When Bob Metcalfe was working for Xerox at its Palo Alto Research Center (PARC), his problem was how to create a network that would enable PARC computers to use the world’s first laser printer. His answer, summarized in a May 22, 1973 memo, was to create Ethernet running at 2.94Mbps. No one knew that Ethernet would become the most popular LAN technology around.
Fast-forward to 2002, and major Ethernet companies researchers and engineers are readying the final version of the 10-gigabit Ethernet (10GbE) standard.Today, 10GbE is a reality, after its IEEE standard IEEE 802.3ae was finally approved in early June. Cisco, Extreme Networks, Foundry Networks and Nortel are already shipping in trial runs, pre-standard 10GbE equipment and the June Supercomm show featured two dozen companies showing it doing its stuff.
Even before it was approved, the technology was already being used in some products. This is because, Val Oliva L2/3 product marketing manager for Foundry and member of the 10 Gigabit Ethernet Alliance (GEA) Board of Directors, explains, ‘It was designed this way because the development of 802.3ae is built upon an ‘alliance,’ which includes vendors from chip to system vendors. No other standard body (GEA) has ever performed a task that created a standard, ensure that there is a working form of the standard, and ensured that there is “inter-operations” of the standard.” In short, 802.3ae was meant for pre-standard, early adoption.
10GbE isn’t you dad’s Ethernet though. For starters, it only runs on fiber-optic at this time. Besides simply running at 10Gbps, 10GbE also has traditional Ethernet’s Media Access Control (MAC) protocol and its frame format, minimum and maximum frame size. However, since 10GbE is full-duplex only-there is no half-duplex option-Carrier Sense Multiple Access/Collision Detection (CSMA/CD) isn’t needed or implemented. In addition, 10GbE works only on optical fibre. Besides being faster, 10GbE has greater range than its ancestors. The 10GBASE-EW version can reach up to 40 kilometers.
Don’t look to deploy10GbE from your client’s office wiring closet to their desktops anytime soon. The office still belongs to Ethernet and Fast Ethernet. No, according to the GEA white paper, 10GbE Overview White Paper, 10GbE will server as a backbone switch to switch and server to switch technology that will enable administrators to extend gigabit to the desktop. This will greatly enhance applications such as videoconferencing, streaming video, high-end graphics and medical imaging.
Seamus Crehan, senior analyst for the Dell’Oro Group, also thinks that, “Storage Area Networks (SANs) could be a really big application market for 10GbE. Indeed, Adaptec, Intel, HP, and QLogic engineers think that Internet small computer system interface (iSCSI) could be used to transport data I/O over 10GbE-borne IP.
Where many 10GbE proponents really see 10GbE coming into its own though is outside of Ethernet’s traditionally stronghold of LANs in the wider world of metropolitan area network (MAN)s and wide area network (WAN). In this arena, a unified, Ethernet-based network topology, could ease network management and reduce network-related costs. According to Bruce Tolley, VP of GEA and manager of emerging technologies for Cisco, 10GbE reduces the need for non-Ethernet technologies like ATM and “leverages the installed base of 250 million Ethernet ports” making Ethernet cheaper than other transport technologies and “In the end, economics always matter.”
Will this new Ethernet make the grade? Crehan thinks so; “It will be very successful in LANs and very likely to be highly successful in MANs and WANs.”
The Why of 10GbE
According to Crehan, what’s driving 10 GbE’s development is that since “Ethernet has become an extremely prevalent technology, because it’s inexpensive and easy to use, 10GbE was just the next evolutionary step. Although there isn’t a huge amount of bandwidth drivers now, as Gigabit switches become more prevalent, 10GbE will come up.” He goes on, “people can’t wait until the bottleneck exists, they need to work on standards today.” Thus, 10GbE was a “forward thinking move.”
On the WAN and beyond service providers will be able to have entire networks running on Ethernet, thereby creating a unified network topology. Oliva point blanks says that 10GbE is meant “to push Ethernet in the MAN and WAN, everywhere. In the end, Ethernet will dominate L2, just like IP dominated L3.”
Frits Reip, Senior Product Marketing Manager for Nortel, also has said that 10GbE will have a 84% interface cost saving over SONET and will giving a savings of 30 to 60% per user over other managed network services. In the end, 10GbE adoption will be as much driven by market performance as throughput performance.
Besides the technical differences mentioned earlier, 10GbE also differ from its forefathers in two significant ways. The first is that the optical transceiver or physical medium dependent (PMD) interface for single mode fiber that will work with both the LAN physical layer (PHY) or WAN PHY.
This has caused some people to think that there are two different kinds of 10GbE. That’s not true. Oliva says, “There’s one version and it’s 802.3ae. There are several 802.3ae PMDs (Physical Media Dependent) and they are classified into two categories: WAN PMDs (“W”) and LAN PMDs (“R”).
Another point of difference is that if you go with the WAN PHY option, you transparently transport 10GbE over existing SONET OC-192c fibre. This optional PHY incorporates a simple, inexpensive SONET framer and operates at a data rate compatible with OC-192’s 9.953Gbps speed. The long and short of this OC-192 compatible asynchronous Ethernet interface is that it will enable service providers to seamlessly run 10GbE over existing OC-192 compatible fibre and transponders.
Or, you could even, as Richard Cunningham, a Cahners In-Stat analyst, points out, “run SONET on some wavelengths on a single fiber and run 10-GE on other wavelengths on the same fiber, as long as you don’t have optical amplifiers in the chain.”
Most of 10GbE runs over single fibre optics, but one type, LX4 PMD, is based on Wide Wave Division Multiplexing (WWDM). LX4 uses four wavelengths of light over a single pair of fiber optic cables. With this, the four wavelengths are “bonded” to create 10 Gigabit speeds.
One advantage that 10GbE has over SONET and other optical technologies is that it can be run on dark fibre. Dark fibre is the unused capacity available on dense wave division multiplexing (DWDM) equipment. By making use of this untapped resource, 10GbE will enable service providers to get more use out of their existing fibre.
New Uses For a New Ethernet
Oliva sees the MAN and WAN as 10GbE’s destination. “The challenges for Ethernet’s acceptance into the MAN and WAN are several, and the 802.3ae (10 Gigabit Ethernet) resolves two of those challenges – bandwidth and distance. To move Ethernet’s into the ‘final frontier,’ which is WAN, it needs to make sure that those challenges are knocked off. There’s more challenges, but 802.3ae is a move in the right direction.”
Tolley, agrees writing “with 10 Gigabit Ethernet backbone networks, service providers will be able to offer native 10/100/1000 Mbps Ethernet as a public service to customers, namely offering the customer twice the bandwidth of the fastest public MAN services OC-3 (155 Mbps) or OC-12 (622 Mbps) with no need for the added complexity of SONET or ATM and no need for protocol conversion.” But, the LAN is also important. He notes that 10GbE switches will be used for server interconnect, campus backbones, and aggregating Gigabit switches.
From the outside looking in, Crehan agrees for the most part, but he also cautions that
“There’s already a large, recent installed base of SONET equipment and service providers aren’t looking to junk that anytime soon.”
Crehan also expects that “many switches will have multiple interfaces ATM to OC48 to 10GbE.” In any case, “you won’t see companies fork-lifting new equipment in.” Instead “service providers, who are already trying to squeeze out as much revenue out of what equipment they already have,” will gradually migrate to 10GbE.
He sees the first “significant volume deployment will be dependent on the enterprise as LAN backbone and maybe dark fibre in campus situations.” But, the “price needs to be more attractive then it is right now.”
At first, the prices won’t be that alluring. Crehan says that the first prices you’ll see will be from “20K to 90K list price per port on 10GbE with the shorter reach will be much cheaper than the long reach versions. At 10K distances,” for example, he says, you’ll see: 70 to 80 grand today.” But he expects, “aggressive price drops of from 30 to 40% price will be seen quickly. We’re very early now, but as more vendors come in the prices will continue drop aggressively. On a dollars per gigabit, at least as cheap as gigabit Ethernet.” He thinks that soon you’ll “see a single 10-gigabit Ethernet port at prices that will be price comparable to 10 1-gigabit ports.” By 2005, he expects 10GbE to have become a billion-dollar market.
Quality of service (QoS) is also a concern. Cunningham says, “SONET offers a good deal of redundancy, up to and including a full restoration path, so that the system can sense a path failure, and then reroute the signal to the destination via another path with 50 milliseconds. The 50-ms figure, also known as latency, is a carryover from voice traffic; if someone’s speech is broken up in chunks and shipped from here to there with more than a 50-ms differential delay, the result is garbled to the human ear.” Ethernet on its own, plain old Ethernet or 10GbE, doesn’t have that kind of redundancy. Some question though remains as to whether 10GbE needs these kinds of voice-carrier QoS features.
Some observers also wonder whether even at 10Gbps whether 10GbE is fast enough for the long run. A few of them cite OC-3072 SONET, which would run at 160 Gbits per second, as an example of a technology that would blow 10GbE’s doors off. Cunningham replies, “Don’t hold your breath for OC-3072. That’s a long way in the future. Granted, Lucent and the Heinrich-Hertz-Institut in Berlin will be doing some experimental work on in-ground fiber near Darmstadt this summer, but it’s hard to imagine any 160-G stuff deployed before 2010, and I’m being optimistic there. There are a huge number of physics problems to be overcome before that can get out of the laboratory.”
But as for 10GbE, it’s only real problem, according to our experts is price. Crehan expects 10GbE pricing to drop to affordable levels even for a tight economy. If it doesn’t, he foresees a far less rosy future for 10GbE.
Oliva though is sure that 10GbE will prosper. As she says, “Ethernet is IP’s best friend, and in the end, IP wants Ethernet.”
SIDEBAR: Gigabit to the Desktop?
While there may be few deployments of 10Gbe in the next few months, where it is deployed, you might be able to turn this into additional work by suggesting the deployment of gigabit Ethernet to the desktop. With sufficient bandwidth now available over campus WAN networks, gigabit to the desk is much more attractive since a few desktops can’t potentially eat an entire network bandwidth with one gigantic, ill-timed file download gulp.
But, for that to happen, you’ll need to check your gigabit network cards and drivers very carefully. Many network integrators have complained to me that gigabit cards have an abnormally high failure rate and that the drivers, especially on Windows, aren’t all they’re cracked up to be.
All that said, with proper testing, you may be able to persuade customers that so long as they’re ramping up their WAN backbone that they might want to consider giving their most highly used LANs a Gigabit steroid injection.