Ethernet takes a step towards service providers
Ethernet, one of the most successful network standards ever, is poised to take a new role, as a new telco-friendly version emerges, suitable for metropolitan area networks.
In order to become the ubiquitous local area network (LAN) standard that it is, Ethernet has had to see off several competitors, including token ring, ATM-on-the-LAN and FDDI. It now has an unrivalled position on company's local area networks (LANs). The only other network specification to grow significantly is the specialised Fibre Channel, used in storage area networks (SANs), and even this is predicted to eventually be superseded by Ethernet.
So where next for Ethernet? Users who have Ethernet internally want to join those networks together to make global networks, and they have done this for some time, by encapsulating Ethernet traffic and carrying it over whatever private and public networks are available. Ethernet interfaces are being offered by most telcos, for long-distance services.
Ethernet, once carried over coax cable, has long since jumped to fibres, and would have no problem reaching the distances over which those service providers operate. So why can't service providers run native Ethernet for those long distance networks, and simplify things immensely?
The answer is that service providers have quite a lot of different requirements. Long-distance networks have different characteristics to the ones Ethernet was designed to meet, such as reliability, conserving bandwidth which may still be costly, and offering multiple clients different quality of service.
Standards like ATM still rule the world in these areas, for the longest-distance and busiest networks, but a new area has opened up where Ethernet can offer significant benefits.
Metropolitan area (or "metro") networks span anywhere from a city to a region as big as some European countries. There is currently a big bottleneck in the metro. During the late 1990s, masses of long-haul bandwidth was built out, and the last mile began to open out with the spread of broadband. However, for city-wide fast links, there is still not enough bandwidth.
"The last decade has seen LAN capacity expand some 100-fold, and backbone capacity more like 300-fold, while the real bottleneck has been in the metropolitan area, with a bare 16-fold increase," says Nan Chen, president of the Metro Ethernet Forum (MEF). The bottleneck is partly there simply because it is difficult and expensive to lay fibre across cities. However, it is certainly true that, in the current climate, service providers would find it easier to deliver services across those distances if they had access to cheaper equipment.
Most metropolitan area networks are based on the Sonet/SDH optical networking standard, which was designed to provide highly reliable networks tuned to carrying voice traffic on ring networks. To get on in this environment, Ethernet needed some (but not all) of those attributes -- and keep its cheapness.
Enter the Metro Ethernet Forum
New Ethernet standards and specifications from the MEF are virtually complete. The MEF's E-Line and E-LAN standards are being voted on, and will be finalised in August. "Until now, metro networking has come at the cost of of a major step change from Ethernet to more expensive, less familiar and less flexible technologies like ATM or Sonet/SDH," said Chen. "All that is set to change with the advent of carrier-class Ethernet services and transport."
The analysts seem to agree. "In the next 10 years, Ethernet will inexorably take over the metro," stated Michael Howard, principal analyst and co-founder of Infonetics Research. "Of course, there will never be a wholesale change because of the Sonet/SDH installed base, but every year Ethernet will account for a larger portion of metro capital expenditure." Howard predicts a $5.7bn market for metro Ethernet hardware in 2006.
The services market could be even larger. "Even where overall spend is flat or reducing, the metro area is where [telcos'] budget is focused," said Ian Keene, VP and chief analyst for the Gartner Group. Telcos working towards an Internet Protocol (IP) based infrastructure for services will have to spend what money they have on the metro, he said. Gartner predicts a $14bn market for metro Ethernet services in 2006.
Talking the telcos' language
The MEF specifications are designed to give service providers things they understand, and present services that can be sold to users. The standards include "Ethernet virtual circuits", very much like virtual circuits on ATM networks, and service specifications that will give users the familiar "committed information rates" and "peak information rates" they get with existing telco services.
The standards will be referenced by the telco standards body, the ITU (International Telecommunications Union) -- in itself a very big indication of Ethernet's arrival in the telecoms area. Previous Ethernet specifications have been handled by the IEEE's local area network committees.
Making it resilient - a faster spanning tree
A major effort has been put into making metro Ethernet resilient. Sonet networks are up and running again within 50ms of any malfunction, because interruptions to voice calls are very noticeable. Traditional Ethernet networks recover more slowly because of the "spanning tree algorithm" they use to re-map and optimise the network paths.
MEF proposes that metro Ethernet should use a local algorithm to divert traffic along detour paths within 50ms. When the global algorithm catches up and optimises the whole network, these detours can be replaced by more optimised paths if necessary. "It is not efficient if, when the network fails, it stops forwarding traffic while it calculates a new way to deliver it," said Chen. "The local recovery method mechanism may not be optimised, but Layer 3 mechanisms can detect the best path in time."
The drafts of the MEF specifications are already being implemented by product vendors, and a test suite is being developed so service providers can be sure that they are getting standard products. As the products emerge, users should see that end-to-end Ethernet services between locations will be cheaper and more flexible than before.
The competition: RPR is still there Metro Ethernet originally had competition from a specification called Resilient Packet Rings (RPR), which was defined within the IEEE as standard 802.17. Like Sonet, it uses a ring-based topology, but one which is optimised for data traffic. Chen politely says, "these technologies give carriers the option to choose whichever fits best into their needs," but many people see RPR as, at best, a niche standard. RPR is not as scalable as Ethernet, and will not handle Ethernet traffic as easily as native Ethernet, causing more expense to service providers. It is based on pre-standard specifications from Cisco called Dynamic Packet Transport (DPT). It is intended to extend and replsce Sonet, but observers wonder why a service provider would replace its Sonet networks, when it can put in metro Ethernet alongside, and migrate the data traffic away from Sonet to Ethernet. Surprisingly, even analysts who predict a big role for metro Ethernet still believe RPR will have a strong supporting role -- if only because it has adopted a more sensible strategy with respect to Sonet. "Will RPR go away? Surprisingly and resoundingly no," said Howard. "A year ago I thought RPR was dead, but now it lives, and carriers want it. Carriers like rings!" RPR's saving grace, says Howard, emerged last summer: "It will be offered as a Layer 2 method to pack bursty packet traffic onto the world's existing Sonet and SDH rings." A recent Infonetics study of 20 North American service providers showed some 25 percent planning pre-standard (DPT) RPR and 19 percent planning to use the final 802.17 standard in their networks by the end of this year. Meanwhile, however, Ethernet is acquiring carrier class attributes. The MEF is working on making it better able to carry TDM traffic, so that it will ultimately be able to support voice. "With these capabilities, service providers can cap their existing networks, and grow a shiny new metro Ethernet, at much less cost and operational expense than Sonet and SDH based networks," says Howard.