When is a router not a router? It's less a riddle than a serious question facing carriers and vendors alike these days. Drastic changes in network architecture fueled by the addition of video and bent by the evolution of Layer 2 and Layer 3 technologies are changing the function routers serve in carrier networks, tempting one to answer the above question with: Sooner than you might think.

An increasingly sophisticated generation of carrier Ethernet gear is taking more and more work off the hands of today's routers, leading to an evolving role for routers and IP/MPLS technologies in the network. As carriers rebuild their networks for a new generation of services, they must find their own balance of Layer 2 and Layer 3 technologies, even as products based on those technologies play an ongoing tug-of-war over network functions.

“Edge routers are now replicating core router functionality further out in the network, and Ethernet switches are being substituted for routers in many cases because their port costs are much cheaper,” Sam Greenholtz, co-founder of consultancy Telecom Pragmatics, wrote in a June report.

And its not just happening in edge networks. Only weeks after its official christening, Nortel Networks' Metro Ethernet division used the Globalcomm 2006 trade show earlier this month to argue for the replacement of IP/MPLS routers in metro networks with its own 8600 Metro Ethernet Switch. Adding provider backbone transport (PBT) technology to the 8600, Nortel championed PBT as a low-cost alternative to metro MPLS. Based on the IEEE's 802.1ah standard, PBT sets up point-to-point Ethernet service tunnels by affixing a 24-bit media access control (MAC) header atop each existing customer MAC header in packet-based traffic. Unlike Layer 2 carrier networks based on virtual local area networks, which can only scale to 4094 tags, PBT can theoretically issue up to 16 million tags. And it will be simpler and cheaper than MPLS in the metro, the company said, because carrier Ethernet switches need only keep track of their own MAC addresses, not the customers'. That also makes it more secure, Nortel said, because carriers and their customers need not share information about each others' addressing schemes.

But some analysts weren't convinced.

“I was sort of scratching my head,” said Eve Griliches, IDC research manager, of Nortel's PBT pitch. “They're saying they can tunnel more Ethernet channels, which is making it more circuit-like. Any time you add circuits, you have to add end points. It becomes more complex, which doesn't scale.”

Gartner Research Director John Mazur was also skeptical of the case for PBT as a metro network cost-cutter, since — in his view — MPLS' cost per bit isn't bad, either. “You're only going to get so many miles out of a dollar,” Mazur said.

At Globalcomm 2006, Foundry Networks introduced an MPLS-enabled metro access switch, the NetIron M2404, which was designed to extend MPLS capabilities — including traffic engineering and fast re-route — to the edge of the metro access network. The approach seemed opposite in some ways to that taken by one of Foundry's chief rivals, Extreme Networks, earlier this year. Instead of pushing MPLS toward the edge, Extreme advocated Ethernet in the core, unveiling a big carrier Ethernet switch it claimed could scale to support as many as 80,000 subscribers with a 160 Gb/s, 10-rack-unit chassis. The platform reaches that level of scale using a MAC-in-MAC technology similar to the one Nortel has proposed.

That architecture might be a good choice for greenfield networks, said Michael Kennedy, co-founder of consultancy Network Strategy Partners, but probably wouldn't be deployed in America's large incumbent networks. “If I'm building a network in Bangladesh, I'd probably do that,” Kennedy said. “But in the U.S., where we've got a huge embedded base of routing equipment, that's just too radical.”

“People say we're getting past the scalability problem in Ethernet, but I don't believe it,” IDC's Griliches said. “Few people use Ethernet across the core. It doesn't scale, you lose any-to-any connectivity. If you're a smaller provider and you're not going to grow, you may be OK, but [Tier 1 and most Tier 2 carriers] will keep the bulk of their core networks at Layer 3.”

Though big carriers may reject the idea of an Ethernet core, Ethernet's increasing sophistication — with improved quality of service (QOS) and priority hierarchies — is propagating the technology's slow creep upstream, from access to metro toward even core networks. And each carrier must decide where to draw the line between switching Ethernet and routing IP/MPLS, based on perceived trade-offs of complexity, cost and risk. And the requirements for offering video service will loom large in those decisions.

“As the take-up rate for [triple-play] services gets bigger, you're going to want to push the MPLS label-switched paths closer to the edge,” Kennedy said. “To some extent, it depends on how successful the phone companies are at selling video.”

As carriers roll out quality-sensitive video applications, networks must be built to discriminate against services rather than packets. And so the work of routing — examining individual packets — will take a back seat to the important work of examining packet flows, the strings of packets that make up a discrete service.

“If you're designing a network to deliver voice, video and data, you design the network to serve the video,” Kennedy said. “The rest is just trivial in comparison.”

In video-dominated networks, routing will be superceded by subscriber management and network policy management — the job of making sure the most important traffic gets the bandwidth it needs. One example is Tazz Networks' Policy Control System, which is sold by Cisco Systems as its Broadband Policy Manager. The system keeps track of how much network bandwidth is being used at all times and how much is needed for all active services, enforcing programmable network policies such as ensuring the quality of, say, pay-per-view video at the expense of plain vanilla Internet access. In so doing, it lets carriers customize the performance of their networks and the quality of their customers' experiences.

“This market is just starting to happen,” Chad Dunn, Tazz's vice president of marketing, told Telephony in late May.

In turn, the role of the broadband remote access server (BRAS) may be changing as well, as questions mount about the ease of packing every subscriber's high-bandwidth video traffic through a common centralized box. Introducing Gigabit passive optical networking (GPON) platforms in recent months, both Tellabs and Fujitsu Network Communications have argued that their gear can help eliminate the need for BRASs in triple-play broadband networks. Tellabs' 8865, an edge router that supports GPON, identifies packets and the services they relate to and applies QOS policies to that traffic, prioritizing and protecting the most important services.

“In today's network, PPP-over-Ethernet frames have to be terminated in the BRAS and the overhead stripped away,” said Randy Eisenach, Fujitsu's marketing development director. “That puts the BRAS in a bottleneck situation in the network. That was acceptable for basic 2 Mb/s Internet access, but now that you're overlaying real-time services such as video, there's a real issue as to whether a BRAS has the performance to support those services.”

BRASs are likely to continue to do their jobs for Internet and voice services, Kennedy said, leaving video to newer policy management gear. But they'll probably take on more of the important work to be done in securing those access networks, perhaps adding firewalls and security services requiring packet inspection if customers want them.

“Network security will revolutionize the BRAS market,” Mazur said.

Security will be especially important as users interact directly with the network, requesting services that dynamically alter the network's priorities and behavior. Juniper Networks seemed to bear this in mind when it unveiled a new approach to IPTV at Globalcomm 2006 — one with a decided emphasis on security. Juniper's IPTV solution includes protection for the upstream control channel, so hackers can't infiltrate the network by spoofing a set-top and attacking the network.

Add ever-changing security concerns to the list of those already mentioned, and the work done by what we today call a router could one day perform such an evolved set of duties that it may even require a new name. What name? That's yet another tough question to tackle.