The computer industry is a sector filled with superlatives. No sooner has one technology been released than another company sets out to release something smaller and faster. Networking technology is no different in this respect. When Ethernet first came into existence as a slow networking standard in the Xerox Parc laboratories in 1973, it was hailed as a revolution in connectivity. Its 2Mbps throughput was perceived to be blindingly fast, and yet, as we have seen, even 10Mbps network speeds can be woefully inadequate these days when it comes to coping with modern data communications requirements.
Fast Ethernet was developed in the early 1990s to remedy the problem.
Winning out over the rival VG anyLAN technology, it provided companies with the ability to use network interface cards (NICs) that could speak to each other at ten times the speed of conventional 10BaseT 10Mbps Ethernet networks. The important thing about the jump from 10Mbps Ethernet to Fast Ethernet is that it enabled companies to maintain their existing cabling infrastructures. Both standards worked over copper cable, meaning that companies without the resources to run new cabling throughout their buildings would find the technology particularly attractive as a means of increasing bandwidth.
One step at a time
The natural law of the computing market dictates that technological capability should increase exponentially. Consequently, having cracked the Fast Ethernet problem, the vendors turned their sights not on the goal of 200Mbps networking, but rather to increasing it by a factor of 10. Gigabit networking became the target for vendors and customers who had particularly demanding data communications requirements.
The Gigabit Ethernet Alliance was formed in May 1996 by eleven founding members, including such big names as 3Com, Cisco, Compaq, Intel, Sun and Bay Networks. Two months later, the IEEE 802.3z Gigabit Ethernet Task Force was created to develop a networking standard based on conventional Ethernet technology that would be capable of achieving the required data throughput. Draft specifications for the standard were released in January 1997, and in May that year the first technology demonstration took place at the NetWorld+Interop show in Las Vegas. October 1997 saw the first multivendor interoperability demonstration, and the standard was finally ratified in June last year.
The standard divides into three variants designed to support different types of physical media. The first, 1000BaseSX, is a short length fibre standard stretching from 220m to 550m. It is designed primarily to handle horizontal building connections, such as the linking of a single floor in a site to an inter-floor backbone. The second, 1000BaseLX is a long wavelength fibre standard which runs up to 5km using single mode fibre and is designed primarily to handle long-range connections such as those found in between buildings within a campus environment. Finally, 1000BaseCX is a copper standard designed purely for connecting components across short distances.
Although Gigabit Ethernet does share many similarities with its older, slower brothers, there are nevertheless technical differences in the latest standard that all network managers should be aware of, according to Geraint Evans, product marketing manager at Gigabit Ethernet equipment vendor Chernikeeff. In particular, although 1000BaseCX is a copper standard, the elements of the technology targeting medium- and long-range network connections are at present exclusively fibre-oriented. Both 10BaseT and its Fast Ethernet successor are based on copper cabling. Secondly, the standard is designed for point-to-point connections, says Evans, meaning that it is targeted at those companies that need to connect two entities such as a switch and a server together.
Finally, Evans believes that although the standard does comply with conventional Ethernet-based deletion detection functions, these will rarely be used.
Ethernet relies on the Carrier Sense Multiple Access/Collision Detect (CSMA/CD) collision detection protocol, which is based on a need for Ethernet packets to share the same line in a network. If a node such as a desktop client wishes to send a packet of data along an Ethernet network, it waits until the line is quiet and then tries to send the packet out. If another packet arrives on the line at the same instant that the node's packet is sent, then both nodes are cancelled and their sources wait for a random period of time before attempting to send their packets again.
CSMA/CD is only required in a half duplex network connection, because collisions will only occur in this situation. In a full duplex connection, when nodes can send and receive packets at the same time, it becomes irrelevant.
Evans uses this fact to conclude that CSMA/CD, although present in theory, will not be a big part of Gigabit Ethernet's future in practice. A White Paper produced by the Gigabit Ethernet Alliance appears to confirm this fact by pointing out that most Gigabit Ethernet products are designed to work in full duplex mode.
IDC analyst Pim Bildebeek highlights a key advantage of Gigabit Ethernet as its user-friendliness and ease of installation. Nevertheless, he doesn't pull any punches when it comes to pointing out the technological disadvantages of the fledgling networking standard.
One big drawback is that its immaturity means that one piece of Gigabit Ethernet equipment may not necessarily work with another, even though theoretically there should be no interoperability problems, he explains.
Pricing is still an issue too, he says, pointing out that Gigabit Ethernet costs up to $4,000 (#2,424) per port, depending on the equipment that you are purchasing. Gigabit Ethernet prices are often seen as more expensive than ATMs (Asychronous Transfer Mode), because Gigabit Ethernet switches are often very large, says Bildebeek. However, he concedes that if you drill down and compare equipment offering the same functions, then Gigabit Ethernet costs less, and prices are decreasing quickly anyway. Perhaps a more valid drawback for Gigabit Ethernet customers is that the choice of equipment is still relatively limited.
Although Gigabit Ethernet represents a promising boost for ailing corporate networks, its position in the high-bandwidth networking arena is not entirely unchallenged. Although the Gigabit Ethernet Alliance did its best to shorten the ratification cycle for Gigabit Ethernet, the consortium has been unable to turn back the clock. While it thrashed out the technical details of its high-bandwidth offering, ATM has been gathering strength since the start of the 1990s. ATM offers backbone speeds ranging from 155Mbps upwards, and although it is expensive, it does offer a range of advantages, according to Chris Blenkhorn, networking consultant at IBM Networking Services.
The need for speed
The first advantage is speed. ATM port speeds are increasing exponentially, says Blenkhorn, and it is already able to match and exceed the throughput offered by Gigabit Ethernet. This advantage is compounded by ATM's cell-based structure, which enables it to switch data much more quickly than Ethernet's frame-based equipment. ATM has also been designed specifically to handle data traffic with low latency requirements. This makes it particularly useful for the communication of multimedia signals such as voice and video.
Unfortunately, the quality of service technology that can be used with Gigabit Ethernet has not been defined as an inherent part of the standard.
This means that it rests above the Gigabit Ethernet protocol in the communications stack, and must be provided by vendors individually. Certain services and standards are being ratified, such as the IEEE 802.1 P. standard and the RSVP technology being explored by the Internet Engineering Task Force (IETF).
Nevertheless, it remains to be seen how quickly the vendors will implement them, and how faithful they will be to the standards as they appear. There is always a temptation to add proprietary bells and whistles to "enhance" standards for your own equipment.
Blenkhorn believes ATM to be in a better position than Gigabit Ethernet when it comes to backbone connectivity in large corporate organisations.
However, he does see a place for Gigabit Ethernet in smaller organisations that require a convenient method of migrating from their existing Ethernet installations to a higher-bandwidth form of networking.
Information on the Gigabit Ethernet Alliance's Web site highlights different situations in which companies would be most likely to upgrade from a conventional Ethernet solution to a Gigabit scenario.
The first involves upgrading a switch-to-switch connection, enabling switches to exchange information between each other at a much higher rate.
This would enable switches to support a greater number of 10BaseT or Fast Ethernet segments, but unless switch-to-server links were enhanced, these segments would still operate at a 10Mbps or 100Mbps bandwidth.
Consequently, the second scenario involves upgrading the link between the server and the switch, which the Alliance says will enable companies to achieve high-speed access to applications and file servers.
The Alliance says that another likely migration path would involve aggregating fast Ethernet switches with a Gigabit Ethernet switch or repeater. If a company currently uses a 100Mbps backbone connected to 10/100Mbps switches or routers, it would be possible to support 1Gbps links from the enhanced switches to the backbone. This enables the network to support more segments and more users per segment.
Upgrading backbone infrastructures may not be as easy for smaller companies as it is for large ones, however. Currently, the Gigabit Ethernet standard is designed for fibre use in long-range communications, while copper is reserved for shorter distances. A longer range copper standard will hopefully be ratified in March. Until this happens, any long-range copper Gigabit Ethernet switches on the market will be working under unratified standards, although this hasn't always stopped companies installing products in the past.
According to Martin Anstett, product marketing manager at Gigabit Ethernet vendor Syskonnect, this won't be a problem for those companies with a backbone placed in one room, because they will be easily able to connect to a central switch. If the backbone is spread throughout the building, however, fibre will have to be installed. Nevertheless, according to IBM's Blenkhorn the cost of pulling a small number of fibre runs throughout a building should not be prohibitive.
Upgrading existing FDDI links using Gigabit Ethernet connections could be a potentially lucrative source of revenue for Gigabit Ethernet vendors, if the Gigabit Alliance White Paper is accurate. It proposes FDDI upgrades as another likely migration scenario. FDDI is facing a considerable amount of opposition thanks to the gradual uptake of other technologies based on fibre communications, such as ATM. One of its biggest disadvantages is its lack of speed compared to other networking technologies. FDDI provides a bandwidth of up to 100Mbps, the same as Fast Ethernet, but ATM can provide considerably more than that, and Gigabit Ethernet offers higher bandwidths, too.
In future, the Alliance predicts that Gigabit Ethernet will finally find its way to the desktop as companies become more familiar with the technology.
Installing a Gigabit Ethernet NIC straight into a high-end workstation feeding into a Gigabit Ethernet switch would benefit particular niche applications such as desktop trading, for example, where it is necessary to provide video feeds directly to the desktop. Other data-heavy applications that would find desktop Gigabit functionality and advantage include CAD.
However, not everyone is convinced that Gigabit Ethernet will be embraced on the desktop. According to IDC's Bildebeek, desktop Gigabit Ethernet will not appear on the market's radar until at least 2002. He agrees that the technology may appear in a couple of minor niches, but says that in the majority of cases, even if someone produces a Gigabit Ethernet card that sells for a small premium, it will mainly be used for futureproofing.
He implies that most companies will not have a desperate need to use this much bandwidth at the desktop. How quickly will Gigabit Ethernet catch on? Bildebeek says it still represents a very small proportion of the networking market.
IDC figures outlining network equipment shipments in Europe reveal that by the year 2000 there will be two times as many ATM ports as Gigabit Ethernet ports. ATM ports currently represent 2% of the market volume in Europe, while Gigabit Ethernet is so small as that it is immeasurable.
By the year 2000 Gigabit Ethernet will represent 2% of the network ports sold in Europe, he forecasts.
Switches are the biggest market for Gigabit Ethernet, says Bildebeek, simply because Gigabit Ethernet is still very much a backbone technology.
He adds that adaptors and hubs also play a smaller part. Gigabit Ethernet ports on routers also represent a small percentage of revenues.
Bildebeek divides the Gigabit Ethernet market into two types of vendor.
The first vendor sells Gigabit Ethernet equipment as its sole source of revenue. Vendors in this category include Foundry Networks and Extreme.
The second type of vendor is much larger, having fattened itself on other, more lucrative areas of the networking market. The big four vendors in this category are 3Com, Cisco, Cabletron and Nortel Networks.
"In the long run, I see people like the big four dominating the Gigabit Ethernet market if only because they have a very large installed base of Ethernet and Fast Ethernet customers," Bildebeek says. There will still be a market for the Gigabit Ethernet-only players, but it will probably be smaller because of their lack of an installed base, he adds.
CASE STUDY: INDIGO
Gigabit Ethernet case studies are hard to find thanks to the immaturity of the technology. One early adopter was Indigo, an ISP which upgraded its Fast Ethernet backbone to a fibre-based Gigabit Ethernet system in December 98.
The company has roughly 50 Unix-based servers from suppliers including Dell and Sun. In the firm's previous infrastructure, they connected into a Cisco Catalyst 5000 switch. Some smaller servers were aggregated into smaller switches that would connect to one Catalyst 5000 port.
"We decided to change because we were moving premises so we had the opportunity to review our infrastructure. To make the move smooth for customers we decided to build a new infrastructure in the new location and then move our machines," says Brian Boyle, network operations manager for Telecom Eireann-owned Indigo. The firm purchased four FastIron II Gigabit Ethernet switches and a FastIron Gigabit Ethernet backbone from Foundry Networks, and connected its servers into the switches using 100Mbps links. The FastIrons talk to each other using Gigabit speeds.
Boyle explains that thousands of customers log on to terminal servers and need access to the international server which is usually connected to a different switch. "We will also have customers with dedicated leased lines off another switch that will want to utilise the international server," he adds.
Apart from the future-proofing that the Gigabit Ethernet backbone gave Indigo, Boyle adds that it has also made it possible to support more customers.
Because the throughput between the switches is higher, the ISP can add more servers per switch without overloading the switch or the backbone.
Boyle, who has another two switches on order, forecasts that he won't need to add any more switches for at least 12 months.
Microsoft seizes control of phishing sites linked with Russian state hackers
Fitness trackers over-estimate the number of steps their users take, analysis of 67 research reports suggests
Everything we think we know about the imminent Apple iPhone 9, iPhone 11 and iPhone 11 Plus launches
All the latest rumours about Apple iPhone Displays, CPUs, launch dates and even prices
Nvidia brings Turing microarchitecture into the high-end gaming segment