Back Talk by Sam Masud
A conversation with Brian Fink of Global Crossing
telecommagazine.com
Q Let's begin with the fiber. Is it Global Crossing's plan to own the fiber?
A The strategic direction is to own our assets wherever it makes economic sense. For example, all of the oceanic fibers are ours, most of the fiber route miles in the United States are ours, although we've done a few swaps in the United States to get coverage quickly. The direction in Europe is also to own the fiber. Even the oceanic cables to South America and Latin America are ours and also across the South American continent. We want to own as much of the actual assets as possible.
Q Can you briefly describe your network architecture?
A The way we've constructed our IP network is fairly straightforward. We've built a completely meshed network on a global basis and we've interconnected high-speed core routers directly to the WDM layer. We have a second made up of Lucent CBX and GX switches, which like our IP network interconnects directly to the WDM layer. We're also running Lucent's IP Navigator on those multiservice switches. That asset is being used primarily for the ATM and frame business, but we're also using it for our IP service sets to transport voice capabilities. ATM brings to the network measurable or absolute QoS at Layer 2, and with IP Navigator we can map some IP capabilities into the ATM structures.
Q Whose DWDM equipment are you using?
A We have a mixed-vendor environment. We have some from Pirelli (now Cisco), Hitachi, NEC and Lucent. There's been some issue of interoperability but that was worked through rapidly. We're also using some Nortel equipment. Nortel's acquisition of (ultralong-haul DWDM vendor) Qtera will enable us to transport traffic over long distances without regenerating the signal, and there are some situations where we need to do that.
Q Ultralong-haul DWDM equipment vendors promise all kinds of distances, but how far really is it necessary for the signal to go?
A That depends on your network design and the kinds of things you're trying to do with your service structures. Network services that are leveraging optical technology will drive those requirements to a certain extent. For example, if you've got a couple of data centers that are 3000 miles apart and want to do high-speed replication between them and offer it as a service. The requirements are driven by the customer.
Q How many wavelengths do you have in the network core and what proportion of these are OC-48s and OC-192s?
A We don't like to get into specific numbers, but it's safe to say that it's about a 60-40 split between OC-48s and OC-192s, respectively.
Q How important is OC-768?
A We're working with our key vendors to understand that technology and its availability. We've got a track record for making sure that next-gen capabilities are being put into our network. The issue we have to look at is the economics of going OC-768 with fewer channels vs. more OC-192 channels. The bigger issue is: When are we going to get OC-768 interfaces on routers? You have to play those two together, because if you can only get line rate out of a router at OC-192, an OC-768 core loses some of its value. There are some next-gen router vendors touting OC-768. We've been talking to Cisco, Juniper and others. I'm on the board of directors of Global Crossing Ventures and we've invested some start-up capital in Pluris and have them in our R&D lab.
Q Are you getting the kind of port densities you need from the router vendors?
A The thing we've been pushing pretty hard is line rate speed at OC-192, but there have been issues with certain vendors being able to provide that. We're continuing to work with Cisco on the GSR because they currently do not do line rate, although Cisco has plans in place to deliver that and schedules to deliver it to us. That is also why we've got some Juniper equipment in our network. We didn't want to be caught because one vendor couldn't meet our needs, as we'll be adding a lot of OC-192 IP routes in our core IP network next year. We're pushing both Juniper and Cisco in this direction.
Q Is the core router market wide open?
A We've spent time talking to a number of the key players--Nortel, Cisco, Juniper, Pluris and Avici. We made an investment in Village Networks just a few weeks ago. Village Networks is a little different play, although the result will be the same because they are combining the optics and the IP layer. There's also another company out there, Lantern, that we'll be meeting with. We move a considerable amount of IP traffic-- it's growing about 400 percent to 500 percent per year--and that's putting a big demand on our IP network.
Q Do you consider these routers carrier-class products?
A Is there a pure IP switch that's carrier class yet? It's a matter of opinion based on what you put in your requirements. I worked for Bell Labs and carrier class to us was stuff that never went down. The IP world isn't quite there yet. The current product set, what we and others have deployed, still lacks a bit from what is true carrier grade. The vendors recognize this and you'll see improvements in speed and their ability to deliver true carrier-grade products.
Q You have the Lucent NX64000 (formerly Nexabit) as well as Cisco and Juniper. How come?
A We tested the NX64000 but today it's not in our production network. We're using the Cisco GSR and the Juniper M40/160.
Q And what about the Lucent LambdaRouter, which is not a router?
A Correct, it's not a router. Village Networks is building something you could call a LambdaRouter because it actually will do IP switching and routing. We're going to trial the LambdaRouter between London and the United States. It will provide protection, restoral and meshing. Lucent has a product that I think is called SoftWave, which manages and controls the LambdaRouter. We want to evaluate and understand whether there is anything here that we can productize from a transport perspective.
Q How important is MPLS (multi-protocol label switching) for you?
A It's a very important standard and we're supporting it strongly. We're running MPLS on our core IP network today and it's a fundamental requirement in the long term. It's how we plan to interconnect our ATM asset with our IP asset so it becomes a seamless look-and-feel network. We're currently using MPLS for traffic engineering, and we've got lots of [MPLS] tunnels, but we intend to evaluate it for VPN-type service, particularly on our core network.
Q Since you have products from Sonus Networks, what's your VoIP strategy?
A Our VoIP strategy is really simple because it's targeted at the Class 4 market. We're doing over a billion minutes a month and we're growing this business rapidly. We bet pretty strongly a year and a half ago that VoIP assets are going to be less expensive to procure and operate, and frankly without it the voice business becomes a pretty tough business to be in.
Q What about Class 5 services?
A We're not sure Class 5 is a play for us yet, because we're providing a lot of carrier services today. We've also targeted the multinational corporations. The more important features on our voice network are going to be things like voice VPNs, optimal routing, international traffic, these sorts of things vs. Class 5 features.
Q Back to optics. When it comes to optical switches, are you in the all-optical camp or is an optical-electrical-optical conversion okay?
A We really don't see this as a major issue [and] we don't get into a heated debate over which is best. They both have their place in our network architecture for different reasons. Some are better for the edge, some for the core [and] we will use both where appropriate. A meshed network's something that's going to be very important given the $18 billion worldwide network we're building.
Q Regarding the metro: Do you favor the big-pipe optical transport vendors or the multiservice provisioning platforms?
A We haven't announced yet what equipment we're using and which vendor we're working with for the metro networks. As far as the multiservice provisioning platforms go, they're first-generation products in this space and will require a revised network architecture. The available DWDM solutions support wavelength services and will have a broader application when subrate multiplexing is implemented. |
