Packet-Optical Stays Out of Control
It's due to a combination of technological shortcomings and carrier reluctance.
One example: Generalized MPLS (GMPLS) still hasn't become ubiquitous in optical networks, according to morning keynoter Shankar Narayanaswamy, vice president of network architecture for Global Cloud Xchange .
He asked carrier representatives in the audience to raise their hands if they actually used GMPLS -- all he got were a couple of wavy-handed "sort ofs."
On paper, GMPLS does what a carrier would want, providing a familiar way to control the optical network. "I like GMPLS," Narayanaswamy said. "It extends the same MPLS we know and love down to Layer 1."
But it hasn't worked that way in practice, nor in lab tests that Narayanaswamy's people have done. Part of the problem is vendor interoperability, as he notes in this video interview.
Ideally, Narayanaswamy would like a multilayer control plane, something that spans the optical network up through Ethernet and Layer 2.
Such a thing is possible, but many carriers seem reluctant to actually try it. That's partly because of an old problem: the schism between the IP and transport sides of the company. "If I hearken to my experiences with the carriers, there hasn't been a lot of takeup of this application," said Karl Horne, senior director of portfolio management at Ciena Corp. (NYSE: CIEN), during an afternoon session on control-plane options.
GMPLS, which has been an IEEE Communications Society effort, isn't the only control plane work that's out there. The International Telecommunication Union (ITU) and Optical Internetworking Forum (OIF) have completed some specifications, too.
In the OIF's case, version 1.0 of an external network-to-network interface (E-NNI) was completed about six years ago, with the 2.0 version in progress now. But carriers have preferred to stick with the internal network-to-network interface (I-NNI). The E-NNI has just lagged too far behind in technology, partly because it's been hard to get consensus on what to include, Horne said.
It's also worth noting that a control-plane change is a complicated chore for a carrier, said Vishnu Shukla, a principal technologist at Verizon Communications Inc. (NYSE: VZ).
"When we are going to change the control plane, we are changing the process. Changing the process in the network is slow, because there is already a process in place which works," Shukla said. "One of the gating items is to convince our operations people that this technology works and that it makes life easier."
Joe Whitehouse, director of marketing at Metaswitch Networks , mentioned the idea of isolating the route-selecting part of the control plane onto a server, or maybe on a particular line card in a network element. It's an IETF draft called the Path Computation Element (PCE), and it's been pitched as a less expensive and more easily upgradeable option.
It might sound like OpenFlow, the specification for moving router and switch control planes to a server. And, in fact, Whitehouse mentioned OpenFlow at the end of his panel presentation: "It's something we should be keeping an eye on, because it could solve a lot of problems in these carrier networks," he said. (See Interop Watch: Talking OpenFlow & 100G, Why OpenFlow Isn't Like Active Networking and A New Ethernet Contender.)
— Craig Matsumoto, West Coast Editor, Light Reading