The Computerization of Public Networks: an Opportunity and

a Challenge for the European Telecom Industry

Preface

E-mail: dbcons@dbcons.it

Table of Contents

Executive Summary *

Telephone and data networks have existed side by side for decades. Although pure play telephony still contributes the vast majority of revenues for most TNOs, data and now multimedia traffic is growing much more rapidly and nowadays, with the development of the Internet, computer networking technology is in the process of penetrating circuit-switched network infrastructures at a rapid pace.

The explosive growth of data traffic is not the only reason for this convergence. Another very important consideration is that the Internet is also set to fundamentally change telephony in two key areas:

• as an alternative backbone for voice traffic, providing significant cost advantages over traditional circuit-switched telephony;

• to control and monitor telephone calls, e.g., with the emergence of "virtual call centers".

How well other service providers move from here will dictate how well they begin the 21^st century. There is compounded evidence however that most established Western TNOs, not only newcomers that can build infrastructures from scratch, are acknowledging this trend and taking steps to augment/modernize their existing infrastructures. So the question there is likely to be more one of pace of implementation rather than one of whether this will happen or not.

North American innovation didn't come that much from established players such as Lucent or Nortel though, but from startups (neither Cisco - whose market capitalization is about twice that of Alcatel although it is more 5 times smaller when measured in terms of sales or headcount -, Bay Networks, or any other of their likes even existed 15 years ago. Now it is them, rather than Lucent or Nortel, along with a raft of fledgling startups, who are creating many of the products that advanced networks do need, … and the jobs that go along with their rapid development.

That Europe doesn't provide the same fertile ground for the emergence of technology firms is well known but, in this case, there seems to be another missing ingredient which is that we've lacked the proper technology culture. Traditionally telecommunications, somewhat akin the computer industry of the earlier days, has pursued a top down vision of how to develop/deploy centralized infrastructures from powerful central switches (conversely mainframes) down to low functionality telephone sets (or dumb terminals). Data networking has evolved the other way around from the need to interconnect stand alone PCs, then local area networks to each other, culminating in the global networking environments that are emerging today to handle not only data, but video, multimedia, … and pure play telephone calls.

This technology culture is not broadly spread in Europe and it will be virtually impossible for suppliers (incumbents or new entrants) to fill the gap with the US entirely. But examples such as that of Israel, a country with hardly more than 5 million citizens, show that joining the band wagon in some ways is not mission impossible (see Annex I). Now that it plans to merge its IT and telecommunications related research programmes, the EC may have a valuable role to play in this perspective.

ICT technologies will keep playing a key role in the ongoing development of advanced telecommunications products and services. To abandon most if not all of computer networking related technical developments to the US may have devastating long term impact on the European telecommunications industry, hence on competitiveness and employment.

• Computer networking related R&D and its application to global public networks should be included in the 5^th Framework programme;

• Cooperation within individual projects between partners with computer and telecommunication backgrounds should be encouraged;

• Involvement from young innovative firms should be pursued systematically;

• Participation from partners from outside the EU should be encouraged on a selective basis in order to facilitate the infusion of technology know how.

Telephone and data networks have existed side by side for decades. Although pure play telephony still contributes the vast majority of revenues for most TNOs, data and now multimedia traffic is growing much more rapidly and nowadays, with the development of the Internet, computer networking technology is in the process of penetrating circuit-switched network infrastructures at a rapid pace.

The explosive growth of data traffic is not the only reason for this convergence. Another very important consideration is that the Internet is also set to fundamentally change telephony in two key areas:

• As an alternative backbone for voice traffic, providing significant cost advantages over traditional circuit-switched telephony

• To control and monitor telephone calls, e.g., with the emergence of "virtual call centers".

Such a statement may have sounded provocative at best only 12 months ago but the vision which was then heralded by a few harbingers such as MCI or Worldcom is now being endorsed by no less than ATT (which has started to provide voice over IP services), Deutsche Telekom (which took a significant position in Vocaltec, an Israeli startup and early proponent of the technology, in addition to unveiling a plan to deploy voice services over IP), and most recently by France Telecom, in a somewhat surprising but unequivocal change of attitude.

How MCI has dealt with this voice/data convergence over the last two years, and what implications this will have for suppliers of related equipment and software products to TNOs provides a telling example of things to come, as outlined below.

At ComNet 1997 in Las Vegas, Vinton Cerf, MCI's senior vice president of data architecture and an original builder of the Internet, discussed the "vault" technology that will tie together MCI's voice and packet-switched networks. "Customers won't know whether their services are using voice or packet networks", he said. "We're taking each network canvas apart and reweaving them together thread by thread for a whole new canvas on which to paint products and services".

Among its advantages, MCI's vault technology will let voice and IP networks to use the same databases, including those on MCI's intelligent network. In a videoconference with a customer account representative, for example, the voice network would carry the speech while the IP network would carry video and documents. Database dips could be made to determine call setup and search out specific customer data, including credit card validation for service orders or purchases. Vault technology is still in its infancy. But the plan indicates how the company envisions its Internet business development and should give its slower competitors reason to accelerate their own efforts.

MCI has offered commercial Internet services since December 1994, though those first efforts - plagued by high prices and delayed software delivery - did not get off to a flying start. Yet the company consistently kept its eye on the larger goal. Since its launch of Internet services, MCI upgraded its backbone twice, first from DS-3 OC-3 (155Mb/s) and last year to OC-12 (622Mb/s). This latest upgrade was part of a $100-million infrastructure infusion in 1996 to prepare for future IP-based business. Another 1996 initiative was globalization. The final major strategic move in 1996 was to identify three Internet technology partners - Microsoft, Digital Equipment Corp. and Intel. 1997 themes included intranet, globalization, Web-hosting service and advancement of IP to the desktop. Each of the partnerships is linked to these themes, which as a group drive value into Internet service for business.

Through the distribution of Microsoft's Internet Explorer browser software and a customized version of the Microsoft Network on-line service, MCI is looking to solidify a desktop Internet strategy that would port many personal communications applications - fax, paging, remote access and the like - onto the Internet. To gird its Web-hosting business, MCI will use DEC's Alpha 2100 server.

The Intel alliance involves the creation of Web-based marketing and multimedia applications. The first joint offering, MCI Webmaker, is an all-in-one package for creating and managing Web sites. Although this will let MCI showcase some dazzling multimedia capabilities of the Internet, the agreement goes beyond mere product development and focuses on another critical yet undeveloped area on Internet business offerings: quality-services measuring.

This is where MCI blurs the line between computing and the network. Through use of Intel's new MMX chips and hardware from Cisco Systems, MCI and Intel are developing services that will combine intranet and remote access-to-LAN offerings, both with guaranteed bandwidth and latency for multimedia applications. Key to the services are standard IPs such as IP multicast, resource reservation protocol and real-time transport protocol that combine high-end personal computer capabilities with high-speed network transport.

The new services were rolled out in June 1996 in a phased multimedia Internet trial with a distributed laboratory connecting three major U.S. cities: Portland, Oregon; Richardson, Texas, and San Jose, California. The trial has since expanded to include a distance-learning application. Other potential applications could include recording and playback of real-time videoconferences, 3-D graphical presentations, real-time audio and video streaming, stereo sound over the Internet, corporate training on demand and video e-mail.

Also building Intranet and Extranet practices are Internet Service Providers (ISPs). Although U.S. ISPs now number between 4,000 and 6,000, only a handful seem prepared to tackle service provision on such a high level. Some names repeatedly mentioned though are UUNet Technologies, IBM Global Network, BBN Planet and Netcom.

UUNet exemplifies these companies' growing ambitions. What sets it apart is that it is now one unit of a tripartite organization rounded out by WorldCom as parent and MFS as a sister subsidiary. UUNet launched its own extranet service by late 1996. The service promises access via MFS at speeds up to 45Mb/s, end-to-end encryption and premises-to-premises quality-of-service guarantees. In February 1997, the ISP announced plans to invest $300 million in its network, including a backbone upgrade to OC-12.

As corporations begin to expand into electronic commerce and EDI, and employees at remote locations have increasing needs to communicate with each other, it will be increasingly difficult and expensive to bring everyone onto wide area networks (WANs). The new service is expected to bring the economy of the Internet into the world of the corporate LAN/WAN. The market still questions ISPs' capabilities to deliver top-notch reliability however, especially when critical data is run over the Internet, but with the planned merger between WorldCom and MCI, UUNet and MFS truly have all the pieces that let them bypass the local exchange.

How well other service providers move from here will dictate how well they begin the 21^st century. There is compounded evidence however that most established Western TNOs, not only newcomers that can build infrastructures from scratch, are acknowledging this trend and taking steps to augment/modernize their existing infrastructures. So the question there is likely to be more one of pace of implementation rather than one of whether this will happen or not.

The more serious concern has to do with the challenge that his evolution poses for the European telecom industry. To put it in blunt terms, the market for related bridges, routers, LAN/WAN switches and other hardware/software products that represent an increasingly large percentage of TNO procurements is dominated by US players with little if any European indigenous competition to speak of. Over the last few years, the European telecommunications industry has focused on ATM-based or ATM-related technology research with little attention paid to other emerging technology approaches that have developed from the need to interconnect PCs and other computer systems, locally in the first place, then within wide areas and now globally. Major European firms like Alcatel or Siemens has signed commercial/technical cooperation agreements with Cisco and 3Com respectively, and Nokia has recently acquired Ipsilon, but this doesn't make up for a strong local industry that will create jobs and contribute to Europe's competitiveness.

True, the telecommunications equipment industry differs from the computer industry in several major ways. For one thing Europe has long enjoyed a strong presence both as a supplier to its own market and as a major exporter; the recent success of GSM also provides evidence that it retains the ability to develop and deploy winning advanced technologies. Also legacy of earlier monopolies and the existence of wired infrastructures that represent enormous investments, in particular at the level of local loops tends to slow down the pace of change that the computer industry has experienced. The fact remains that networking technologies have already started to make a significant dent into telecommunications infrastructures and have the potential to do to traditional telecommunications equipment what minis did to mainframes and what PCs did to both. This is not a concern that can be overlooked easily by policy makers when tens of thousands of jobs are at stake.

As outlined in the introduction of this document, the current evolution of telecommunications infrastructures is being driven by tow main factors:

• first the need to deal with exploding volumes of data that are being exchanged between computers systems that benefit from permanent or dial-up connections across LAN/WAN and global infrastructures, including intra and extranets;

• second, the realization that these very same infrastructures, can support voice traffic very efficiently (from "Etherphones" to "Voice over IP") with the additional benefit of providing higher levels of functionality for integrated voice/data applications.

The following sections of this chapter outline the main trends/requirements in relationship with these two factors.

One of the biggest changes to the way the industry uses IT has been the huge up-take of intranets over the last 18 months. Access to the intranet is provided by easy-to-use, low-cost, multiple platform browser software installed on the user's desktop or laptop PC. It enables access to information drawn from a wide range of sources, such as corporate databases and multimedia documents. Organizations set up intranets primarily for employees, but can extend them to business partners and customers with appropriate security clearance - thereby creating extranets. Users can also access information from intranets on mobile devices, such as GSM phones.

Any intranet requires the following:

• corporate network: connects computers and allows them to exchange data. A high bandwidth network supports higher levels of traffic, providing the capacity necessary to support the transmission of multiple media such as voice and video;

• firewall: a software or hardware security gateway that acts as a filter between two networks, and which only allows certain traffic through;

• hypertext Transfer Protocol (HTTP): a standard procedure for communication between web browsers (clients) and web servers;

• personal computer or network computer: the hardware device used to support a web browser, and to display information returned from a web site;

• TCP/IP: the most widely used communications protocol for sending data from a web site around an intranet or the Internet;

• Web browser (client): software program designed to request information from web servers and display it once it is returned;

• Web server: software designed to deliver files and data to web clients, via browsers. It can also refer to the complete installation of server software, server content and server hardware;

• Web site: one or more web servers providing information that is available to any user with a web browser. The information is linked together with hypertext links. The home page is the main menu or opening page of any web site. A Universal Resource Locator (URL) gives the intranet web site and individual web pages unique addresses.

The biggest applications for intranets today are internal communications, followed by knowledge sharing, management information systems, customer service, sales and marketing and training. When asked how those applications will be extended in the future, organizations cited a broad range of possibilities, such as global business applications, call centers, software distribution, and marketing. Research organization Input found that intranets are still at the very beginning of their life-cycle, and are largely being used for straightforward, but potentially high rate-of-investment applications such as internal information distribution (71 per cent of respondents).

One step up from intranets are extranets. These private Internets tie together groups of companies that do business together. The group could be a manufacturer and its allied distributors and retailers, along with its vendors and banking partners. They would use the public Internet to transmit and receive between their intranets anything from parts orders and inventory data to payments and invoices. Generally, these transactions require encryption.

In addition to intranet, electronic commerce and electronic data interchange (EDI) applications that can run on the Internet, users also look to the World Wide Web as a way of doing business. Several corporations have already launched serious investigations into how to use the Internet to increase revenues, customer responsiveness and overall business efficiency. And not all companies studying the Internet are in the computer or telecommunications business.

Boeing, for one, has detailed an extensive intranet plan, with which it hopes in 10 years to conduct almost all its business over the Web. Meanwhile, machinery manufacturing vendor Caterpillar is the anchor participant in a test to see how well the Internet can reduce costs, increase customer support and satisfaction, and contribute to global competitiveness. Under the auspices of the U.S.-based National Information Infrastructure Testbed (InfoTEST), a 37-member consortium of corporations, universities and U.S. government agencies, Caterpillar has designed a hypothetical product supply chain that simulates its own distributed manufacturing organizations. The only carrier member at this point is Sprint. The trial of what InfoTEST calls the Enhanced Product Realization system is aimed at determining whether Caterpillar - and by extension any other manufacturing company with worldwide operations - can make design-to-delivery modifications for any product, anywhere, within five days. By the trial's conclusions at the end of this year, members hope to better grasp how the Internet, intranets and extranets work and the benefits they provide.

As users contemplate their networks' future architecture, they increasingly look to companies that clearly have their Internet act together. If any carrier is to gain a foothold in this business, it needs to demonstrate - from the CEO down - that it understands the communications requirements for each major user on a specific level. Part of this "common visioning" means service providers must be able to discuss ways they use intranets themselves. Many carriers are at a serious disadvantage because they have not adopted the Internet into their daily business - let alone culture. Few established carriers besides AT&T, MCI and Sprint can speak intelligently about applications from their company's perspective.

In recent years, the main requirement that has driven many of the LAN technology developments is performance. This desire for performance has several contributing factors:

• Networks became bigger as more users are connected;
• PCs and the applications on them became more powerful;
• Isolated networks had to be connected as there were more and more centralized services;
• The Internet/Intranet revolution caused many more network-oriented services.

As networks become crowded, users seek for solutions that can ease the congestion, but are evolutionary rather than revolutionary. LAN switching is a very good solution for this problem as it provides extremely good performance at a reasonable cost and without affecting the upper layers of the network (i.e., no changes to workstations or servers).

The last issue is the cost revolution. As with many other high technology markets, the networking industry was able to provide more functionality for lower cost. Driven by new silicon and software technologies, vendors were able to provide much faster solutions for the same or lower price. In this competitive market, users were able to get much more powerful solutions at a reasonable cost. For example, the cost of 10/100 Ethernet adapter and its switched Ethernet port is today down to $350 or less (list prices). This is less then the cost of Ethernet adapter and shared Ethernet hub port of four years ago, and is probably around 100 times faster.

Looking into the future, we have to assume that at least some of today's requirements will remain valid. We also have to assess new requirements and investigate their influence on the network.

• Performance is likely to remain a key requirement in the future. The next main bottleneck is probably going to be Routers. Layer 3 switches are going to replace them in many networks, especially in the LAN. This is going to provide additional boost to network performance and will enable new applications on the network;

• The same is true for network management. More powerful tools that support new types of switches are probably going to be required;

• Cost will remain important. There are technology enablers to provide further reductions in the cost of some network components which is always appreciated by users. This cost reduction process will also be influenced by the new mid-size networks: this market segment provides much of the market growth and these users are very price-sensitive.

There may also be some additional requirements:

• Multimedia applications, especially frame-based Video and Voice services, may appear in the LAN and will demand the appropriate level of service. These applications will require new set of switching services and protocols;

• New levels of resilience are probably going to become increasingly critical for users. As networks become more central to business operations, users will require more reliability and network uptime, even at the expense of performance or functionality.

• Security will also become more important. As networks become the universal connection medium, it will become increasingly important to protect valuable business assets from unauthorized access - both external and internal.

It is clearly time to look beyond pure performance and port counts. Network Managers need a consistent set of tools if they are to efficiently monitor, trend and troubleshoot today's cell- and frame-switching networks, let alone the enterprise-wide, multiservice (data, video, voice), multilayer (Layer 2 and 3) switched environments of the future.

US West Inc., a potential newcomer to the long distance market is already doing exactly that. It's announced a push to implement xDSL nationwide. It already has frame relay for data. Now it wants to leverage its infrastructure and layer voice on top of it. Its protocol? IP, of course. Worldcom Inc. is also well aware of the changing telecom environment. That's what makes its proposed buyout of MCI Communications Corp. so compelling - and so potentially lethal to its rivals. Worldcom already owns the premier ISP - Uunet Technologies Inc. If it gets government approval, it will add MCI's voice expertise to Worldcom's unmatched IP experience. Being in control of circuit - and packet-switched technologies could give it a huge edge going into the next century - on a worldwide scale.

The net is a turning point. Carriers are pushing vendors to develop faster and more robust backbone routers and switches using new technologies. And customers are pushing carriers to deliver higher bandwidth for business-critical applications. The real issue is whether carriers can deliver what customers demand: the same kind of security, reliability, and ease of use they expect from telephony services.

Multicasting ease these constraints. Say a bank wants to educate its younger customers on the basics of buying a home. Instead of sending an informational brochure, it might decide on using a 30-minute video instead. Today, the bank would have two choices in the making and distribution of that video: Mass-produce VHS cassettes and pay the postage to mail them, or put the video up on the Web site - where it would encounter some of the limitations we have already discussed. But with multicasting, the video could be made available to tens of thousands of users at the same time.

At it doesn't take much to imagine what other applications may flow from this. Besides the obvious implications for national broadcasters, multicasting could also be used to distribute software updates and a range of multimedia products. Multicasting makes sense from the carrier perspective, too. Because duplication of packets is delayed until the last possible moment, multicasting helps conserve network resources. And with push technologies emerging almost daily, it's not hard to see how multicasting could play a significant role in alleviating network congestion.

But multicasting is just one key aspect of the next-generation network. Another major technical issue that needs to be addressed is security. Security has been the subject of a number of earlier FAIR Working Papers (WP Nr.7: Securing Electronic Networks - WP Nr.9: International Governance of Cryptography - WP Nr.13: Erosion of Privacy and Security in Public Telecommunications Networks - WP Nr.14: The use of Encryption in On-line Services) and will not be discussed any further hereafter. Suffice it to say that strong security is an absolute prerequisite to the take-off of electronic commerce.

Strong encryption has to be made available to all. It's vital. And it might give consumers the assurance they need when making purchases over the 'Net. Right now, it's about as easy for someone to intercept a credit card number over the Internet as it is for a waiter to lift one from a carbon slip in a restaurant. Fortunately, the theft of confidential information on the Internet isn't that common. But it doesn't take much to scare the public off (or at least make them nervous). That's why we need strong encryption. Without it, the long-term viability of electronic commerce is in doubt.

Internet service providers (ISPs) are developing premium levels of service for future higher bandwidth, real-time enhanced offerings. In the short-run, these services are likely to be provided as intranet offerings, essentially commercial private network services which guarantee certain levels of performance within their boundaries. Naturally, customers will need to interconnect these intranets, and the vendors will have to build standards, gateways, settlement processes and so on, in order to accomplish this interconnection. Then the ISPs will have, in effect, established the rules for entry into the Internet II club.

In addition to the practical attempts made by network participants, industry standards bodies and consortia are extremely important in the ongoing evolution to utility level functionality. The establishment of ground rules are needed to allow competitors to compete fairly on the basis of better functionality and marketing, and without creating unwelcome market confusion which inhibits the new market's overall development. So far, this initiative has received industry support from organizations such as the IETF, International Telecommunication Union (ITU) and MPOA (Multiprotocol over ATM). Typically, these organizations will not only help to establish the standards which ensure interoperability across vendor equipment and legacy technologies, but also promote testing to ensure that the standards are practically implemented.

In this area, there is a need to extend definitions for technologies such as Internet telephony gateways, multimedia gatekeepers and the application level schemes to implement newer standards such as RSVP. Naturally, once new standards are implemented, products will still need to be tested, and work will be needed to implement true compliance and interoperability.

On a more practical level, local bandwidth (LANs and local distribution) must be increased to support widespread use of multimedia on the campus or building. The expected rate of adoption of the enabling new technology, Fast Ethernet (100Mbps), Gigabit Ethernet, switched Ethernet, switched Token Ring, new multiprotocol switching system, IP switching and even desktop ATM should result in a significant portion of enterprise desktops being enabled for multimedia by 2000. The decision to adopt broadband capabilities for enterprises installing new LAN systems today is easy as the cost difference is almost negligible per port.

Another practical issue is the local access technology used for wide area transport. For large campus environments, the answer is high-speed broadband connections which are shared by large number of users - these broadband connections may be multiple T1s, T3s, or ATM connections. The solution for small offices and telecommuters is ordinary modem dial-up and high-speed ISDN now, and ADSL and cable modem technologies in the future. Ordinary modem dial-up can comfortably support audio and data collaboration and with ISDN, it is also possible to get enough bandwidth to enable decent quality video. ADSL and cable modem technologies will be very useful when they are deployed by local access providers because they offer the potential for higher bandwidth TV quality video.

WANs with different QoS (Quality of Service) classes including isochronous or real-time delay and jitter minimization quality, are enabled today through ATM technology. The more common router networks are beginning to advance to the QoS enablement stage as routers using the emerging RSVP protocol begin their initial deployment. RSVP is too new today to be in use for universal services in the near term since multimedia and other applications have not been modified to use it yet. Nevertheless, RSVP should be useful for limited deployment of video applications in the next 12-18 months. RSVP is not, and never will be a general panacea for managing network congestion and so ultimately, the network has to be engineered for adequate bandwidth regardless of the sophistication of the protocol. Even Frame Relay is being developed to support voice and limited video services. The key is, as always, managing the bandwidth and preventing oversubscription of the network which will unduly delay sensitive real-time applications.

It is important to recognize all the work that has been accomplished in addressing market needs with the multimedia technologies that are available for users today. The key is to implement the highest value-added applications first, this is analogous to the incandescent light bulb which was the first application for electricity because it expanded the usable day. Real-time multimedia applications can likewise shrink the globe for international enterprises with high-value applications such as:

• international voice and fax services over IP to reduce cost;
• Internet websites tied to call centers for better sales and customer support;
• desktop video over IP for improving collaboration in industry/specialized applications;
• desktop video multicasting for effective delivery of distance learning and other broadcasts.

IP data networks have enormous momentum to become the common utility infrastructure for communications and information services. Users should learn how to adapt to the opportunities these early implementations present. They gain competitively by establishing better means of interfacing with customers and working internally, and gain critical experience as they do. Vendors can similarly offer the market what is feasible and economically practical to develop and deploy today. Those who adopt a wait-and-see attitude do so at their own risk - the market and technology of the Internet are not governed by the same life cycle as the old telco services. The day of the geographical communications monopoly is ending and service providers that miss the call may yet evolve into raw transport providers that have a lot in common with old fashioned electrical utilities.

The past couple of years have seen two major trends in high-speed transmission. The first is the on-going improvement of WDM (wave-length-division multiplexing) technology. NTT, which boasts the leading WDM research lab, is using the technology to pack larger and larger amounts of bandwidth into fiber. The company has moved way beyond the terabit level, and fiber itself has a theoretical capacity of around 50 to 75 Tbit/s.

In the commercial domain, WDM is so cost-effective that standards work on faster serial speeds in fiber (like OC192c [10 Gbit/s] and OC786c [40 Gbit/s]) has pretty much come to a halt. Some carriers have concluded that using WDM to send multiple OC48 (2.5 Gbit/s) channels is more economical; in fact, MCI already uses an OC768 link over WDM. Right now, the commercial standard is 8 or 16 channels per fiber, a density referred to as Dense WDM (DWDM). But researchers already are packing thousands of channels into one fiber, which means DWDM will soon be supplanted by even denser technologies - and marketing departments will have to come up with other terms to describe them.

The second fiber trend is continued improvement in analog signaling techniques. As we all know from modem technology, it's possible to send more than 1 bit per pulse by varying the amplitude in the pulse. In a fiber optic transmission, the norm was just 1 bit per pulse - until recently. Now researchers are sending as many as 3 bits per pulse, effectively tripling the theoretical capacity of a fiber to about 150 Tbit/s.

The innovations of the late '80s and early '90 have moved quickly into products. The most important of these are probably the improvements to the optical amplifier. Signals in fiber degrade over distance and need to be amplified every 100 kilometers or so. This used to be handled by optical-electrical amplifiers, but these devices put major limitations on how many channels could be signaled through a fiber and how fast it could be done. Upgrading a link meant upgrading all the amplifiers in the fiber path. But now, with optical amplifiers, that's not a problem. Transmitting more channels simply means upgrading the equipment at the ends of the fiber path - and the optical-electrical amplifiers already in place can stay in place.

Switching technology is booming too. Commercial routers and switches now incorporate switched backplanes capable of moving 20, 40, or even 100 Gbit/s. And researchers are doing even better: At Stanford University, Nick McKeown - with help from Texas Instruments Inc. and Cisco Systems Inc. - is putting the finishing touches on the Tinytera. It's a near-Tbit/s switch the size of a soda can, made using CMOS technology. Optical packet switching isn't capable of those speeds just yet, but it's improving quickly - and it will be available if and when electronic switching runs out of steam.

Meanwhile, routing technology is making rapid progress too. With Internet data rates increasing, established vendors, startups, and research labs have made 1997 a year of technical innovation. Commercial routers can now forward well over 10 million packets per second or pps., allowing for transmission rates in the 10 gigabit/sec range. (As a rule of thumb, for every gigabit of bandwidth, a forwarding capacity of 0.5 million to 1 million pps is needed, assuming average packet sizes of 128- to 256-byte).

Even more stunning, the least expensive routers can forward packets almost as fast as the most expensive routers, thanks to ASIC technology (although the most expensive routers can do more per packet, such as applying firewall filtering). Industry and university experts generally agree there are only a few remaining obstacles to building terabit routers.

Moore's Law predicts that by 2002 chips will be about 10 times faster. Smart engineering can often boost that by a factor of two or better. Recent research on routing table management at Washington University and Lulea University in Sweden is promising a factor-of-10 performance improvement. And parallelism can probably buy us that last factor of five.

Given this backbone network potential for extremely high speeds, the question is where is all the data needed to fill that network going to come from? Well, we all know that wireline technology is improving. Cable modems and xDSL both furnish megabits to the home. Also keep in mind the work of UCLA's Leonard Kleinrock, whose latest statistics show that 90 percent of all LAN traffic has a destination off the LAN (a stunning reversal of the trend just a couple of years ago). So much of that edge capacity shows up in the middle. Wireless is improving, too. A recent Ph.D. thesis at MIT showed that by making better use of power management in radios, it's possible to deliver 200 Mbit/s per user. Also worth keeping in mind is that we'll soon be able to attach little wireless antennas directly to silicon chips, making all sorts of new communications patterns possible - not to mention very cheap.

Switching and routing have evolved rapidly in the last 25 years. And it now appears we've reached some stability in these two fundamental networking technologies. The new generation of LAN internetworks will consist of hybrid switches and routers and be based on frame technologies such as Ethernet. Public wide-area nets will also be a hybrid but will primarily handle cell-based ATM. Because of the Ethernet dominance at LAN level, integrated Giga Ethernet/ATM switching will also be in strong demand though.

So are routing and switching technologies mature? Absolutely not. The cells-vs.-frames debate will rage on. What's more, these two cornerstone networking technologies are moving up the protocol stack. Routing has traditionally been based solely on destination host numbers. In the future it will also be based on source host or even source users, as well as destination URLs (universal resource locators) and specific business policies. The forces that have changed switching and routing in the past are still at work, and will lead to even more transformation. A brief history of these key technologies reveals some of what the future holds.

In the beginning, routing was very simple. It had to be, because packet switches were minicomputers, which could route 1,000 packets per second, more than enough to handle a couple of 56-kbit/s trunks and 100-kbit/s hosts. By the 1980s, the arrival of LANs meant routers had to be 100 times faster. Special-purpose processors, ASICs (application-specific integrated circuits), and buses made this possible. Meanwhile, routing became a bit more complex every year. Routers had to handle multiple protocols, some with Layer 3 routing and some with Layer 2 bridging.

By the early 1990s, the growth of the Internet had prompted the development of many more routing algorithms, including EGP (exterior gateway protocol), BGP (border gateway protocol), and multicast routing. Not only did routers have to carry many data protocols, but also they had to understand many routing protocols. Routing tables became enormous - as many as 50,000 entries in the core of the Internet. Algorithms that used to work well for small networks broke down under the demands of huge global routing calculations. Routers began to have a "fast path" for cached routes and a "slow path" for new routes. Performance became nondeterministic. So every year, routers became bigger and faster, but also more complex and difficult to administrate. And this trend continues.

Switching also has become increasingly intricate. The first switching technology was circuit switching, installed worldwide for the public switched telephone network (PSTN). In an effort to apply the advantages of circuit switching to data networks, ATM was created in the early 1990s. Routers were already staggering under the data load alone, but it was clear that the data networks would begin to carry voice, image, and video. The best-effort datagram approach was never designed to carry traffic with demanding QOS (quality-of-service) requirements, while switching had been doing so for decades. Likewise, there were questions about whether routing could scale to millions of switches, while switching seemed adaptable.

ATM technology has many advantages. It also has a serious drawback: It uses fixed-length cells and connection-oriented signaling, but almost all the computers in the world communicate with connectionless datagrams. ATM products came too late, 20 years after the development of Ethernet and the Internet, and 15 years after the introduction of the personal computer, when datagram technology was entrenched. Building end-to-end ATM networks means adding a lot of hardware and software to the computers at each end. The designers of ATM forgot the lesson of the PSTN: simple termination points. For telephony, that means connection-oriented signaling. For computer networking, it means connectionless, for historical, financial, and technical reasons. End-to-end connections are not efficient for short bursts of data, which typify e-mail, Web surfing, and LAN communications. And with fast Ethernet adapters selling for $64, the dream of end-to-end ATM switching is now dead. Switching is not going to replace routing.

However, something more interesting is happening: hybridization. When ATM was introduced, many LAN vendors did not know whether to try to stop ATM or embrace it. A few years later, it's clear that Ethernet and TCP/IP rule the LAN infrastructure. But ATM hardware has some compelling advantages. So the next-generation LAN switches, particularly for gigabit Ethernet but also for ATM, represent a cross between switching and routing. They still inspect each packet and route it on a connectionless basis. But like switches, they employ dedicated hardware for traffic forwarding - the "fast path" is now the only path.

In the WAN, things are more complex. Routers are an absolute requirement for connecting between LANs and WANs. Ethernet rules the LAN. But ATM will be quite common in the WAN, particularly in the telco world. So how will switching and routing coexist?

In the early 1990s, the challenge was simply to make ATM work with LANs. The extremely simplistic idea of ATM as a logical wire (RFC 1483) was followed by classical IP over ATM (RFC 1577). The first effort of the ATM Forum was LAN Emulation, mirrored in commercial offerings by one-armed routers. Then, in the mid-1990s, the focus shifted to developing shortcut flows to avoid routers. The ATM Forum began work on multiprotocol over ATM (MPOA), recently standardized. Commercial offerings began to feature route servers. Both 3Com Corp. and Ipsilon Networks Inc. (now a subsidiary of Nokia) came out with faster versions of IP.

Today, the emphasis is on merging switching and routing. Many companies' products use this idea - IP Navigator from Cascade Communications Corp., Cell Switched Routers from Toshiba Corp., and, most notably, Tag Switching from Cisco Systems Inc. The IETF (Internet Engineering Task Force) is working on a standard version - multiprotocol label switching (MPLS). The idea behind these sorts of schemes is to move routing to the end edge of the network and rely on faster, simple switching in the core.

Many entrepreneurs are giving up on the idea of ATM and switching however, and building routers to handle 50 million or 100 million pps. They reason that restricting the problem to just one protocol, TCP/IP, and dedicating hardware to the job is enough to get the necessary performance. This approach solves the problem at the product level, not at the network architecture or protocol level, a considerable simplification.

Some vendors are adding traffic management to routers, while others are putting it in a separate box. Traffic management can take place in the firewall, the access concentrator, and now the Web switch. It will become far more specialized as customers demand fine-grained control of flows. We are already seeing the emergence of load balancers, which intelligently distribute requests across multiple Web servers. Soon we will see traffic managers that can handle class of services, class-based queuing, and per-flow queuing. Eventually, routing, switching, and flow management will be handled individually for each user and Web URL.

Thus, in the future, you may be sent on one path when you casually browse the Web for CNN headlines. And you may be routed an entirely different way when you go to your corporate Web site to enter monthly sales figures - even though the two sites might be hosted by the same facility at the same location.

Some parts of a Web site may be set for business policies and different service classes, like audio and video. An order entry form may get very low latency, while other sections get normal service. And then there are Web sites comprised of different servers in different locations. Future routers and switches will have to use class of services and QOS to determine the paths to particular Web pages for particular end-users. All this requires use of Layers 4, 5, and above.

But whatever the product, if the current situation prevails, the vast majority will be imported into Europe, not designed or manufactured locally.

Along with European Network Laboratories (ENL, Paris), DATA COMM reported in its February 1998 issue an exhaustive evaluation of seven gigabit Ethernet switches from six vendors. They tested devices in both edge and core configurations. Edge devices take departmental or workgroup traffic, typically from Ethernet or fast Ethernet networks, and shunt it onto a gigabit Ethernet uplink that usually attaches to a backbone network. Core switches, in contrast, are put to work in the data center: They move traffic among large numbers of edge devices or even high-speed servers equipped with gigabit Ethernet adapters.

Reams of results and record-shattering performances say it all: By delivering on its promise of high speed and huge capacity, along with true QOS (quality of service), gigabit Ethernet looks like a worthy competitor to ATM - and it's available at just a fraction of the price.

Not only did these devices deliver a full gigabit per second, but also even the slowest ran at better than 97 percent of capacity: What's more, gigabit Ethernet can be used to build enormous switched networks and move traffic through them at blinding speed. Some devices routed traffic just as fast as they switched it, at nearly 12 million IP packets per second. Finally, these boxes boast flow control and prioritization schemes that make it possible to blast packets through even the most congested links - and run delay-sensitive apps like voice and video over gigabit Ethernet connections.

While the technology leaves frame formats untouched, handling traffic from Ethernet or fast Ethernet networks without any conversion, it almost always runs in full-duplex mode. That means more bandwidth. A single gigabit Ethernet switch port can handle traffic from as many as 20 half-duplex fast Ethernet links or up to 200 half-duplex Ethernet ports.

LAN switches are simple, cost-effective and offer excellent performance. So when we need to increase the capacity of the LAN, why not migrate the entire LAN infrastructure to operate exclusively on LAN switches? The answer is that, in some cases, this is indeed possible - but more often than not there is some need for routing in the LAN:

• Historically, the widespread use of routers in existing LAN infrastructures has led to the adoption of IP addressing schemes in which multiple IP subnets exist within the LAN.
• Unlike LAN switches, most routers offer a range of access controls which can form a useful part of the enterprise LAN security policy.
• A fully switched LAN which has no routing in it would have to operate as a single large broadcast domain.
• Some LANs contain a mix of technologies, such as Ethernet and Token Ring.

There are some solutions to these problems nevertheless, in many real networks the practicalities of the situation will dictate that routing functionality is required, for one or more of the reasons we have identified. This used to mean that if conventional LAN switches are deployed, it was also necessary to deploy routers. Now Multilayer switches are changing that.

A multilayer switch is a device which has multiple LAN ports over which stations can communicate either by means of Layer 2 packet forwarding (as in conventional LAN switching), or by means of Layer 3 packet forwarding (as in conventional routing). The type of packet forwarding that is used in each case is whichever is needed for any given pair of stations to communicate. In practice, this depends on whether they are members of the same subnet - in which case Layer 2 forwarding is used - or if they are members of different subnets, in which case Layer 3 forwarding comes into play.

A multilayer switching device can be logically viewed as a Layer 2 switching fabric which has a Layer 3 forwarding function attached to it by a high-capacity connection. A number of LAN port interfaces are attached directly to the Layer 2 switching fabric. Just like a conventional router, the Layer 3 forwarding function has one or more IP addresses and MAC addresses associated with it that end stations use to send IP packets to it, for forwarding on to different subnets. Similarly, if the Layer 3 forwarding function supports other protocols such as IPX, it would like an IPX router from the point of view of the end stations. For other Routers, Multilayer switches looks like another router installed in the network.

LAN switches generally achieve very good price/performance by applying hardware-based frame forwarding techniques to the process of moving packets from one LAN segment to another. This is relatively easy to engineer, since the forwarding decision is based on a simple MAC address look-up table, and the forwarding process involves no change to the content of the packet.

Routers, on the other hand, generally exhibit much lower forwarding rates at considerably higher cost. Three reasons for this are:

• the packet forwarding decision in a router is considerably more complex than that in a LAN switch;

• a number of changes have to made to the contents of each packet received from one LAN segment before it can be forwarded to another - for example, the entire MAC packet header has to be replaced, and the Time-To-Live field has to be updated in the IP packet header;

• if packet filtering is used to implement network security, each packet may have to be compared with a number of complex filtering criteria.

This additional complexity has meant that many routers implement the packet forwarding process mostly or entirely in software - which accounts for the inferior price/performance characteristics of routers compared with LAN switches.

Next generation switches will include even more functionality at a lower price. The internal expertise in switching will lead us to advanced solutions in the following areas:

• any layer switching - provide a switching solution that handles all layers of the protocol stack to achieve better optimization of network resources, higher security and better service-oriented switching;
• resilient solutions that adapt in real time to the changing network conditions, and provide a self-healing capability to the system;
• better integration of the Network Computing devices with the network's switches.
• higher performance and higher port counting in a market competitive cost.

The Internet protocol (IP) has evolved to become one of the most flexible networking protocols available today. Despite its flexibility, IP networks have some characteristics which in the past, made it difficult to use in corporate networking environments. However, recent advances in IP technology - the increasing widespread use of IP-based applications and a growing body of experience in the management of IP networks - mean that IP-based networks are now appropriate and robust enough for the vast majority of today's corporate networking needs.

The ability to share information across a single, multi-site enterprise that crosses international boundaries, is a strategic necessity in business today, and as companies grow and diversify, the number of sites where decisions are made and business is conducted is also increasing. Connecting these locations securely and reliably using the latest technologies can be costly and complex since it requires significant ongoing capital investment and evolving expertise. Moreover, as service-level demands increases, in-house resources can become strained, jeopardizing overall network reliability and diminishing effectiveness.

One way of protecting investment and remaining flexible for the future is to make the decision based on the protocols carrying the data, not the protocol of the communications medium. Using IP, companies can choose the telecoms medium best suited to their needs while keeping the corporate infrastructure stable and flexible. This strategy protects investment in the network while also maintaining open options for future transport technology.

Consider a global network, or a fully meshed global network. Each individual connection (PVC for Permanent Virtual Circuit) into the hub site or interconnected sites incurs the cost of a PVC and committed information rate (CIR). In a fairly small environment (nine to ten connections), frame relay would be an ideal solution. However, when considering a very large national or global network, a fully meshed network, any-to-any connectivity, high bandwidth and expected growth, IP becomes an extremely attractive network solution.

IP is a connectionless protocol which removes the need to configure point-to-point connectivity. IP, by default, has connectivity to every site. Such fully meshed networks also provide a greater level of redundancy, as traffic has any-to-any network connectivity. Furthermore, with a fully meshed network, the customer can better protect against the loss of a hub site. If a hub site experiences a power outage, local loop problems or router failure, in a fully meshed environment, full network connectivity will not be lost as remote sites are not primarily dependent on the hub to communicate with other sites and/or devices. The remote sites will simply send traffic via other network locations.

Internet service providers, equipment vendors and software developers say they can now give the security, performance, availability, and multiprotocol support of a private network over the cost effective and global Internet. This connectivity is provided by a virtual private network (VPN) or extranet, and the technology is currently being considered primarily as a means of extending the reach of private networks. In addition, VPNs may address locations where traditional private network connections cannot be economically justified.

IP network providers have applied tremendous resources to the planning, testing, deployment and day-to-day operation of the backbone network. The redundant mesh topology of an IP network can now ensure robust and reliable interconnectivity operating at peak efficiency and providing a high level of service. According to vendors of VPN technology, by taking advantage of the various authentication and encryption features these devices boast, it is possible to forge industrial-strength links over the most porous of public networks: the Internet.

So in addition to the cost of the physical components, the buying criteria can now also take into consideration the level of management, security, monitoring and performance included in the price of the service.

But firewalls aren't enough. What is also required is network wide mutual authentication and centrally controlled access authorization in a distributed environment. Once the connection is established, unique site-to-site data encryption at the customer local area network (LAN) level ensures confidentiality of all network traffic, which is unintelligible and unchangeable to unwanted visitors. IP network providers can protect data by offering end-to-end network based data encryption, where the traffic is protected before it leaves the customer site.

At the backbone level, the Internet is a crude indicator of the shape of things to come. A second-generation 'Net is in the offing - one that will ultimately deploy a mix of multigigabit and terabit routers and ATM (asynchronous transfer mode) and SDH/Sonet (Synchronous Optical Network) switches, along with high-speed access schemes like xDSL (digital subscriber line) and cable modems.

The TNOs know what is coming. Some have started to invest in packet switched backbones. The ISPs (Internet service providers) know it. In fact, voice over IP and other enhanced offerings are key to their economic survival. The IP equipment suppliers are counting on this paradigm shift to create a market for voice over IP gateways. Sure, some of these devices come from startups like E-Fusion Inc., Inter-Tel Netsolutions Inc., Netspeak Corp., and Vocaltec Inc. And some of these newcomers may not survive; others could end up billion-dollar players. But internet-working giants like Bay Networks Inc., Cisco Systems Inc., and 3Com Corp are taking voice over IP very seriously. The same goes for other heavyweights like Lucent Technologies Inc., Microsoft Corp., Northern Telecom Ltd., and Siemens AG.

Most important of all, though, are the economic drivers. Voice over IP and packet switching will win out in the end because they can deliver services far more cost-efficiently than today's circuit-switched technology.

It all sounds great. But a couple of things are definitely needed if voice over IP is going to go anywhere, starting with gateways, devices that translate between the circuit-switched and packet-switched worlds. Cisco is positioning its 3600 router as a voice over IP gateway; Lucent has a gateway. And nine other vendors are shipping products, with about 20 expected soon. Sure, some of the products aren't much more than toys. But the same was true for the PC in the early 1980s.

And gateways are just the beginning. Nearly every major vendor has either announced or delivered an IP voice product. Some are offering IP-based central office equipment, typically incorporating high-density modems and Layer 2 and Layer 3 switches that can field voice, video, and data. Cisco is melding its AS5300 ISDN/analog modem bank with the Catalyst 5500 switch. Ascend Communications Inc. is bundling its TNT dial-up gear with its GRF gigabit router. And 3Com is adding voice over IP and fax to its Total Control Hub.

But don't let the presence of ATM and SDH raise any doubts. IP is the protocol, and it will be routed. ATM and SDH will serve as the transport. And running IP over SDH will help overcome some of ATM's inherent inefficiencies.

The economic case for a network to support both native ATM and existing non-ATM services is particularly strong however. While non-ATM services could be moved from current networks on to an ATM backbone, a more likely scenario is that the ATM infrastructure will provide additional capacity for non-ATM traffic to cope with growth in this area. This in turn would allow new investment to be channelled into a future network architecture rather than current non-ATM infrastructure.

As data services such as frame relay and LAN interconnect are only designed to support variable-rate, bursty traffic, it is hardly surprising that this makes up the majority of the traffic on current ATM networks. Native ATM services, on the other hand, support both variable- and constant-rate traffic on a connection-by-connection basis. Growth in native ATM applications will lead to a greater mix of traffic types being carried by ATM networks as new applications make use of the additional capabilities.

The development and deployment of native ATM applications that exploit the full potential of ATM networks is, however, to some extent dependent on the widespread deployment of ATM networks. Increasing the deployment and reach of carrier ATM networks will help to stimulate the deployment of native ATM applications which in turn will stimulate the growth of ATM networks. Adaptation of non-ATM traffic into ATM traffic can help increase the load on ATM networks and so accelerate expansion of the ATM network. At the same time, this can help avoid having to make a choice between investing in expansion of the existing non-ATM network to cope with additional demand or expanding the native ATM network to cope with future demand.

If all networks are considered, voice traffic is characterized by being a constant-rate traffic stream with very stringent requirements on timing and delay across the network. Data traffic, on the other hand, is typically a highly variable-rate (bursty) traffic stream and can tolerate both delay and delay variation when crossing the network.

ATM becomes increasingly attractive as a unifying network protocol when the mix of constant - and variable-rate traffic increases. While ATM is quite capable of carrying only constant-rate traffic or only variable-rate traffic, there are other solutions that can do this and which may be more efficient. The strength of ATM is that it allows both traffic types to be mixed within a single network. Finding ways to increase the traffic mix over ATM networks by moving non-native ATM traffic on to the network (particularly constant-rate traffic) greatly assists in justifying the deployment and expansion of ATM networks. Adaptation is the term given to the process of converting non-ATM traffic into ATM traffic. During the development of ATM standards, the need for adaptation was recognized and standards for how adaptation should be performed were defined.

A number of adaptation schemes have been developed to support the many different types of non-ATM traffic. Currently, there are four main adaptation mechanisms in use within ATM networks (Table 2), and each is identified as an ATM adaptation layer (AAL):

• AAL1 is optimized for carrying synchronous bit-streams over an ATM network. The common term for this service is constant-bit rate (CBR)

• AAL2 has only recently been defined by the ITU and supports the carriage of intermittent bit-streams, which have a precise timing requirement, over an ATM network

• AAL3/4 is designed to carry variable-rate data over an ATM network. AAL3/4 is mainly used to support SMDS/CBDS services over ATM networks

• AAL5 is also designed to carry variable-rate data over an ATM network, however, AAL5 is much more widely deployed than AAL3/4, and is additionally used for carrying Internet protocol (IP) and frame relay traffic.

While ATM adaptation provides the mechanism to map non-ATM traffic into ATM cells, it does not address the issue of mapping non-ATM services over an ATM network or to an ATM service. One issue involved in the mapping of services is the association between a non-ATM traffic stream and the corresponding ATM virtual circuit. How this is achieved is partly dependent on the non-ATM services being mapped.

A good illustration of the numerous issues that can arise from adapting non-ATM traffic to ATM is the mapping of a 64kbps voice circuit over an ATM network. The adaptation scheme used is AAL1, which makes use of one byte of the cell payload to carry AAL1 protocol information, leaving 47 bytes for voice data. Once filled, the ATM cell crosses the ATM network and terminates at the destination adaptation unit where the data is extracted from the payload the bit-stream regenerated.

While the time the cell takes to cross the ATM network is typically very low, the delay experienced by the voice traffic is not. This is because of the 'packetization' delay incurred while the cell payload is being filled before it is sent. The delay is of the order of 6ms as a bit arrives every 15µs and a single cell holds 376 bits (376 x 15.5 = 5.8ms). One consequence of this delay is the increased probability that echo cancellation will be needed on the voice connection.

Another factor which needs to be taken into account is the possibility that cells may experience delay in crossing the ATM network, or in extreme cases may even be lost. In either of these situations, the destination adaptation unit may need to send out bits in the outgoing 64 kbps bit-stream when it has no cells available to extract the data from. The standards for adaptation do not specify what should be done in such circumstances.

Any significant shift of non-ATM traffic from traditional networks on to ATM networks will require adaptation in the service provider network on a much larger scale, with equipment that can provide ATM adaptation in bulk. Such adaptation will need to cope with both static and dynamic service mappings that require interaction with signaling.

Of course, the adaptation to and use of ATM must be totally transparent to the user of the non-ATM service. For example, a telephone user should not be able to tell whether the call is being carried end-to-end over an existing TDM network or carried for part of the way over an ATM network. To date, most of the focus in ATM equipment development has been on building ATM switches. The adaptation equipment that exists today has typically been engineered for customer premises and enterprise networks.