Voice and Data Worlds Collide

by Susan Almeida, Vice President of Consulting Services

Strategic Networks Corp.

As we hurdle through the information age, we're at the beginning of a once-in-a-century tectonic shift. For the first time, data is about to eclipse traditional voice traffic. As the data snowball barrels forward, voice calls will become a minuscule piece of the information traffic streams coursing around the world. Simply put: networks will be designed for data, and traditional voice calls will piggyback along the cheapest path. Also for the first time, voice is taking on a strategic role-clever voice/IP applications that will radically transform business, much like call centers, Web sites, and e-mail already have. What is the enabling technology? Packetized voice, i.e., techniques that chop voice up into bite-sized pieces for network transport, including Voice over IP (VoIP), Voice over Frame Relay (VoFR), and Voice over ATM (VoATM).

These two trends-explosive data growth and strategic voice/IP applications-are nothing short of monumental: they are fundamentally altering the global telecom infrastructure. The shift from traditional circuit-switched networks to data- centric networks is more fundamental than the last great transition in the global telecom infrastructure: the move from analog to digital. That shift created economic activity measuring billions of dollars through new central office switches and transport services. Today's sea change from circuit-switched to packet-switched nets will make that transition look like child's play. In the wake of this great re-buildout, some very big players too slow to move will be shunted to walking-dead status; others with the innovation and courage to act will be catapulted to the head of the race.

Let's explore the impact on infrastructure and operational efficiencies first, then the role of strategic voice/IP applications.

Follow the Dollars

For the last century, circuit-based voice traffic loomed monolithic; data always the second cousin trying to elbow its way through infrastructures optimized for carrying phone conversations: skinny, rigid telephony pipes. But if we zoom forward, traditional phone calls will be but a tiny sliver of the pie. By various accounts, the swing will be between the year 2000 and 2010. In some U.S. regions, like Silicon Valley, the pendulum has already swung.

As the center of gravity shifts to data (specifically, IP), there's a huge sucking sound as investment dollars follow the traffic. Until now, the lion's share of funding has funneled into a telephony-centric infrastructure. This is why most Internet traffic today, including voice over the Internet, actually runs over telephony nets.

But the days of telephony-centric circuit switching are numbered. Why? Because service providers and enterprises alike are now pouring their investments into the high-growth area: data. Why waste precious dollars funding infrastructure optimized for the slow-growth phone call market? With the bulk of capital dollars and senior management attention flowing to packet-switching nets optimized for data, it becomes ever more expensive and cumbersome to maintain a separate voice infrastructure. Network planners don't like to run multiple nets when fewer will do. It is massively expensive and inefficient to maintain over the long run. The natural proclivity is to consolidate infrastructures wherever feasible. And integrating voice and data is a time-honored recipe for saving costs-higher volume and complimentary traffic patterns means better dollar-for-dollar use of the fixed infrastructure investment. The new wave of innovation in packetized voice has service providers and end users alike itching to once again integrate data and voice. But unlike the 1980s integration over TDB networks, this time it will be over data-centric infrastructures.

The itch to integrate is not lost on the equipment vendors. If you take the pulse of the equipment vendors you'll find massive brainpower and engineering dollars being pumped into packetized voice. It is inevitable that technical barriers will drop-and voice will be an integrated stream flowing through the global packet-switched infrastructure. Service providers and enterprise users will be able to collapse their half-dozen separate data and voice infrastructures into a souped-up, IP-centric, packet-switched infrastructure. Separate telephony-centric, circuit-switched infrastructures will eventually shrivel up and die.

Why Packet Switching?

Simple: it's the architecture that works best for data and it's also the technology that squeezes the most out of underlying bandwidth. Some argue that with fiber build-out, Wave Division Multiplexing (WDM) and switching advances, bandwidth will soon be free and bandwidth optimization mute. The reality: demand will continue to outstrip supply in many parts of both public and enterprise nets. For most net planners, then, bandwidth remains a scarce and expensive resource to be optimized, at least for portions of their nets. While traditional circuit-switched nets designed with telephony in mind rely on fixed, nailed-up connections; packet-switched nets-whether cell (ATM), frame relay, or IP- offer fluid, elastic connectivity that squeezes the most out of expensive pipes. And with the data market innovating at "Internet speed," packet-switching costs are driving down faster than was ever conceivable in the traditional telephony switch world. Packet-switching equipment has nose-dived from $60/packet/sec in the '80s to one tenth of a cent per packet/sec with today's high-end routers and ATM switches. The blazing innovation from the data world is not only driving down equipment costs, it's also making packet- switching acceptable for carrying voice. Further, it's pushing bandwidth efficiencies to new heights via advanced voice compression and silence suppression (silence is up to 60% of the average phone conversation). Perhaps one of the most powerful arguments for why packet-switching will be the next infrastructure is that the cost of carrying a voice call is about to drop an order of magnitude.

Let's take a look at what's pumping into the marketplace for packetized voice. The key technologies are: digitization and packetization of analog voice traffic, voice compression and tight control over delay and loss. To get good voice quality you need minimal delay, predictable delay and consistent performance. Natural sounding speech needs more: echo cancellation, background noise regeneration and low loss. Loss is even more critical with packetized voice because a dropped compressed packet is far more noticeable than noise on a traditional phone call.

We've already seen quantum leaps in all of these areas-and the advances show no signs of abating. Take voice compression: while used for 20 years to squeeze the most voice into fixed leased lines, now the dramatic gains are being made in packetized voice. Today's voice compression algorithms already achieve up to 10:1 gains versus traditional 64kbps voice circuits. With 2kbps voice compression now on the drawing boards we will soon achieve near 30:1 gains over a traditional circuit-switched net! Voice quality is subjective so delay benchmarks are tricky, but on the low-end of the traditional telephony delay scale, a satellite phone call takes about 250ms to transmit.

Let's look at how packetized voice performs within the major data-centric networking models: ATM, frame relay, and IP packet switching. You can think of these packetized voice technologies as falling into two categories: packetized Voice Over Transport Services (VOTS), which includes VoFR and VoATM; and VoIP.


The VoFR and VoATM approaches are similar: analog voice calls from a CO switch, PBX, key-system, or handset are passed to a frame or ATM switch. The voice traffic is then digitized, packetized (into frames or ATM cells), compressed, and otherwise conditioned; then switched over a frame relay or ATM transport net to the destination switch, where the process is then reversed.

Voice over ATM. There are two main applications for VoATM: enterprise nets and as transport within service provider nets. For enterprise nets, you connect traditional telephony equipment (i.e., PBX) to your ATM switch, which then converts and packetizes the voice stream. The packetized voice stream is then carried over a private or public ATM network. Many service providers are already using ATM as a multi-service platform to provide frame, ATM, and IP services to users; carrying voice streams moves infrastructure convergence one step closer.

ATM was designed from the start to handle voice as well as data and video streams. VoATM has been available in ATM switches since last year, pioneered by vendors like GDC (Middlebury, CT) and Nortel (Brampton, Ontario). The ATM solutions include compression, echo cancellation, and very tight control over delay through built-in ATM Quality of Service (QoS) mechanisms. Another important development: applying a voice edge/core architecture similar to data nets. The concept: move processing-intense voice routing functions to the edge; and perform high-speed voice switching in the ATM core. The advantage: it cuts delay and improves sound quality by eliminating multiple conversions/compressions that would normally occur if the traffic threaded its way through intermediate tandem switches or PBXs. With key Voice and Telephony over ATM (VTOA) standards now finalized, we'll be able to make broader use of these approaches.

Voice over Frame. Frame relay has come a long way since its data-only roots. VoFR works analogously to the ATM example. But, in the case of frame, the main application is for the enterprise. And since there are so few private enterprise frame relay nets, the transport is a public frame relay net. Leading frame relay access device (FRAD) vendors like Hypercom (Phoenix, AZ), Nuera Communications (San Diego, CA), and Motorola (Schaumburg, IL), have supported voice/frame for the last two to three years. Most FRADs today provide some level of VoFR support, again, including digitization and packetization, compression, echo cancellation, and bandwidth management/traffic optimization. Initially, these were all proprietary, but the VoFR standard FRF.11, covering coding and frame conversion, compression and silence suppression, was ratified in May of 1997. The latest innovations include voice switching in the frame switches themselves. This provides similar quality and delay benefits as the ATM-based voice switching discussed above; while expanding the FRAD's voice calling capabilities beyond branch office-to-regional office point-to-point calls. In terms of delay control, frame relay is not as deterministic as ATM, but again, the frame market has not stood still here. Simple prioritized frame relay has been implemented and is now offered by several major service providers like US West !NTERPRISE (Englewood, CO), MCI (Atlanta, GA) and Sprint (Westwood, KS). US West !NTERPRISE, for example, guarantees 50ms nationwide for its prioritized frame relay service. And the recent ratification of FRF.12 provides a standard way to segment long frames into shorter frames so voice frames are not stuck behind large data frames.

Voice over IP. Among the packetized voice choices, the industry darling in terms of hype and investment is VoIP. And with good reason: VoIP shares the cost advantages of its VoFR and VoATM cousins, but, more strategically, adds tremendous value via new voice/IP applications. Here, we're not referring to audio streams or audio clips on Web sites. Rather, we're referring to the new class of applications that merge Internet capabilities with telephony functions. This strategic value is what's going to tie the bow on V/IP. IP has already won the protocol wars as the native data networking protocol for enterprise and public data nets -- next on the agenda is packetized voice.

Again, focusing on the operational efficiencies side of the equation first, let's look at VoIP. VoIP works similarly to "VOTS" services, but instead of converting to frames or cells, the voice traffic is converted to IP packets. One of the key enablers of VoIP is the "VoIP gateway" which digitizes, compresses and packetizes voice for transport over IP nets, such as internets, intranets or extranets. Various market researchers have pegged the VoIP gateway market in the $1.8B range by 2001 while the dial equipment market is sized at $7.5B in the same time frame. Nearly every major network vendor has delivered or announced some form of VoIP product. For VoIP gateways, VocalTec Communications (Herzliya, Israel) and Nortel's Micom Communications (Simi Valley, CA) were early leaders. Lucent Technologies (Murray Hill, NJ) offers its Internet Telephony Server, Newbridge Networks funded Vienna Systems Corporation (Kanata, Ontario), and Cisco Systems (San Jose, CA) has its 3600 and 2600 series routers. The momentum can also be seen in 100-plus companies that participate in the Voice over IP Forum.

You may say, "Well, this is great, but the quality of voice over the Internet is terrible." This may be true today, but the quality is improving daily. In a VoIP system, delays are introduced by the VoIP gateway (one on each end) as well as by the end-to-end delay across the wide area network(s). Today's gateways introduce delays as high as 400-600ms. Multiply this by two (since there is a gateway at both origination and destination) and you've got serious delay. But Digital Signal Processing (DSP) advancements are slashing this down to the 100ms level. Motorola is a leader in this area-with claims of VoIP processing speeds estimated at the 40ms level.

At the wide-area IP network level, performance issues are being addressed on several fronts. At the raw bandwidth front, advances continue to be made in high-speed access and backbone routing, switching, and transmission capabilities. This is the year where high-speed access networks get real. In the U.S., xDSL and cable modems are expected to move from tens of thousands of subscriber lines to hundreds of thousands; hitting the million mark in 1999. Faster dialup, ISDN, T3 frame relay and ATM access continue to rollout. At the physical and transmission level, WDM and SONET deployments are bumping raw transport by orders of magnitude without service providers having to bankroll time-consuming fiber installations. The routing and switching fabrics are cranking up with next generation ATM switches and a new range of multiple-gigabit routers from companies such as Avici Systems, Inc., (Chelmsford, MA), GigaPacket Networks Inc. (Littleton, MA), NeoNet LLC (Westboro, MA), Torrent Networking Technologies Inc. (Landover, MD), Berkeley Networks Inc. (San Jose, CA), Juniper Networks (Santa Clara, CA), Ascend Communications (Alameda, CA) and Cisco. All of this is adding up to high-performing public IP nets. MCI, for example, is guaranteeing less than 95ms delay across its Internet backbone; Concentric Network Corp. (Cupertino, CA), a new entrant, offers serious credits if it doesn't hit its 125ms guarantee. In practice, these IP nets are performing at levels which are quickly closing the gap with the VOTS nets (at 50-70ms). Layered on top of this raw forwarding power in the IP nets is the rapid implementation of H.323 standards-protocols for transmission of video, voice and data over an IP network-and more control over quality of service via QoS, Class of Service (CoS), and Quality of Access (QoA) techniques.

As gateway and network latencies continue to fall, VoIP performance will soon surpass that of conventional satellite phone calls. And, once voice is digitized, it can be manipulated in many ways-for example, to produce CD-quality voice. At some point, packetized voice will become the new "toll-quality" standard that traditional telephony calls aspire to reach.

So, What's the Delay?

Despite the clear march towards packetized voice, the switch will not flip overnight. The worldwide data infrastructure has quantum leaps to make in capacity, reliability, QoS, billing and calling features like call routing. In essence, the data world has a lot to learn from the telephony world, and we expect a lot of leadership from the traditional telephony/transmission vendors, like Lucent and Nortel, masters of these techniques who are now very serious about IP nets.

VoFR has the installed-base edge in the enterprise, but will be increasingly encroached at the low end by VoIP and at the high end by VoATM. Both VoIP and VoATM have critical dependencies that could stall their adoption and boost the life of frame (remember ATM to the desktop?). For IP, it will be the expected leap in performance and latency controls. And today's first generation VoIP products and standards are far from complete. With some exceptions, most of today's VoIP gateways are PC-based, with low port counts, virtually no built-in reliability, large footprints, and are voice-centric, with little in the way of advanced telephony features. The next generation hitting the market is bringing carrier-class scale and reliability, SS7 integration, and integrated voice+fax+data support. There's virtually no interoperability today, but standards bodies are furiously tackling inter-gateway signaling, call control, address translation between IP addresses and telephone numbers, security and billing.

In the meantime, vendors from all corners of the internetworking world, the telephony/transmission world, and startups, are jockeying for position. Hold onto your hats as we face architectural upheaval during yet another round of piranha-like integration-this time, of VoIP capabilities into every imaginable network product. VoIP gateways have already made their way into FRADs from Hypercom, Nuera and Motorola. Prime examples of integrated products in the pipeline are integrated remote access servers, from vendors like Ascend, Lucent, 3Com Corp. (Santa Clara, CA), and Cisco, adding VoIP and SS7 capabilities; and VoIP modules being added to PBXs and COs from companies like Lucent and Nortel.

For VoATM, possible stallers include technical issues around converging on a scaleable approach to map IP QoS to ATM, as well as market issues. For enterprise nets, public ATM services have been caught in a chicken-and-egg mode: enterprise users won't sign up unless the service is ubiquitous, service providers have been waiting for enterprise demand. Recently, this is starting to break open. As for VoATM as a means to carry CO-switch voice traffic inside of service provider nets, the question is whether this will scale enough to move from trials to production.

The Real Power of Voice/IP

So far, we've discussed the operational efficiencies of using packetized voice. But in the case of VoIP, don't miss the real power. V/IP is not just a cheap way to run phone calls. It is ushering in a whole new range of voice services that add tremendous value to an enterprise. Many of the V/IP start-ups are focusing on the application side of this market -- this is the stuff that will drive a boom in all manner of new business-transforming applications. There are endless possibilities, but three very real, already-on-the-shelves applications highlight the power of this nascent market.

The first is use of voice/IP applications to spur e-commerce. The key example is voice-enabled Web pages. How it works: a consumer browsing your web site is interested in buying your product. Rather than disconnecting the PC connection; then dialing your 800 number to get more info. The consumer clicks on your customer service web icon ... a VoIP connection is made to your agent who can talk directly with the customer ... push more Web pages or order forms to the customer ... and close the deal on the spot. Add a video camera, and you're interacting near face-to-face as well. Bottom line: more touch equals more sales. Lucent's Internet Call Center is a prime example of this application.

The second class of applications address operational efficiencies. Workgroup collaboration and unified messaging are two already available that can completely transform workplace productivity. These become as mainstream a part of the business day as cell phones, e-mail, and audio- conferences today. Microsoft's (Redmond, WA) Netmeeting and Lucent's MMCX point the way to next-generation workgroup collaboration. Here, you combine the power of phone calls, video conferencing, electronic whiteboarding and collaborative editing in one platform. With unified messaging now a reality (like Lucent's INTUITY Internet Messaging), a user gains full portability. Listen to voicemail while surfing the web ... have e-mails read over the phone while on the road ... have faxes automatically routed to the hotel or email ... etc. These applications address what's needed for a businessperson to successfully work in today's hyper-speed business world: mobility and 7x24 convenience.

Cashing in on Convergence

So as we move to this new converged world where the Internet meets telephony, what is the best way to navigate through the next few years?

For the enterprise user, it boils down to finding the cheapest way to carry fully-featured voice at near-toll- quality while still tracking usage and costs and minimizing management. And, more importantly, leveraging the next wave in strategic business-transforming applications so your company is not left in the dust. Service providers face equally daunting decisions: which type of telephony and converged V/IP services to offer and what is the best internal infrastructure to run them over? Hello! It's your wake-up call. Hundreds of billions of dollars are on the line.

Copyright 1998, Susan Almeida

Strategic Networks, All Rights Reserved

This article appeared in the Spring 1998 Issue of 21st

21st, The VXM Network,