There has been interest in carrier Ethernet for a decade or more and -- let's be honest -- more than a little hype, too. In the early days, the focus was on how Ethernet was going to displace SONET and Synchronous Digital Hierarchy as a low-level optical technology.

Then we were going to have Ethernet to the home, or maybe to every business site. Recently, with the advent of Provider Backbone Transport (PBT, also called Provider Backbone Bridging and PBB-TE), we've heard people say that Ethernet was going to replace MPLS. Is all this just part of a technofantasy, or is there something going on here?

There's something going on. When I surveyed ten network operators about their priorities in 2005, carrier Ethernet didn't rate in the top three for any of them. Today, ten out of ten rate carrier Ethernet as either No. 1 or No. 2 among their technology issues, and the reason they give is PBT. With PBT, Ethernet acquires traffic-engineering features that many believe are as good as or better than MPLS; that, of course, is why there's so much fuss about the battle between PBT and MPLS today.

The real story here lies deeper, however. Network operators are looking hard at what should have been the real issue of next-generation networks (NGN) all along, which is how these networks can best form the foundation for all the services future enterprises and consumers will buy.

It's a foundation - maybe

Ethernet isn't a replacement for SONET, nor is it the basis for enterprise-transparent LAN services or something; it's a contender for the foundation of NGNs, and that's not only big news to the industry, it's a big change in the Ethernet mission. If Ethernet is going to be useful as the foundation for flexible service delivery, it needs some critical capabilities, and vendors are starting to step up and offer them.

One capability that's critical is scalability and traffic engineering, which PBB and PBT provide. These let Ethernet infrastructures scale handle not only major metropolitan areas but whole countries, and simultaneously provide for stringent service-level agreements (SLA) and controlled failover modes to handle node and trunk problems.

Nortel has been the vendor champion of PBT from the first, Nokia Siemens and Huawei Technologies have joined in, and smaller players, such as Extreme Networks and Meriton Networks, also have been very visible in their support.

A related challenge is the need for a control plane. PBT achieves its benefit in part by dispensing with all the discovery and adapting that take place with standard Ethernet bridging, but you can't route traffic or engineer capacity if you can't find nodes and endpoints. PBT was designed to use an independent control plane, and two vendors have stepped up to provide one: Avici Systems' Soapstone Networks business unit, which is exiting the router business to focus on control plane development; and start-up Gridpoint Systems. Both vendors offer carrier-Ethernet control-plane tools, and both have demonstrated their ability to create and control predicable, Ethernet-based service infrastructure in a number of trade shows and events.

The third challenge is the support of services, which is what this is supposed to be about. Service support for infrastructure means support for the three connection topology models that the Metro Ethernet Forum (MEF) defined years ago--E-Line for point-to-point, E-LAN for multipoint and E-Tree for multicast.

Hammerhead Systems just announced full support for the MEF models for carrier Ethernet and MPLS, as well as interworking between MPLS and PBT (Hammerhead also announced a partnership with Soapstone).

Network operators BT Group and DT (Deutsche Telekom) have expressed a level of commitment to carrier Ethernet and PBT, and it's pretty likely that in 2008 at least four other major operators will join them. Carrier Ethernet and PBT have got the vendors named here good engagement with operators world-wide; in fact some of these vendors tell me that they're almost consumed with requests for information and for devices to test.

Some love it, some hate it

Not everybody loves PBT, particularly router vendors that favour IP/MPLS. Cisco, Juniper Networks and Alcatel-Lucent are counted in the camp of PBT opponents, though they all surely are considering PBT support as operators become increasingly strident in their demands to hear about it. Ericsson's position is less clear, but I've recently heard there is a movement within that company to provide support for PBT in some form. Foundry Networks is said to be looking at PBT as well, but there are no references to it on the company Web site.

The two main drivers behind PBT are stringent SLA control and cost. The adaptive behaviour of IP, with its dynamic reconfiguration and routing, makes it difficult to write enterprises the same kind of SLAs they had for frame-relay services, which inhibits convergence. This is one reason an ex-BT executive has given for BT's interest in PBT.

While advances to MPLS -- particularly T-MPLS -- promise similar non-adaptive behaviour, carrier-Ethernet switching products are reported by operators to be about 40 percent less expensive than routers, so PBT has a significant cost advantage over T-MPLS, if there are no other reasons to deploy routers.

Whether the network is a green-field site -- having neither significant router nor significant Ethernet infrastructure -- is the big issue for PBT. If a network is a green-field, building a carrier-Ethernet PBT network would be significantly less expensive than building a T-MPLS network using routers. Metropolitan networks seem a pretty sure place to deploy carrier Ethernet and PBT, although IPTV in the form offered by Alcatel-Lucent deploys IP features in these networks.

In wider-area applications, where some routers are almost certain to be used, the cost advantages of PBT may be smaller. The enhancements to carrier-Ethernet control-plane and service models may make the difference between carrier Ethernet as a niche player and carrier Ethernet as a full-scale infrastructure alternative to IP/MPLS.