The idea that Ethernet could become as ubiquitous and cost-effective in public carrier networks as it already is in private networks is the driving concept behind an Institute of Electrical and Electronics Engineers Inc. study group on "Ethernet in the First Mile."
The EFM study group, a precursor to full-blown standards activity within the IEEE, is developing standards for carrier-grade Ethernet that could be deployed as a low-cost broadband alternative to DSL and cable modem, and as a replacement for SONET.
If the group's efforts are successful, carriers would be able to save money by deploying Ethernet-based capital equipment, which is less-expensive than ATM or SONET switching gear. Network executives would reap the benefits of less expensive customer premise equipment for WAN connectivity. And both businesses and consumers would benefit from broadband access that would cost less, be easier to get and offer more bandwidth than DSL or cable modems.
Before all of this can happen, there needs to be a fundamental change within Ethernet. While several competitive local exchange carriers (CLEC), such as Yipes Communications Inc., Telseon Inc., XO Communications Inc., Cogent Communications Inc. and Fiberstreet Inc., are deploying Ethernet-based public networks, the key will be convincing the traditional carriers that Ethernet has the operations, administration and management features necessary to be considered a carrier-grade transport technology.
To address those issues, the study group is defining carrier-grade Ethernet for three topologies: long-distance, point-to-point single-mode fiber; voice-grade copper loops; and passive optical networks. In addition, operations, administration and management enhancements, such as remote monitoring, remote loopback and remote failure indication, are being worked on.
"I see the change to Ethernet in carrier networks occurring fairly gradually at first, but that won't last long," says Jonathan Thatcher, chair of the IEEE's 802.3ae 10G Ethernet standards group and principal engineer at World Wide Packets.
"The incumbents will eventually be driven to go to Ethernet by competition, availability of low-cost carrier grade products, improvements to the standards and most significantly, by demand from large corporate customers," he adds.
Today, Ethernet rules the LAN, while SONET is the dominant transmission technology for the public WAN. So, in order to route Ethernet-based TCP/IP traffic into or out of the public WAN, some form of translation, such as ATM, packet over SONET, DSL, cable-modem or serial point-to-point high-level data link control encapsulation is required.
Because more than 95 percent of Internet communications begin and end as Ethernet formatted frames, the idea behind EFM is to use Ethernet instead of SONET as the primary transmission technology for public multiservice networks.
Ethernet traditionally has been a private LAN topology supporting point-to-point distances up to .062 miles over unshielded twisted pair copper wire, up to 1.2 miles using multimode fiber optics, and up to 3.1 miles using single mode optics, at speeds of 10M, 100M and 1,000M bit/sec.
However, during the past five years, advances in optical technologies combined with removal of the "collision domain" limitations of true carrier-sense multiple access with collision detection operation have allowed Ethernet to stretch to many tens of miles over single mode fiber (even over a single fiber).
Also, recent efforts within IEEE 802.3 on 10G Ethernet as well as in International Telecommunication Union activities in X.86 and generalized frame processing have attempted to define how to put Ethernet frames into a SONET wrapper for transmission over the public WAN backbone.
Physical layer technologies already standardized for ANSI T1E1.4 very-high bit rate DSL (VDSL) can also be borrowed and adapted to provide high-speed (10M bit/sec and above) Ethernet over voice-grade twisted pair cabling at distances longer than .49 miles.
The carrier perspective
At first glance, there may not be a huge incentive for incumbent carriers to change from a SONET model to an Ethernet transmission model because providing more bandwidth at lower costs is not necessarily in their best interests.
However, there are immediate and powerful incentives for their most profitable segment - large multinational corporations - to force these carriers to provide Ethernet-based access services.
For example, a carrier executive who didn't want his name used says, "If we deliver a 45M bit/sec DS3 to the corporate premise, it will cost our customer around US$50,000 [$25,000 per end] in capital expense for DS-3 interfaces and $42,000 per year in WAN support staff costs to support that connection. If we deliver a 100M bit/sec Ethernet to the site, use flow control or traffic shaping to limit it to 45M bit/sec, and charge the same monthly rate, we can still help our customer save lots of money. The customer's capital cost to connect to the 100M bit/sec Ethernet link drops to around $400 [$200 per end] and the staffing cost is $24,000."
The carrier executive adds, "In a reference design that we did for a hypothetical virtual private line central office service, we designed it using Ethernet switching, ATM switching and traditional SONET circuit switching. Ethernet was cheaper by a factor of 3-to-1 over circuit switches and by a factor of 10-to-1 over ATM switches."
But he added that Ethernet was still too immature to deploy because there is not enough operational and management capability provided by the underlying technology.
Ethernet must become more robust if it is to be successful in the carrier marketplace. Carriers must be able to ensure the integrity of their networks and meet customer service-level agreements without fundamentally impacting customer traffic.
Basic features that have long been taken for granted in SONET, such as remote loopback, and link statistics, have not existed within Ethernet at all. The EFM study group is addressing these issues, but the specifics of how to do this have not yet been agreed. There is debate on whether to do these functions completely out of band from the bearer traffic or whether to try to leverage the rich set of in-band SNMP-based statistical and diagnostic capabilities developed for the enterprise Ethernet market.
Ethernet to the home
The real prize for Ethernet may be residential broadband services. With more than 120 million residences in North America, only about 5 percent are currently subscribing to any form of high-speed (more than 300K bit/sec) broadband Internet service.
The deployment and adoption rates for DSL and cable-modem services have been frustratingly slow for many potential customers. This process has been further complicated by the financial failures of a number of DSL-based CLECs. Finally, by the time currently planned DSL and cable networks are finally deployed, future performance enhancements required for broadcast video, HDTV, Web access and home entertainment services may be well beyond the scalability and capability of those deployments.
"I see the subscriber access network as the last horizon for Ethernet. We have done 10M bit/sec, 100M bit/sec, 1,000M bit/sec and now 10,000M bit/sec. The cost of Ethernet switching is dramatically lower than competing technologies such as DSL and ATM. Further, with support for running Ethernet on low-quality voice-grade copper at much longer distances than 100 meters, new opportunities will open up in Ounstructured' wiring at hotels, older office buildings, hospitals, airports and multitenant units," says Howard Frazier, chairman of the IEEE 802.3 EFM effort.
Frazier adds that the price/performance ratio of Ethernet-based equipment for residential broadband services can be two orders of magnitude better than current DSL equipment. And DSL requires an ATM backbone, which is more expensive than an Ethernet backbone. "So the cost of the whole network infrastructure can be less in absolute dollars at the same time it delivers much higher performance to the subscriber," Frazier says.
More than 77 companies and 125 people attended the March EFM Study Group meeting, with as many as 200 attendees at the May meeting in St. Louis. Vendors included Cisco Systems Inc., Nortel Networks Corp., Intel Corp., Alcatel SA and Lucent Technologies Inc., plus many newcomers.
Cisco is shipping a Catalyst switch that uses a proprietary implementation of Ethernet VDSL to deliver up to 15M bit/sec over voice grade copper cable at distances up to 5,000 feet. This so-called long-reach Ethernet is being targeted at hotels and other multi-unit buildings.
"But for this technology to be truly disruptive, which we believe it can be, it must be standardized," says Bruce Tolley, manager of emerging technologies at Cisco.
Ruby is a 28-year veteran of the computer and data networking industry. He recently retired as vice president and CTO of Avaya Inc.'s Enterprise Internetworking business. He can be reached at email@example.comWhen will it be ready?
After a "Call for interest" in November 2000, the IEEE 802.3 EFM study group began its efforts in January. This study group is expected to transition to a full standards effort as the IEEE 802.3ah Ethernet in the First Mile (EFM) this fall. While it is still early in the standards process, if EFM is able to complete its work on schedule, we would expect to see a "ballot-ready" draft document from IEEE 802.3ah EFM by January 2003.
The standards effort is focusing on three areas:
Long-distance, point-to-point optical EthernetThe technical objective is to standardize operation of 1,000M bit/sec (1000BaseX) Ethernet over a single fiber (instead of a fiber pair) at a distance greater than or equal to 6.2 miles. This new fiber-optic physical layer interface could be used to provide lower cost, single-mode optical interfaces, using half the amount of fiber currently required for Gigabit Ethernet at distances twice that currently specified by the standards. State-of-the-art 1300nm and 1550nm single-mode optics will be leveraged to support this work. This will be a relatively minor enhancement to existing Ethernet, using technology that is generally considered to be low risk. This kind of connection would be most applicable to high-speed corporate networks.
Voice-grade copper Ethernet
The technical objective is to support point-to-point distances greater than or equal to 2,500 feet at speeds greater than or equal to 10M bit/sec. Technical presentations have shown operation at speeds near 50M bit/sec for shorter distances with operation at lower than 10M bit/sec speeds at much longer distances. While some study group members want to just "steal" the physical layer specification developed for ANSI T1E1.4 very-high bit rate DSL, there are also other proposals that will meet this requirement. This activity will likely require more work than long-distance optical fiber, but in the true spirit of Ethernet it will "borrow" from past work.
Passive optical Ethernet
Perhaps the most aggressive and controversial of the three topologies being considered, the Ethernet Passive Optical Network (EPON) technical objective is to develop an 802.3 compatible physical layer interface for operation at 1000M bit/sec, at distances greater than or equal to 6.2 miles, with support for at least 16 or more end points, in a shared, point-to-multipoint topology. EPON expects to borrow heavily from work already done within the post, telegraph and telephone administration-sponsored Full Service Access Network (FSAN) consortium.
A passive optical network (PON) uses a single fiber from the central-office head end (or other active location) to service 16 to 32 end points. The optical signal would be split by one or more "passive" splitters mounted midstream in the network. In this topology, any end point could "receive" downstream traffic from the head end at up to the full 1,000M bit/sec, but would have to share the "upstream" bandwidth with up to 32 other nodes, thus limiting its share to about 30M bit/sec.
The PON architecture has inherent advantages in that less fiber is required to service more users, plus power is not required in manholes, on poles, or at the curb. PONs may also have a lower-cost electronics structure, however, given the still declining cost of very inexpensive Ethernet switches, it is not entirely clear that this will continue to be the case.
Unfortunately, unlike the ATM PONs defined in FSAN, Ethernet operation over point-to-multipoint networks is currently undefined. Architecturally, Ethernet interfaces today expect the underlying network to be based on either point-to-point links or a shared peer-to-peer topology. Without more work, the EPON will "break" certain behaviors that higher layer protocols such as IEEE 802.1 bridging and others expect. Thus, it is expected that more work will be required to define the EPON than the prior two.