Universally accepted standards: What a pain! Communications-hardware vendors all used to use proprietary compression technologies to squeeze voice traffic through frame-relay networks. But last spring, the Frame Relay Forum announced an interoperability standard, FRF.11, that seemed to finally put an end to all that. For the first time, companies could shave 35 percent or more off their intracompany long-distance bills without committing to a single vendor's hardware and software.
The key to these glad tidings was G.729A, a voice compression/decompression (codec) protocol. Hardware vendors almost universally agreed that G.729A was good enough for toll-quality voice. In fact, the protocol provided nearly the same voice quality as the Public Switched Telephone Network (PSTN), which cost thousands of dollars a month more. Service that had previously cost a company $120,000 annually cost only $40,000 with frame relay. A bonus was the fact that interop
erability could convince antsy customers that frame relay was mature enough to trust for both data and communications.
But the plot was thickening. The proposal was unraveling even before the ink was dry. France Telecom North America (FTNA), the University of Sherbrooke, Lucent Technologies (based on work done at Bell Labs), and other companies had all provided some technology pieces to G.729A; each contributor now wanted a piece of the licensing action. Telephony vendors could end up paying dearly if they incorporated the codec in their systems.
How dearly? That was the other problem. Intellectual property claims were nearly impossible to sort out, leaving vendors in the position of ignoring G.729A in favor of each one's own proprietary codecs. So much for inter-operability.
This was all the more painful because frame relay has improved its voice quality greatly. Two years ago, some disparaged frame relay because of annoying delays in two-way conversations. But although it's still not perf
ect, frame relay today almost equals PSTN under pristine network conditions.
Worse still, frame-relay fans must quickly sort out their problems now that a competitor, IP, has emerged. As corporations construct IP intranets, many see the next logical step to be voice services on those networks. Maybe. Headlines may buzz about voice over the Internet, but anyone who has listened to such calls knows that, right now, they're more a parlor trick than a Fortune 1000-level solution. The real potential for voice on IP is for calls within an enterprise, to connect headquarters with remote subsidiaries -- the very turf that frame relay has been trying to claim.
Now, as firms wonder if they should find alternatives to expensive long-distance service, they're also asking which
is best. Frame relay and IP both offer hope for tomorrow's single-pipe data/multimedia dream. The big question: Is either technology mature enough for you to commit now? Here's how to decide.
Giving Frame Relay a Voice
Frame relay's variable-size packets efficiently do what their original design dictated: move blocks of data across WANs. Hardware at each end of a link handles error correction and flow control, so frame relay avoids the overhead burden of its older cousin, X.25. Plus, service providers typically sell frame-relay service for less than the cost of T1 or fractional T1 lines. That can add up to significant savings for data-intensive businesses.
If a company uses a public frame-relay network -- offered by AT&T, MCI, Nynex, Sprint, US West, Wiltel, and others -- it taps into the public frame-relay "cloud." That saves it from having to buy its own routers and switches. (See the sidebar "Saving with Frame Relay".)
A bonus is
dual dial tone
, which reduces long-distance charges further (see the figure
"Leaky PBXes"
.) This feature -- disparaged by long-distance providers -- allows subscribers to call remote numbers for the price of a local call. A caller in New York,
for example, dials a prefix number, sending the call to the frame-relay network, and hears the first dial tone. The caller then dials an outbound extension at a PBX in a remote office in, say, San Francisco, and hears a second dial tone. Then the caller can dial a remote customer as if making the call from San Francisco. Companies are reluctant to admit they do this, but one spokesperson said, "We hear it works just fine."
Coastal Construction Products connects its Jacksonville, Florida, headquarters to six remote offices using frame relay. In 1995 came the decision to add voice to the fractional T1 and 64-Kbps data lines. "We knew that if we could run voice over those networks, we'd pay for any additional equipment from our reduced long-distance phone charges," explains Jack Caven, MIS manager.
The company spent about $55,000 for Micom equipment, including its software FRAD. (FRADs -- frame-relay assembler/disassemblers -- package data into frame-relay packets; today, vendors sell integrated har
dware that combines traditional FRAD with routing, switching, and multiplexing.) The frame-relay link itself cost about what the former leased line did.
Caven estimates the company recouped its investment in 18 months, but reduced costs were not the only benefit. "We began to have better communications within the office," he explains. "Because we could call Miami as easily as calling the office next door, our people began to communicate more -- not long conversations, but more short conversations to check availability of products." Conference calls among remote staff members had previously been expensive, with the telecom provider patching calls together. "Now conference calls are free," Caven adds.
With Coastal's circa-1995 equipment, voice codec algorithms are old and compression is only to 16 Kbps (versus 8 Kbps, the current standard). Consequently, the system is primarily for intra-office calls.
IP's Say
Voice over IP offers similar savings: Voice gateways produce a voice/fax la
yer on an IP intranet. Gateways can be simple. For example, Micom's V/IP is a standard ISA card that plugs into a business-class PC connected to a PBX and a network. The V/IP card digitizes voice and puts it in IP packets (at the sending end) and unpacks the IP packets (at the receiving end). Similar PCs run at each remote facility.
IP-based intranets move voice and faxes through an enterprise well. And the Internet can be an important low-cost link for one-way calls, such as checking voice messages or sending a fax while on a trip. And IP also provides dual dial tone.
IP gateways create a directory of phone numbers and IP addresses associated with each destination gateway. To place an IP-network call, users need only to dial a single-digit access number to reach the IP network, a number to reach the destination office, and, finally, an individual's telephone extension. The gateway sets up the call (often in 1 or 2 seconds).
Vienna.way, a call-processing server in the Vienna Systems product
family, performs traditional PBX duties so that users can place or receive calls through their PC's IP gateway or a special serial-interface telephone. The server runs on Pentium PCs using Windows NT or Unix. Multiple Vienna telephony cards (with four or eight DSPs) in each PC can support up to 96 simultaneous calls to the PSTN. To handle more users, you can string together multiple servers.
VocalTec's Telephony Gateway 3.0 provides similar capabilities using VocalTec software and Dialogic telephony boards. Unlike Vienna's product, the VocalTec gateway runs only on NT (using 200-MHz or faster Pentium PCs). VocalTec recently announced Atrium, intranet software that conferences multiple callers, even if some use traditional telephones and others use PC connections to the IP network. The software, which was due to ship this summer, costs $2400 for a one-user license.
FTNA is currently testing voice over IP, using Micom's gateway boards. The project's first goal is to add telephony to the Sprint frame
-relay data/e-mail network among the company's New York, Chicago, and San Francisco offices.
FTNA focuses on voice and fax over IP as part of intranet experimentation. It gets first-hand experience in future development of voice-over-IP services. "If we're running voice over IP at layer three [of the OSI model], we can use frame relay or ATM [asynchronous transfer mode]," says Jean-Francois Mulé, manager of information systems.
ABC Bücherdienst, a Regensburg, Germany, bookseller, is testing an innovative use of voice over IP. The company recently hired sales agents in Boca Raton, Florida, to handle inquiries from European customers after normal business hours. When Bücherdienst closes, the headquarters PBX routes sales calls over a leased line to Florida. A customer talks to a German-speaking sales agent and may not even be aware that the call has traveled outside Europe, despite some degradation in voice quality, according to Michael Gleissner, managing director.
"We're loo
king at the Internet as a way to enable us to shift our operations internationally without a huge telecommunications cost," Gleissner says. "It's hard to get experienced people in Germany who are willing to work at night." The system, which is about a month into a six-month beta test, provides many standard telephony features, such as voice mail and fax capabilities, Gleissner adds.
Lucent provided all the hardware, software, and services in exchange for the bookseller's being a test site. The system was not "plug and play": Lucent has been upgrading software, as often as twice a week, to tweak voice quality. But the quality of voice over the Internet is still volatile, depending on how the conversation connects. "You quickly figure out the call isn't going through the normal telephone lines," Gleissner says. "But the quality is improving every week."
Speech Quality Evolves
Such encouraging implementations of voice over data networks are a recent change. "If you asked me a year and a half
ago if voice over frame relay had a chance to succeed, I would have said no," admits Tom Jenkins, broadband consultant for TeleChoice, a Verona, New Jersey, telecommunications consulting and market-research firm. "But I've changed my mind."
Voice over frame relay had earned a dubious reputation for high latency, taking over 100 milliseconds on average to send packets across a network. (The human ear starts to notice delays with latencies of 50 ms. At 300 ms, conversation becomes difficult. At 500 ms, conversations are annoying.) IP also suffers some of the same problems as frame relay.
A new generation of codecs, such as G.729A, not only compress conversations more efficiently but work with telephony applications that ingeniously interleave voice and data so that data sneaks through during the silences in all conversation. This is known as
silence suppression
. (A Bell Labs study found that silence can make up as much as 60 percent of a typical conversation.) Vendors estimate that silence s
uppression can reduce bandwidth requirements by about 3.5 Kbps.
Compression algorithms vary widely in how tightly they squeeze voice signals, ranging from 32 Kbps to 4 Kbps. (Note: Older codecs didn't compress smaller than 32 Kbps -- still a large chunk of bandwidth in 56-Kbps networks. Newer codecs, such as G.729A, offer higher compression -- 8 and 4 Kbps -- with relatively high voice quality.) Common codecs include pulse code modulation (PCM) and adaptive differential pulse code modulation (ADPCM), used by PSTN in the U.S. and by postal, telephone, and telegraph (PTT) systems in Europe. Both achieve high-quality audio with unnoticeable latencies. Unfortunately, they consume 64 and 32 Kbps, respectively, unacceptably high for a 56-Kbps frame-relay pipe.
Algebraic code-excited linear prediction (ACELP), a more recent technique, underlies the G.729A standard for 8-Kbps compression. ACELP can produce "near-toll-quality" sound in subjective tests.
G.729A is a cousin of G.723, the compression sc
heme pushed earlier (as part of H.324) by Intel and Microsoft for videoconferencing over PSTN. G.729A was developed because G.723 needs significant computing -- about 30 percent of a standard Pentium 100's power. G.723 also has a longer frame size -- 30 ms -- resulting in 90- to 100-ms latencies. With a smaller frame size of 10 ms and only 30- to 35-ms delays, G.729A became a simpler and higher-quality choice for voice applications. (Latency is 3 to 31Ž2 times frame size.)
Still, G.729A might never see widespread use. Even representatives from companies that helped develop the specification secretly hope a single company will offer a better alternative that's unsaddled by licensing problems. Nevertheless, networking analysts see this as a significant, if incomplete, gain. "It sends the right signals" that the industry is working to make voice over frame relay viable, Jenkins says.
Voice hardware also can control the flow of different data types. Because voice and fax communications break dow
n if there is too much delay, FRADs and gateways give them higher priority when packets travel through the frame-relay pipe. Data traffic remains in the sending hardware's buffer until the hardware sends the higher-priority packets.
Timing Is Everything
For the time being, and even if G.729A finds resolution, your safest choice still is to buy FRADs from a single vendor. Unresolved standards issues beg the most fundamental question for both frame relay and IP: Is the time right to combine your voice and data traffic?
The answer: Only for select applications. Neither technology offers enough quality for a large corporation to scrap traditional voice services. For example, Kevin O'Donnell, senior vice president at Florida food distributor Bonacker & Leigh, says he has noticed steady improvements in voice over frame relay in the past two years. But the quality still is not high enough "to talk with my best customers." For internal
conversations, however, especially after workers
get used to slight delays, frame relay is acceptable. "When you're saving $15,000 a month in long-distance charges, you get used to the sound quality pretty fast," O'Donnell quips. (See the figure
"How to Send Voice over Frame Relay"
.)
Mulé estimates that 90 percent of FTNA's New York-to-San Francisco calls travel over the IP network. His rating of IP voice quality? "Pretty good for our internal purposes." Unlike over the Internet, you can control the quality of calls over a private intranet.
Nevertheless, the relatively short payback times of hardware costs for voice over frame relay or IP make it easier for large companies to commit part of their voice services to one of these technologies. Long-distance savings can pay for a FRAD in half a year, so even if the technology changes in a year or two, you will probably recoup your costs. (Voice-capable FRADs range from about $2000 to $10,000, with most in the $4000-$6000 range.) And once you launch a combined voice/data
network, later transition to one-stop-shopping services is easier.
Just don't expect comfort in numbers: Today, according to Jenkins, voice represents only about 3 percent to 5 percent of the traffic over frame-relay networks. Despite alluring cost-savings potential, many companies are reluctant to commit to the technology because of equipment that changes rapidly and, ironically, a lack of standards. Also, frame-relay and IP can't yet provide advanced telephony features, such as cost allocation and minute-by-minute call tracking.
By the year 2000, even if standards work out, voice may not reach 10 percent of frame-relay traffic. This is partly because companies don't want their communications to be jeopardized by network downtime, and partly because of intractable turf wars between communications and information-systems (IS) managers. "Turning all your voice services over to the IS department isn't a good move for empire-building," Jenkins says.
How You Choose
If you have the pion
eer spirit and are ready to run part of your voice communications over a data pipe, first prepare yourself. Hardware vendors will bury you with proof of how each FRAD or IP gateway offers the best voice quality. Instead, do your own investigating: Bring loaner equipment into your organization for real-world testing. While service providers theoretically supply the same frame-relay services, customers note anomalies that are seemingly dependent on how calls get routed within a single enterprise. "We recently upgraded our multiplexers at each office, and invariably one or two of the sites couldn't use the frame-relay system," says O'Donnell. His suspicions focus on differences in how data travels to each location when it comes down from the fiber-optic backbone.
Jenkins suggests asking hardware vendors for recommendations for frame-relay service providers, and asking service providers for hardware recommendations. Try to test equipment from several vendors at the same time for comparisons. Audio quality i
s highly subjective, and comparisons might be the only way to quickly judge how well the hardware is delivering high-quality voice data.
Voice over IP, on the other hand, benefits from the widespread commitments to intranets that companies everywhere are making. "A corporation needs an intranet," Mulé says. "Once it builds a full TCP/IP network, adding voice to it becomes a good solution. You just put gateways where you need them."
Solution Sells
Once single pipes routinely handle voice and data, debate over frame relay versus IP will probably become irrelevant. Instead, service providers will sell results rather than technologies. Your traffic might use frame relay, IP, ATM, or combinations of each. "We won't see service providers selling a technology," says Jenkins. "Instead, they'll say, 'connect into my network; we'll deliver the data.'"
Where to Find
ACT Networks
Camarillo, CA
Phone: 805-388-2474
Inte
rnet:
http://www.acti.com
Data Networks Speak Up
