100Base-VG, etc.), but it has not survived without some scratches. Its main competitor now is not diverging standards but a completely different form of network -- the asynchronous transfer mode (ATM) network.
At this point the Institute of Electrical and Electronic Engineers (IEEE) 802.3z Gigabit Ethernet standard is at the Draft 2 (D2) (released in March 1997) stage with a final promise slated for early 1998. The delay is to allow vendors to work out any last minute bugs, "features," and general incompatibilities that may arise. With a lot of help from the Gigabit Ethernet Alliance and the cooperation of many vendors, it has kept to a mostly straight path from start to finish.
Gigabit Ethernet follows the same communications principle as its forbears; it supports carrier sense multiple access with collision detection (CSMA/CD). This, in English, means that it is a shared network where multiple stations all receive the same signal and separate out the data directed only for them through addressing. Additionally, the hardware can detect if there is already a signal on the line to avoid collision or can simply try again when there is no collision. This technology is half duplex simply by design, meaning that you can only either send or receive at one time; you can, however, make it full duplex and allow machines to send and receive at the same time using either multiple wires or some clever methods. Most vendors are implementing it using a switched hub, and only one device is typically on the port.
The specification allows for copper and fiber standards. The copper version, known as 1000Base-CX is limited to a maximum of 25 feet and requires 150-ohm shielded twisted-pair (STP) copper at a serial line rate of 1.25 Gbps over a special cable known as Twinax; IBM mainframe shops may be familiar with it. Most Ethernet products we use worldwide are based on unshielded twisted pair (UTP), however, for the sake of simplicity and reduced cost.
The IEEE committee recognizes this fact and is pursuing this goal as well. The simple fact is that at gigabit speeds STP Twinax provides much better data integrity from electromagnetic interference and moving to UTP might reduce the distance even further. To counter that, 1000BaseT, as the traditional UTP copper system is labeled, cannot function very well even theoretically over the old two-pair wires. The IEEE 802.3ab committee was formed separately to try and solve this with four-pair UTP supporting distances of 100 to 200 meters.
With fiber-based Gigabit Ethernet things gets more complicated. There are two main types of fiber used by most systems in the world: single mode fiber (six to 10 microns or micrometers thick) and multimode fiber (62.5 to 125 micrometers thick). The difference between the two is that single-mode fiber can go much longer distances than multimode fiber; in the ATM world it is 20 km versus 2 km. Combined with this difference is the fact that you can use different frequencies of light for communications; it's a tradeoff between frequency chosen, the cost of the equipment needed to transmit at the given range, and distance supported.
Instead of recreating the wheel, they took a page from Fibre Channel technology. Fibre Channel is already used now as a high-speed substitute for SCSI drives supporting gigabit access speeds needed for video editing and data warehouse applications. The fiber-based Gigabit Ethernet systems will use the same physical signaling system as Fibre Channel to communicate at 850-nanometer light, called 1000Base-SX for short wavelength light, providing a distance of 300 meters. A second system using 1300-nanometer light, called 1000Base-LX for long wavelength, will support up to 550 meters. Single-mode standards will probably come out as well, supporting even greater distances.
How does it perform?
So the question will definitely come up: how fast is it? To the novice, it is easy to assume that because it says 1000 Mbps, that is truly what you get. Unfortunately, that has never been true with Ethernet. Leaving the consideration that it is a shared medium and that the actual speed is divided across all the systems aside, Ethernet performs somewhere between 60 to 90 percent of the rated speed. According to the Gigabit Ethernet Alliance, tests with chipsets and hardware from AMD have indicated that it is possible to push 720 Mbps in half-duplex Gigabit Ethernet from one machine to another. Whatever vendors may claim, Gigabit Ethernet suffers from the same fate of all Ethernet products: it doesn't really live up to its name. Nevertheless, 720 Mbps is quite a huge amount of bandwidth.
Product interoperability at the D2 level is promising. Although switches and hubs can function well, it can cause a problem when it comes to network routing. When you are pushing so much data across the network, your hub may just outpace your current routers. The bottleneck comes into effect when you have such kinds of switches in combination with older, less powerful routers. Even today's best routers have difficulty routing packets at several hundred megabits per second. Of course, a new generation of routers and switches are on the market that can work with 622-Mbps ATM interfaces, and some prominent vendors are hoping to put in Gigabit Ethernet interfaces as well.
Will servers actually push that much traffic individually? In many cases they will. CAD stations and scientific applications can easily push hundreds of megabytes of data per image. Large network backups could be done in a shorter time, and group videoconferencing or distribution could become a reality. It's possible that a Web site could push that much data, but it's usually more of a case of routing than it is of bandwidth.
And it's not just a matter of application. Most PC systems today cannot support such sustained speeds. You will have to wait for 64-bit PCI buses that run at higher speeds (like 66 MHz or so) to be able to push that much data. Furthermore, even the best Ultra-SCSI systems will not keep up; the Fibre Channel disk subsystems from Sun, Compaq, and other vendors can. Larger Unix server systems will fit more into this category with higher-speed buses such as the ones from Digital, Sun, and Silicon Graphics.
You can be sure, however, that this hole will be filled. New servers and workstations will eventually fill this gap -- probably by the time the standard is official and products are out. In other words, now is still not yet the time to buy the equipment that can utilize this technology.
The vendors can't wait. Not only are the traditional hub and switch vendors brimming with excitement over their pre-standard products, entirely new companies have placed their bets on it with products of their own. The list is lengthy with almost 30 vendors at the starting line -- prominent names like 3Com, Cisco, IBM and Hewlett-Packard. The newbies betting the farm on this technology include Packet Engines, Extreme Networks, and Foundry Networks. The Gigabit Ethernet Alliance has 11 members including hub and switch makers, network interface card and chip vendors, and server hardware manufacturers. Many of these same vendors will continue to sell ATM products as well. This is Gigabit Ethernet's prime competitor. The market is just a little different for each, but there is still much overlap, leading customers to argue whether they need one or the other.
Whenever someone tells me that Gigabit Ethernet will thrash ATM, I chuckle quietly. It certainly has a lot of advantages behind it; it is simpler and more familiar; it may be cheaper eventually; it does not require reeducation in engineering techniques; and it may not require new protocol stacks. However, the fact is that ATM is only now starting up and even at the most basic level it can scale up to higher speeds, whereas Gigabit Ethernet is a dead end. It's very likely Gigabit Ethernet will require as much recabling as ATM. And although the industry is working on bringing quality of service requirements to Gigabit Ethernet, this is more like taking a Geo Metro, forcing a dragster's super engine into it, and thinking that you will get the precision control of a Ferrari.
Gigabit Ethernet provides a solution for many LAN situations and will probably be much more successful than ATM in that environment. It takes the flexibility of ATM but provides the bandwidth and possibly some of the quality of service benefits in a familiar technology. At $2,500 to $3,500 a port, it is an expensive upgrade for simple applications such as departmental servers; it is something targeted at the corporate data warehouse and heavily used corporate intranet servers. One thing's for sure, Gigabit Ethernet interfaces are much simpler, so it is more likely that the prices will go down on the supporting chipsets faster than it will for ATM.
By the middle of 1998, you will have a selection of Gigabit Ethernet products to choose from. It is predicted that this industry will hit the $1 billion revenue mark by the year 2000. Pre-standard products are already available, but I would advise against picking them up unless your vendor can prove interoperability and show an upgrade plan when the official standard arrives. You will very likely see a mix of Gigabit Ethernet and older Fast Ethernet within the same product. After all it is more likely that you simply won't need gigabit speeds for everything, and this saves you from buying multiple devices to solve the problem.
Whatever the case may be, Gigabit Ethernet is following in the footsteps of its predecessors. It will probably never sell as many ports as 10BaseT before it becomes obsolete, but Gigabit Ethernet will provide high-speed server solutions into the next millennium.
About the author
Rawn Shah is vice president of RTD Systems & Networking, Inc. He has worked with many different aspects of the LAN world and is currently strongly investigating the world of ATM and DSL and their implementation and implication on the future of voice, video, Internet, and data networking. Reach Rawn at firstname.lastname@example.org.
If you have technical problems with this magazine, contact email@example.com