Reality Check!
Top network strategists weigh in
from the June 4, 2001 issue of Broadband Week
As the hype of the year 2000 gives way to the harsh reality of the new millennium, telecom industry survivors are looking for the strategies that will ensure they'll still be around for SUPERCOMM 2002.
Broadband Week contributing editor Annie Lindstrom recently conducted a virtual roundtable discussion with optical network strategists from five of the nation's top service providers to discuss their real-life plans for advanced networks. They were Tom Afferton, AT&T Network Services district manager for advanced transport technology and architecture planning; Mike Coghill, Global Crossing senior vice president of global engineering; Joe Lawrence, Level 3's principle architect; Mike Rouleau, Time Warner Telecom's senior vice president of marketing; and Jack Wimmer, WorldCom's vice president of architecture and advanced technology. An edited transcript of those conversations follows.
BBW: What are your plans for 40-Gigabit connectivity?
Afferton: There are two things that AT&T has done to prepare for 40G. The first is to make sure that we have fiber in the ground that will support it in a cost-effective way. We announced at the end of 1999 that we were rolling out 16,500 route miles of new fiber. That has been in progress over last year and will continue through this year, giving us a nationwide footprint of new fiber on top of our existing network. Working with AT&T Labs and our vendors, we are pretty comfortable that it will support 40G.
The second thing we wanted was to have optical line systems that will support 40G when the electronics become available. We recently announced that later this year we will be rolling out a 1.6 terabits per second-capable DWDM system from NEC, which is forward compatible with 40G. Initially, it will be equipped with 10 Gbps lines, and we will get the 1.6 Tbps capacity with 10G. But the view is that 40G is in the road map for that system without having to change out the amplifiers. I don't have a specific time frame for 40G, but we have made the decisions we can make now to be forward compatible with 40G.
Coghill: We have done some testing in Europe. We were able to pass a 40G signal about 10 kilometers to 15 kilometers or so. It wasn't a test of how far we could go, it was more to find out if we could we get 40G equipment working right on our existing fiber.
When we look at 40G and why we would want to have bigger and bigger pipes in our network, it really gets down to the issue of the cost per bit per kilometer. I think the use of 40G or any large pipes in your network is very dependent upon your network topology, meaning how many express routes do you have, what are the distances you might go between drop points on those express routes. Because the advantage I get in cost per bit per kilometer at 40G if I have to access it a lot of places at 2.5G or 10G can be lost by having to do all that add/drop.
Plus there are other issues we have to look at such as existing repeater spacing and all those types of things to see if it can work over our existing infrastructure. If you look at North America and Europe as it evolves for us, we have a lot of add/drops, and every major city we drop a lot of traffic in and out in all the first tier cities and a number of the second tier cities. So, the jury is still out for us. Since we are even now turning up 10G L-band equipment and finding out some of the trials and tribulations of that, such as Raman pumping, even working through those things and having combination L-band and C-band, we are learning things that we and some of the vendors didn't know about until we turned up the systems. So I think we have a lot of work to make that stuff work effectively and be able to manage an all-optical network with that infrastructure before we introduce 40G.
Lawrence: We are a big believer in 40G. We definitely see that it will bring some significant cost savings in the fairly near term. There are some folks that are talking about 40G being three to five years out, but we believe it is going to be viable within the next year or two, and then continue to get better. We expect it to be the same sort of technology shift that 2.5G to 10G was. Level 3's philosophy is one of driving down the cost of communications, so we are not going to jump on the bandwagon of a technology for technology's sake. We want to lead the way when it's economically viable, and we believe that is going to be in the next year or two.
Rouleau: I guess we think about 40G at several levels. Number one is what do we do in the local arena, which is our primary business, and what do we do on our regional infrastructures? Also when we think about a 40G solution we are thinking about OC-192 and OC-192c, and there is continuing interest in 10G Ethernet, and where we go then beyond that. At this point we see 40G as kind of a multiplexed application to deliver 10G interfaces. I don't know that anyone is asking me for an OC-768 yet.
Wimmer: We did a trial with Siemens, which we announced in March. That was a 3.2 terabit, 80 lambda, 40G trial over a few kilometers of field fiber. It was a technology trial intended to confirm the feasibility of 40G and to convince ourselves that we could really use it. We did that successfully. It went very well. It was our first opportunity to look at systems running at that kind of capacity.
The rough time frame for deployment is fourth quarter 2001, or first quarter 2002. Then we have to get it through test cycles, so we are into the early or middle part of next year before we are ready to go on any kind of a scale basis. But we may do some field trials sooner than that.
BBW: What are your plans for deploying all-optical switches in your networks?
Wimmer: We have them in the lab now and as soon as we can get them to pass muster, which we are optimistic will be this year, we will deploy them next year.
The horserace has been between the OEO boxes and the photonic boxes, which are not quite out of the starting gate yet. But the good news is that the demand hasn't gotten so far ahead of us that it's a problem yet either. We did deploy some small 1x1 Nortel optical switches a year ago in our UUNet backbone when we put OC-192 ports on our Internet routers and trunked them across a wavelength at OC-192c data rates between New York, Washington, D.C. and Chicago. We learned a lot by doing it.
Rouleau: We have looked at OEO switches and brought some things into the lab, but we haven't deployed anything at this point. The costs are compelling, at least on paper. But then you have to do the back office integration, surveillance, NMS, OSS ... all those kinds of things. So, we are analyzing them for use in the metro. Where do they fit, what kind of applications can we support? I think that once that infrastructure is in place, the interesting thing will be to move control of the network and provision out to the end user. But I don't know that anyone that has optical switching with that kind of customer interface just yet.
Lawrence: All optical switching is a very interesting technology, but I don't know if it's really brought its benefits to the plate yet. We are much more interested in flexible optical add/drop. Right now, most optical switches provide too much of a performance impact to your link budget. Ideally, what you want to do is put an optical switch in the middle of the network so that traffic that doesn't need to stop off at that site can pass right on through without being regenerated. There are some specific players that provide that capability, but, in general, the technology is not there yet.
We definitely see a role in the mid-term for channel-based optical switching, which really is flexible optical add/drop in some ways. But we definitely see electrical switching playing a significant role, integrated with the long-haul. Ciena has announced that we are using its CoreDirector platform. We are using it to do STS-1 cross connect to support private line services and for its protection capabilities.
Coghill: We have deployed Lucent's LambdaRouter on our Atlantic Crossing sub sea systems to provide protection switching. It is a 256 by 256 switch right now, but it can be expanded to a couple thousand ports if necessary. It's not likely we will do that in this current application. Today, we are using less than 100 ports.
We believe that OO switches will eventually be included in the core of our network. But there is still some work that has to be done around signaling and other network elements to truly see the advantages of that optical switch. We installed the LambdaRouter to get some experience. Over time we will use them to do optical VPNs and intelligent network restoration. The latter might be a year to 18 months away. Issues of interoperability need to be addressed as well.
Afferton: In general we are more application and function focused as opposed to particularly obsessed with individual technology. Going into SUPERCOMM I'm going to be looking at photonic technology, but more from an application standpoint. We are rolling out intelligent OEO switches that will be fundamental to our architecture this year. We have not announced our vendors or specific timetable. We are looking for functionality such as automated provisioning, mesh restoration, the ability to monitor the capacity of our network as we roll it out. If photonic switches come out and show that they can do similar things with better cost or better scalability we will certainly consider that. But right now we are going with OEO because they meet our needs.
BBW: What is the impact of GigE on optical network strategy?
Coghill: It won't really have a big impact because it is a last mile issue. We do have optical Ethernet products being made available. In our world, our intention would be to offer a carrier class GigE solution that would have as its underlying architecture a resilient packet ring, or RPR. The problem today with using GigE for the last mile is that it's really a LAN technology that does not have the reliability or restoration capabilities we are used to in the SONET infrastructures we are used to using in the last mile. It has problems like spanning tree control and things that when there is a failure can take 10 to 15 minutes to restore itself.
Lawrence: We expect it to have a very profound impact. We are very big supporters of the use of GigE in the network wherever it makes sense. So today, in our metro networks, we have colocation facilities from which we run a native Ethernet network straight over fiber for data services in parallel with our TDM infrastructure. The reason we went in that direction is both the cost per bit and the scale at which we could operate the system and manage it was better running it over native Ethernet. We expect 10G Ethernet over fiber to come and to be very cost effective.
Rouleau: Actually, it's quite synergistic. We have been deploying GigE services for a while, and we have number of customers. We think about GigE as blade level integration into what we are doing for our optical networks anyway. So when we deploy a multiservice access device, it has GigE or 10/100 as an interface. We groom that right into the SONET infrastructure we are deploying. We think of Ethernet as another type of transport so it's the full 10 Mbps, 100 Mpbs. We sell a half a gigabit solution and a full gigabit solution that can be point-to-point or point-to-multipoint. We have not introduced switching yet, but we will, where we can start to control the level of bandwidth on an extranet application.
I can put a card into a Cisco 15454 service multiplexer that gives me a dozen 10/100 ports very cost effectively, because I already have that equipment there and am already serving customers' private line TDM needs and their voice needs.
Wimmer: It's having an impact in terms of things that we will be rolling out going forward, not in terms of capacity demand. It hasn't become a big capacity hitter yet. It will start probably initially in the metro space, because GigE is not a long-haul transport technology.
However, 10 GigE potentially will change that when it becomes a standard later this year, because the 10 GigE standard incorporates many of the management hooks that SONET has in it today. So I will have tools available to me to do fault management, performance management and troubleshooting kind of things in a live network. So GigE has the potential to compete with OC-192 for some data transport needs.
BBW: What do carriers most need from vendors to make building optical networks easier or more cost effective?
Coghill: Quite frankly what we need from all the vendors for any network are the OSSs that allow us to provision, manage inventory and manage the service layer. They all say they have that, but quite often it's for their own specific product, but you know in these networks we use a number of products. We can connect them together to create a transmission path, but we can't always connect them together through an OSS element manager. And that is essential, especially when you start talking about VPNs and the ability for customers to do provisioning or network reconfiguration on their own.
I think most vendors are starting to put more focus on software control and service environment, given that that's really where we can start to reduce our costs. Part of the operational efficiencies we would hope to gain in the future through using new technology would be to be able to have fewer people and more expert systems that can scale to grow with our business.
Lawrence: The thing that I most want from vendors is for them to focus on building a system that is optimized to provide the best unit cost. Too often vendors are making compromises in their system design to accommodate fiber constraints, old fiber types and edge applications, as opposed to designing a system that brings silicon economics to a majority of customers and then worrying about edge applications later.
The other functionality that we are looking for is more intelligence, better remote inventory and control. We are very much interested in systems that have GMPLS capabilities and automated switching.
Wimmer: Optical cross connects are on the top of the list. Also there are some tools that we need to do performance management and operational things. These all-optical networks will happen and work well, but we have taken away the SONET tools, so we need something to replace them. I expect there will be a lot of focus on that SUPERCOMM this year. There are a lot of vendors cooking things.
Longer term, as these things grow and scale, and they will, then you need to be able to think about automating management of them and you start thinking about signaling and control layers. There are several industry standards activities geared toward defining a signaling and control protocol and network that would allow you to communicate between the optical transport layer and the data layer and the back office systems it takes to run all these things. That's not an immediate priority, but having the ability to automate these things over time is going to become very, very important.
Afferton: Let me use DWDM as an example. I get asked a lot about what I think about ultra-long haul, or Raman or 40G. I come back to the fact that we are not particularly obsessed with any one specific technology, what we are obsessed with is ultra-cheap.
When I talk to start-up vendors and they ask me, 'What are the opportunities, what are you looking for?' I say, 'I want to see some compelling unit cost savings.' Now certainly operational savings are part of that, but so are straightforward capital costs. As I go to SUPERCOMM, in the DWDM space, I'm going to be looking for vendors that have a tangible technology differentiation that says this will allow you to continue to ride that unit cost curve down by significant improvements over the systems we are planning to roll out.
BBW: How has the economic slowdown affected plans for rolling out optical networks?
Rouleau: We have tried to stick to our business plan and I think have done, if you look at our first quarter results, a pretty good job at managing our business. We have something that's unique. We have more buildings on-net and more fiber in the ground than most other providers, and as a result, we can deliver services pretty quickly to end users. So our service levels are our differentiator and our execution is a differentiator.
We have always have managed our cap ex spend to profitability. We have a sales-costing tool that we deploy to all of our sales people where we look at all the cost to build out to a building or deliver a service. And if we don't generate a 30 percent return over the life of the agreement with no residual value, then we don't deploy, we walk away from the business. There are instances where we have taken lesser because the building is a good opportunity, but we do not build speculatively to buildings. We always make sure we have revenue coming out of that building before we build into it.
Afferton: We can't talk specifically about our capital plans or any numbers, but we are still building our network. We still have customer demand that is growing, and we still have a need to continue to grow the capacity of our network. So, we certainly haven't stopped. As I mentioned, we have an increased emphasis on unit cost. We want to have efficient growth of the network where we continue to maintain best in class unit cost.
Because my job is to work with new vendors, the changing economy is affecting the stability of vendor space I see out there. I need to ask them how much funding they have and when it's going to run out. Beforehand, funding was never an issue. So certainly on the supply side we are seeing a change. However, my door continues to be open to startups. I'm sure that as much time will be spent at SUPERCOMM on the floor as it will in private suites with startup companies.
People are talking about a shakeout and reduced opportunity for selling out there. There are certainly going to be some vendors that don't have substance that will fall by the wayside. But there are start-up vendors out there that have good solutions, innovative technology and a solid team and we are certainly going to continue to talk to them.
Wimmer: Technically, we are still rolling along. The economy certainly hasn't caused us to speed things up. If we look at the number of packets flowing through the network, we really haven't seen a downturn in traffic. We still see a very healthy growth of data traffic going through the network, which says the network still needs to grow. The issue is using up the capacity that's been invested and not getting it out there too far ahead. Maybe there's a lag here and we haven't really seen the economy's effect on traffic yet, I don't know. But technically our plans haven't changed: it's just a matter of how rapidly we invest.
You know, from one perspective, if it slows down a little bit, maybe that's not bad. It gives us a chance to get a little more organized. I mean that seriously. If the world were growing at 1000 percent per year then I might be forced to put some boxes out there that have a short life cycle. So from that view, it's probably not a bad thing. It will shake out some of the suppliers and we will find out who the strong suppliers are and who's not. That's kind of sad, but it's true.
|
