PDF of this tutorialEthernet is (by a huge margin) the most successful networking technology ever. After all, 99 percent of all TCP/IP packets (including Internet traffic) traverse an Ethernet somewhere (and more likely five or six Ethernet LANs in the World Wide Web). And the future is even brighter!
Optical Ethernet to the Consumer
FTTB is a reality for the Fortune 500 today and will be a reality for 95 percent of non–small office/home office (SOHO) businesses within just a few years. A carrier that runs fiber to a business enables the delivery of all of today’s communications services and, likely, all of tomorrow’s. And Ethernet services are the least expensive services that can be provided over that fiber today.
FTTH is already in trial deployments in several communities. Several studies have been completed suggesting that building an FTTH network is no more expensive than a full-scale DSL buildout or a two-way cable television (CATV) upgrade, costing in the neighborhood of $1,000 per home passed. (See http://www.tml.hut.fi/Studies/Tik-110.300/1998/Newtech/ftth.html for a study by Matti Rantanen of the Helsinki University of Technology.) The potential capacity of an FTTH network is orders of magnitude greater than DSL or CATV, so it is clear that for any new buildout, laying fiber is the logical choice. And just as in FTTB, Ethernet is the least expensive technology providing optical-network access today.
It seems clear that Ethernet services will be coming soon to businesses near you—and to residences soon after. The only questions are how soon, how fast, and how expensive? In many metropolitan areas, some service providers already provide Ethernet services by the megabit of capacity: you can buy from 1 to 100 megabits, as needed. A few service providers already sell Fast Ethernet (100 megabit) Internet access for $1,000 per month—a truly compelling price when compared to the Old World SONET– or ATM–based services.
Still, is Fast Ethernet fast enough? Of course, that depends upon the business, but the cost of laying FTTB dominates the capital expenditures, and today’s Fast Ethernet infrastructure can be readily upgraded tomorrow. The switching cost and Internet-access cost are still much higher for Gigabit speeds, but those costs are rapidly declining. For Ethernet services between locations for a business, Gigabit Ethernet may already be needed to support file servers, backup servers, and other intranet applications. After all, most workstations installed today come with fast Ethernet built-in, implying that the switching and office-services infrastructure should be significantly faster to avoid bottlenecking.
Residential applications are more limited. 10-Megabit Ethernet should be adequate for Web browsing for the near future, and the bandwidth needed for telephony is negligible. The most bandwidth-hungry application recognized today is TV, especially premium services such as video-on-demand. A single channel or movie at broadcast quality requires about 4 Mbps; DVD quality requires only about 9 or 10 Mbps; and even HDTV will probably only require 20 Mbps. Of course, these figures are per-channel-viewed, and the industry should plan on an average capacity of perhaps two channels per household at a time. Still, even the most aggressive of these numbers imply that a single Fast Ethernet service to each home (not shared) is more than adequate for the services we know of today.
Optical Ethernet Area Networks
Large optical Ethernet networks are changing the definition of the LAN. “Local” might even be “global.” The original barriers in an Ethernet LAN (3-km span, 1023 nodes, 1 optical repeater) have long since been vanquished. Today, the practical limits are more because of the need to terminate broadcast traffic or to provide security between management domains, or because of today’s limits to the number of MAC addresses that an Ethernet switch might support. Note that VLANs already start to address these issues; and larger VLAN–enabled switches in the future are at least likely to control the problems and isolate them for a really large router to handle.
The practical limits to the size of an optical Ethernet are not geographic; rather, they involve bandwidth, node count, and overlying protocol (broadcast traffic, routing-table size, etc.). As VLAN and other Ethernet services become more common, we may even see large corporate networks simplify into a single, optically connected Ethernet LAN, with only a few large routers providing the necessary functions of security, address management, and interdomain routing.
The largest optical Ethernet networks are likely to belong to carriers, interconnecting all of their points of presence (POPs). Even today’s optical Ethernet technology could be used to create a nationwide (or even global), very-high-speed optical Ethernet that can be used to interconnect the large core routers in each POP. Such a network would have fewer router hops, faster link-failure recovery, and lower cost than today’s “normal” network of OC–192 POS routers.
Beyond 10 Gigabits
Just as the growth of 10-Megabit Ethernet led to the need for 100-Megabit Fast Ethernet, and just as the growth of Fast Ethernet led to the need for Gigabit Ethernet, the growth of Gigabit Ethernet is now driving the market to 10-Gigabit Ethernet. This trend is not likely to stop anytime soon. Servers—whether Web servers, file servers, e-commerce servers, or others—must have greater bandwidth than the customers they serve, otherwise, those customers will feel frustrated with inadequate performance and possibly go elsewhere for service. The best example is Web servers. If the average Web browser is using a 56k modem, a server on a T1 line can simultaneously handle approximately 30 customers. But the broadband movement has already started, and millions of consumers are now accessing the Internet from DSL and cable networks. These consumers access the Internet at speeds up to 10-Megabit Ethernet (typical connectivity for a cable-modem service), and for them, a service provider limited to a T1 line is already unacceptably slow. This requirement to always be faster than your customer (and by a factor of the number of simultaneous accesses) drives the need for T3 (45 megabits) or even greater speeds today and will drive the need for Fast Ethernet and Gigabit Ethernet tomorrow.
Some carriers are already deploying gigabit optical-Ethernet services today. They may limit their customers to a few megabits per second, but the links are gigabit-capable; and someday the fees for gigabit-scale Ethernet services will be affordable. Then, even a 10-Gigabit–Ethernet transport will be inadequate. Engineers are already dreaming of the next step, and arguing over what speed it will be. 40-gigabit speeds (SONET OC–768) have already been demonstrated, so that is a possible Ethernet target. But the Ethernet “purists” insist that the only logical next step is to move the decimal point one more time—to 100 gigabits.
In the meantime, the protocols and techniques for bandwidth sharing over parallel links exist, work well, and are used in thousands of sites. It is a simple step to run parallel optical-Ethernet trunks, each on a separate wavelength, all multiplexed over a single fiber pair using DWDM technology. In this way, a point-to-point Ethernet link could have scores of 10-gigabit channels, with an aggregate Ethernet bandwidth of perhaps 400 gigabits, today! Using recently announced DWDM capacity of 160 wavelengths, 1600-gigabit-per-second links could be implemented. So in a sense, Terabit Ethernet is already available—of course, this kind of network requires very large Ethernet switches at the ends of that fiber.
The limits on optical-Ethernet bandwidth may just be the limits of fiber-optic bandwidth— perhaps 25 Terabits per second for the available spectrum on today’s fiber, which is still well beyond the capabilities of today’s lasers and electronics. Still, extrapolating from recent trends gets us to that level in only 5 or 10 years.
One final thought: nearly every computer “geek” in the country would love to have gigabit Ethernet to their home; and once they get it, they are sure to create many new “killer” applications that consume all that bandwidth, thus driving the need for even more. The cycle may be never-ending.
Figure 5. Gigabit-to-the-Subscriber—to the Global Area Network
