The New Order in Networking
By Tom Nolle, Network Magazine
Jul 10, 2000 (12:00 AM)
URL: networkmagazine.com
It probably seems to most readers that the industry has been talking about 21st-century networking for at least a century already. The buzz is particularly frustrating because everyone knows that the year 2000 isn't a neat boundary beyond which all manner of technology changes will explode into being.
However, we really are going to have 21st-century networking in this century, and it's really going to be different-very different, in fact-from the networking we've known. It's also going to be different from the kind of networking we thought we were going to get.
There are enough objective facts in the networking business and technology space to let me predict, with some confidence, what is going to happen over the next decade, and how those events will affect service providers, service consumers, and equipment vendors. I can draw a roadmap with fairly certain mileposts, and I can identify some areas where there are clear issues to be dealt with-issues that will change some of the details of the future of this industry.
Predictions entertain most people, gladden a few, and infuriate a few too. I hope these predictions will also make all those people thoughtful.
THE DRIVERS
Networking is a supply-and-demand business, and both supply and demand are changing quickly in response to trends that emerged in the 1990s-most notably the Internet. In addition, the regulatory framework for networking is changing, and any highly regulated business is extremely sensitive to even minor shifts in governmental rules and policies. The interplay of these factors sets the direction that networks will take; they create three legs that support the “stool” of our network of the future.
The big story in regard to demand is, of course, the Internet. Starting in the early 1990s, the popularization of the Internet created a new industry, caused major changes in the nature of retail sales and fulfillment, spawned a microculture in our society, and generated more exaggeration than perhaps anything technology has ever produced.
In a global traffic-and-dollars sense, the Internet isn't big stuff. Almost none of the ISPs are profitable even now, and the total revenue for Internet access and transport alone (excluding hosting, consulting, and so on) amounts to less than 1 percent of service-provider revenue worldwide.
Despite this, the Internet has contributed the most important piece of the puzzle of future networking: It has defined the service model for public data networking. We get voice through RJ-11 dial tone and touch-tone keypads. We get public data through a service demarcation that looks like a gateway to the Internet. By promoting Internet-specific information appliances on all our desktops, the Internet has created a community of users that all public data carriers in the future will have to sell to. Connectionless IP user interfaces win as the application interfaces of the future, for both corporate and personal use. That's the first leg of our stool.
It's also clear that the infrastructure of 21st-century networking is fiber optics. Fiber optics as a communications phenomenon that began in the early 1980s with the deliberations that resulted in SONET standardization. Dense Wavelength Division Multiplexing (DWDM), the new-generation optics, expands the capacity of fiber strands enormously, even as more and more companies are laying new fiber.
A DWDM-based network with hundreds of wavelengths (or lambdas, as vendors sometimes call them) per fiber will let network builders create rich infrastructures of almost unlimited flexibility. When coupled with technology to switch traffic from one lambda to another as it moves from fiber strand to fiber strand (as Sycamore Networks, Cisco Systems, Lucent Technologies, and Nortel Networks now claim to have), DWDM lets carriers build networks without traditional switches or routers in the core at all. These networks are effectively featureless pools of bandwidth waiting to be exploited. In short, DWDM wins at the core. That's the second leg of the networking stool.
However, how do we use that featureless bandwidth pool, particularly given just how much bandwidth we're talking about? A single fiber with 512 wavelengths, each one operating at OC-48 (achievable today, at least in the lab), could carry a total of about one and a quarter terabits per second-nearly the total of all today's voice and data traffic combined. The core network of even five years from now will be able to carry perhaps ten thousand times the current traffic volume.
What would this do to carrier revenue streams? If supply explodes in the absence of comparable demand, the result would be a price disaster, driving service providers bankrupt and ending public communications stability worldwide.
However, that won't happen for two reasons. The first is because falling revenues would starve the carriers for money to expand, thus halting the build-out of networking. Second, the Telecom Act of 1996 ensures that network progress won't come to a screeching halt.
The Telecom Act of 1996 is the regulatory third leg to the stool that represents the stable network of the future. Regulators who constructed the Telecom Act realized that the access network-the last mile-was a constraint to the development of new applications. However, new applications would justify all that fiber bandwidth and generate the revenue needed to offset falling prices and profits on today's limited-bandwidth services like voice (which, worldwide, generates 88 percent of service provider revenues).
Thus, the regulators devised a complicated formula that would enable the Incumbent Local Exchange Carriers (ILECs, primarily the RBOCs) to become national competitors in the future data market (most of which is long distance). They balanced that gift with a requirement that these ILECs offer competitors wholesale access to their local infrastructure.
Some will argue that the role of the ILECs, particularly the RBOCs, is declining. TV commercials have denigrated these carriers as stodgy and conservative-even fossilized. The truth is, they're highly successful businesses, and their primary role in network access today cannot be doubted. That they would dominate future access is not surprising. The FCC recognized that dominance, and recognized that only by encouraging the RBOCs to modernize would the FCC achieve the goal of universal broadband access.
Like any compromise where billions of dollars are at stake, the Telecom Act has been tested and redefined through the legal maneuverings of all players. The regulatory dancing finally seems to have come to an end, though. In November 1999, the FCC issued a formal ruling (technically, an “order”) that will finally jump-start the deployment of broadband digital service in the local access network, and thus create the foundation for high-bandwidth application growth in the future.
The FCC formula is simple: To the extent that ILECs build out a new digital infrastructure to serve the public goal of universal broadband networking, the government will shield that new infrastructure from the Telecom Act requirement that the ILECs unbundle the elements of their networks and offer them wholesale to competitors.
Because of this ruling, the RBOCs have all developed plans for the access network of the future, which is the first concrete piece of our 21st-century infrastructure. The technology winner in the access space, by a landslide, is ATM, and so it's with ATM access that our story of the future network begins.
ARCHITECTURE OF THE FUTURE
To provide the kind of regulatory shielding the RBOCs want, the future local access network will push ATM equipment outward from the Central Office (CO) to about 500 feet from residential and business users. The RBOCs will do this using a modernized set of fiber-based remote concentrators.
Remote concentrators aren't new-most new residential services are already provisioned using these devices. What's new is that the concentrators will be based on ATM, and will provide a mixture of traditional voice and advanced data services. The exact nature of the deployment of these new devices may vary from RBOC to RBOC; in this section, I'll use SBC Communications' (www.sbc.com) Proj-ect Pronto as a model because the most details are available about it.
As shown in Figure 1, users' fiber and copper loops will be connected to the remote concentrators (called “neighborhood gateways” by SBC), which will then be connected back to ATM switches in each CO. These switches will be linked via ATM trunks to other COs, as well as to the service equipment that provides voice and data services to users.
This infrastructure will provide analog voice (the staple of today's market), as well as Digital Subscriber Line (DSL). Eventually, most users will be connected this way, except for those so close to their COs that direct copper wire makes more sense.
To offer voice service, the fiber remotes will monitor the copper loops connected to them for an indication that an old-fashioned “off-hook” event has occurred (called “loop start” or “ground start” condition in telephony). When that happens, a small digital encoder will convert the analog signal from the phone into digital bits, then into ATM cells.
The result will flow over an ATM Permanent Virtual Circuit (PVC) to a kind of ATM-to-Time Division Multiplexing (TDM) gateway product, which will then connect it to a Class 5 voice switch. This gateway will eventually be displaced, as the big Class 5 vendors like Alcatel, Ericsson, Lucent, Nortel, and Siemens provide ATM customer interfaces for their switches.
The Class 5 switches will also use the access carrier's ATM network to move calls from office to office, using the ATM network as a “virtual tandem switch.” This eliminates today's costly practice of using voice switches in hierarchical networks to create local call paths.
Another gateway product, or perhaps an existing voice switch, will link this virtual tandem network to the Interexchange Carriers' (IXCs) toll networks, a role played by Class 4 switches today. If a toll carrier wants to use an IP infrastructure for long distance calling, this toll switch gateway can be a Voice over IP (VoIP) media gateway and controller.
When a customer wants a high-speed data service, the same copper loop can be used to carry a digital bitstream (again formatted as ATM cells). A splitter at each end of the loop separates this digital data from analog information in order to prevent problems with existing phones and switches. The digital portion will appear as an Asymmetric DSL (ADSL) service, and the data traffic will flow over a separate ATM virtual circuit to the CO ATM switch, and from there to a data service gateway device (a Data Service Point of Presence, or DSPOP).
The DSPOP integrates the core network with the new ATM-access edge. If we assume that core networks are protocol-specific and contain a large number of electrical devices like routers and voice switches, the DSPOP will be a form of access concentrator, not unlike an ATM-based version of a Class 5 switch or an access router.
DWDM seems to be foreclosing that kind of core, however. New versions of DWDM offer low-cost “wavelength hopping” from one fiber to another, eliminating the use of electrical switching/routing in the core. The core network-once a mixture of switches/routers and fiber strands-is becoming almost a pure fiber network. Electrical devices will be confined to the edge of this network, where traffic levels are too low to justify dedicated wavelengths.
However, how deep an electrical layer can we expect to have as broadband access multiplies? Because a DSPOP collects the traffic of one or more CO's worth of users, it supports thousands or tens of thousands of customers. Traffic volume here is thus likely to justify a direct connection between the DSPOP and the optical core.
Thus, the combination of fiber optics progress and ATM-access network traffic concentration tends to squeeze the electrical components of the carrier network into a thin band between access handoff and core.
To accommodate the core network's shift from electro-optical to pure optical, many vendors are using Multiprotocol Label Switching (MPLS) to create a virtual transport network to connect their service-layer devices, as Figure 1 and Figure 2 show.
Because MPLS label-switched paths behave like circuits, and because they can be created over ATM, fiber, or IP core networks, they help shield service-layer feature software from changes in the underlying network devices. Because MPLS happens to be useful in that consolidation mission today, and because MPLS can carry any form of future traffic, it's likely that core networks will migrate to an optical/MPLS structure before the end of this decade.
The high traffic concentration at the DSPOP also means that the DSPOP owner (a service carrier, to create a name) will have enough clout to choose from among the growing number of specialized fiber network players (AT&T, WorldCom, Sprint, Qwest, Level 3, Williams, and more), or to connect to several of them to draw various services and features for sale to the end user.
This creates the network architecture shown in Figure 2, where a “stack” of core network options is combined with a series of DSPOPs to create a variety of voice/data/video services to sell to the user. Each service rides an ATM virtual circuit to a device on the customer premises (a “premises service agent”), where the carrier and the user join to manage and monitor their various service relationships.
The combination of premises service agent Customer Premises Equipment (CPE) and SPOP can be expected to develop for every type of service. In fact, an ILEC's voice service architecture can be viewed this way; the Class 5 switch (and the gateway, if used) is the Voice SPOP (VSPOP), and the phone is the premises service agent. While these two devices could behave similarly to traditional devices today, they could also be combined to create other types of service relationships in both the voice and data space.
Figure 3 shows both traditional voice service and a new-generation voice architecture, both based on the CPE-VSPOP model. In both cases, the user signals for service using the instrument of choice-a phone, in this example. From this point, however, the models are increasingly different.
Traditional voice, today and in the future, links the signaling device directly to the Class 5 switch, which presents it with dial tone; collects the dialed digits; and interprets the user's request for service as a local call, long distance call, special feature request, and so on. That the future network can work as the present one does means that phone service won't be disrupted, and that users and carriers can retain most existing equipment.
New-generation voice introduces a service agent device between the user instrument and the traditional voice environment. The service agent could be in the CPE, as previously noted, or installed as a module in a VSPOP device.
In this new configuration, the user's service request doesn't go directly to the Class 5 switch, but rather to the service agent. The service agent first determines the nature of the request and then either routes it to a special feature-fulfillment network, or to the conventional voice network. Thus, new and exciting voice features can be created without disrupting the equipment, practices, or stability of the current voice network.
This service approach potentially opens an advantage to Competitive Local Exchange Carriers (CLECs). Though the CLECs can't get access to ILEC network elements as easily (if at all), they can still wholesale the services of the incumbent network, meaning the ATM virtual circuit connections to the customer. With a VSPOP, a service agent device, and some wholesale agreements, a CLEC could use its own equipment to create special custom-calling features such as unified messaging and hand off the low-value local calls to the incumbent!
With normal voice services and DSL access infrastructure, a CLEC is hard-pressed to maintain a 20 percent return. However, premium custom-calling and wholesaled ATM customer connections could create returns as high as 80 percent.
A similar architecture based on enhancing the baseline service already consumed by the customer can be applied to the data space, where the “baseline service” is the Internet.
As shown in Figure 4, the data service carrier would start its business with an agreement to wholesale Internet connections from one of the major ISPs (as most small ISPs already do today). This default Internet service would be connected to a DSPOP located in each of the areas where the data service carrier desired to offer service. From the DSPOP, virtual circuits would be wholesaled to each customer location.
To provide service differentiation, the service carrier would then acquire a set of alternate core network services from a long-haul carrier (Level 3, Qwest, Williams, or any IXC or value-added network provider). These alternate core services might take the form of a QoS-capable routed network, an ATM Switched Virtual Circuit (SVC) network, an MPLS network, or virtually any other kind of network.
When a customer requires a kind of IP service behavior (secure transport, high QoS, and so on) not available on the default Internet path, the service carrier would introduce a traffic forwarding rule into the DSPOP. The DSPOP would divide the customer's traffic stream to route the application flows designated for this special service to the alternate network, while routing the rest to the Internet. The forwarding rules and core network service connections would be controlled by a new service management system, the VPN Operations Support System (VPNOSS). For more information on the VPNOSS, see The Phone Bill of the Future, January 2000.
One example of how this type of architecture could be applied is a user VPN, where the company had adopted one of the “private” IP addresses sanctioned under RFC 1918. Many companies have used these addresses (particularly the Class A address 10.x.x.x) on their own IP networks, not caring that the addresses can't be routed on the Internet because they wouldn't be unique to a user.
With a forwarding rule to divert addresses in this range to a “virtual router” unique to the individual user, the DSPOP-equipped service carrier could create a public IP service that seamlessly blends the user's VPN and Internet service. This service would also maintain secure barriers between the two by preventing Internet addresses from being directly routed to or from the private address space. (Of course, this can be accomplished in other ways.)
This “Internet-plus” behavior could create an IP VPN that replaces part or all of a private line or frame relay network. It could also create a short-duration special path to expedite e-commerce transactions, or to create NetMeeting-style collaborative relationships. Because public IP services of all types are created as what appear to be “handling enhancements” to the Internet, current buyer equipment and practices would be compatible with the new services, facilitating their introduction and adoption.
The networks of today are built from service-aware hardware: Voice networks are built from voice switches, and IP networks are built from routers, and so on. Providing user service is simply a matter of connecting users to the network of their choice; that's the dilemma that multi-service networks pose. Given that the information appliances we have today- phones, faxes, and PCs-don't support a common protocol, any convergence dictates either the displacement of existing devices or universal adoption of conversion devices to link the consuming appliances to the network.
This network model solves that problem. In our future architecture, the access network migrates almost entirely to ATM. The core network is free to adopt any protocol or architecture it chooses, but will become increasingly dominated by pure optical routing as DWDM develops. As such, it will become service-independent as well.
This leaves premises service agents and SPOPs that form the access-to-core boundary as the only places where service intelligence and features can be created. It's a new approach to network-building, and it's inevitable.
NEW NETWORK WINNERS
So, who wins in the new network? In the carrier space, the service carriers do because services are the high-margin profit source that drive everyone's business engine. The best a carrier could ever hope to obtain in revenues from broadband access alone wouldn't amount to more than a third of current Local Exchange Carrier (LEC) revenues.
Except for the ILECs, which have to play a subsidiary game for regulatory reasons, no facility carrier can ultimately avoid becoming a service carrier. The ILECs are already planning their independent advanced services subsidiaries (SBC's plans are included in its filing for a merger with Ameritech, which was approved in fall 1999). Those CLECs that survive will certainly adopt the service-carrier role.
In the race to supply hardware for the new RBOC ATM-access network, it appears that Alcatel has a commanding lead, with Marconi in second place. Both companies have leveraged their experience with ATM access gained in the European market to steal the march on domestic leaders like Cisco, Lucent, and Nortel.
The Alcatel acquisition of Newbridge will give Alcatel a credible ATM switch to match with its ATM fiber remotes, giving the company a shot at the whole RBOC ATM equipment space-a space that could be worth as much as $35 billion over the next five years.
In the core network, there's something to smile about for everyone. Cisco's dominance in the IP service space would be threatened if the incumbent facility carriers-the RBOCs and IXCs-created a consolidated network using only one technology, which would almost certainly be ATM.
However, core networks will probably not follow the technology lead taken by the access networks. Future traffic growth will be largely IP-based, so future core networks will probably be dominated by IP as well. The question is whether IP alone can provide the range of service quality control and security that buyers want. Many observers think that a combination of IP and ATM, or IP and MPLS, will be required.
For Cisco, that could be a big win because the company has probably the best track record with facility carriers in deploying ATM/IP and MPLS networks. AT&T's business-based IP network uses Cisco MGX switches, and it was the basis for a big AT&T roadshow in spring 2000-a roadshow that showcased virtually the same kind of network architecture I've discussed here.
For Lucent and Nortel, this new-age architecture model could add up to real challenges. Both companies had probably hoped their incumbent positions with facility-based carriers like the RBOCs would carry them into both the access network and the core network. In the new-age architecture outlined here, however, both players lose (at least for now) to Alcatel and Marconi in the access network, and find themselves facing Cisco for the IP-dominated core networks-which will be most core networks.
For all the major equipment vendors, the key factor will be the extent to which the new service-agent-and-SPOP architecture for service creation can be made to look like an evolution of the existing IP and voice equipment architectures.
The truth is that none of the incumbent players has an ideal approach for service-building in this future network, because all of them proposed architectures that created services and network connections from a common base of equipment. With access and deep-core networks going to a service-independent mode, none of these architectures is the ideal solution for the network of the future.
In the service layer of this new network, it may be the start-ups that win. There are at least a dozen new companies targeting the premises service agent and SPOP spaces; some will even target both. A half-dozen more are feverishly developing the service software to create, manage, and bill the new services.
One reason for the start-ups' interest in these new spaces: While most big carriers won't buy infrastructure products from a start-up, most are willing to buy service-layer technology from such a firm. Cascade Communications, the market Cinderella story of the networking age, was propelled by its success with the ILECs. Other start-ups are hoping history will repeat itself. That might not be necessary, of course. All the major equipment vendors have already shown their willingness to buy up small, innovative firms. It's likely that the dawn of the new-age network will also be the dawn of a second round of acquisitions.
THE ISSUES
Who wins and who loses in a vendor sense will be of great interest to some (particularly vendor employees and investors), but of relatively little interest to others-in the long run. Similarly, the fate of the CLECs, the ISPs, and the ILECs will not affect the service choices of most users-in the long run. However, this doesn't eliminate other issues, most of which relate to how services will evolve in the near term.
At the technical level, the big questions arise from the nature of the RBOCs' ATM-access networks. Their ATM-access networks won't be “real” ATM networks at all; rather, they'll be subsets of ATM designed specifically to support an evolution from a business model dominated by the sale of voice connections to one with a much broader base in the data world. In these ATM networks, customers will probably not be able to call one another directly at the ATM level; they'll have to work through SPOPs.
Similarly, early plans call for the use of ATM's Unspecified Bit Rate (UBR) service, which offers no QoS guarantees, for all ATM-access connections other than those targeted at standard telephone voice. This would effectively prevent competitors from acquiring data-grade connections and leveraging them to undermine the ILECs' voice business.
Finally, current disclosures suggest that at least residential consumers would have access to only one data connection at a time, despite the fact that ATM could theoretically support dozens of simultaneous connections. This would favor full-service players over incremental (meaning smaller) carriers.
The idea that ATM networks wouldn't provide full ATM service isn't something vendors (or users) have contemplated, yet this will clearly hold true. Eventually, the RBOCs may increase the level of ATM features they expose, but the timing of that process isn't predictable. Thus, “access ATM” for now has to be considered separate from “standard ATM,” and products and planning must reflect this idea.
A second technical issue is the impact of this new infrastructure on MPLS. Today, MPLS is driven by the (let's face it) opportunistic interests of the IP and ATM communities. In the future, it's likely to displace much of both, so it's not clear whether the mission of MPLS over the long term will be supported by the IETF standardization process. The recently created MPLS Forum may be the answer to this, but it's too early to say.
The paramount question the new network poses, though, may be at the public policy level. Policy should balance public interest in universal broadband services with public interest in competition (and possibly lower prices) for current service business. The carrier industry in the United States is a $200 billion market, and there are many players, both inside this market and outside it, who would like a large share of that service revenue stream.
Since we don't know how much revenue would arise from new broadband-based services, we don't know how much revenue loss in the current service areas the carriers would risk as a price of modernization. Giving twenty million homes DSL at $250 per year (not counting Internet connection costs) would add only $5 billion to the service pie. Asking the LECs to risk perhaps ten times that much in revenue loss from current services is clearly asking a public stock company to do something irresponsible. Many of the limitations on RBOCs' “access ATM” stem from their need to protect their service rate base.
Is this sleazy (as some have already suggested)? No, it's the price we're going to have to pay for a painless migration to a completely new service provider business model. If policy-makers permit competitors to use the new broadband access network against those that must build it, no new access network will be built. Without it, no real change in the nature of our services is possible.
Most users would readily agree that the trade-offs the FCC and the ILECs have entered into to secure widespread broadband services will be worth the price. Some competitors won't agree, of course. There will surely be regulatory wrangling in the future, and there may also be a new wave of public debate over the nature of the trades the FCC has, and will, make.
None of this will be very productive, and the real danger is that it will disguise the far more complicated and important issues, like those listed, that relate to the details of the new network we'll all share.
We are about to see a complete modernization of the access network, and a complete shift in the service provider business model. That something this sweeping would produce an entirely different infrastructure shouldn't surprise us. That not everyone will believe in these changes isn't surprising, either-but it doesn't make them any less inevitable.
We're finally going to get the kind of radical service changes that have been written about for a decade. For some of us-as with most wish fulfillment-we won't get exactly what we want. It's time to face that truth, and then face the new network future. |
