Lessons Learned
BY LAURENCE J. FROMM
As we approach the two-year milestone of the commercial introduction of the world's
first IP telephony gateway, it is an appropriate time to reflect on some of the lessons
learned in the life of this nascent industry.
VOICE COMPRESSION DOES NOT MATTER
Many of the early IP telephony pundits proclaimed that the major cost driver for IP
telephony is voice compression, which can reduce bandwidth use 10-fold over circuit
switched telephony. Besides ignoring the deployment of digital circuit multiplication
equipment in telephone networks throughout the world to achieve the same level of voice
compression on circuits, this prognostication failed to anticipate the deployment of some
of the next generation carriers.
It has been widely reported that Qwest, for example, is deploying G.711 (uncompressed
64 Kbps) voice coding on their IP networks. Many expect that enterprise deployments will
also favor G.711. This does not mean that bandwidth is free; rather, it means that some
network operators would prefer to invest their capital in networking technology rather
than gateway technology. Some wide-area network operators get a better return on their
capital investment by deploying the latest in wave division multiplexing equipment than in
putting more and faster digital signal processors (DSPs) at the edge of their networks.
Similarly, some enterprise deployments will prefer to address bandwidth needs by investing
in gigabit Ethernet infrastructure rather than deploy DSPs at the edge to compress voice.
VOICE COMPRESSION DOES MATTER
Other prognosticators predicted that very low bandwidth voice coders (3 Kbps or less)
would never have an impact on the market. While this prediction is correct to date (to the
chagrin of some startups that aspired to sell large volumes of low bandwidth voice coders)
the jury is still out on this one. The main argument against very low bandwidth voice
coders is packet overhead. In a packet network, of course, information travels in packets.
Internet Protocols
The Internet Protocol version 4 (IPv4) header is 20 bytes long (containing destination
address, source address, length, type of service, time to live, etc.) IPv6, with expanded
address fields, takes 34 bytes. Within IP packets, IP telephony travels on User Defined
Protocol (UDP). UDP headers are 8 bytes long, including source port, destination port,
length, and checksum. Within UDP, Real Time Protocol (RTP) is used. RTP, of course, has
its own header of 12 bytes, including sequence and time information. Finally, within RTP
the payload is carried. The header overhead, then, for IPv4 is 40 bytes. The payload for
G.729a, for example, at one frame per packet is 10 bytes. (G.729a compresses voice to 8
Kbps with a 10 ms frame size.) Hence, 80 percent of the IP bandwidth used by a G.729a
session at one frame per packet is for overhead!
Is Smaller Better?
By throwing a lot of DSP power and complex compression algorithms at the problem, one can
get good voice quality at as low as 2 Kbps (or lower!) - a 75 percent reduction in
bandwidth use from G.729a and a 97 percent reduction from G.711. But because of the
immutable packet overhead, this investment reduces the total bandwidth need by only 15
percent (from 50 bytes per packet to 42.5 bytes per packet). Of course, one could send
packets less often - every two frames, or every three frames - order to reduce packet
overhead and bandwidth utilization. Reducing packet frequency, however, increases latency
and degrades voice quality.
Hope Springs Eternal
The good news is that packet overhead is immutable no longer. The IETF is working on a
protocol for link level compression so that 40 byte packet overhead for IP/UDP/RTP can be
reduced to just two bytes. Hence, payload optimization much more directly translates into
bandwidth optimization, renewing interest in some network operators for lower bandwidth
voice coders. Indeed, some next generation telcos are already deploying routers that
support link level compression in order to minimize packet overhead.
To Compress Or Not To Compress?
What about the argument that paying to compress voice is a relatively poor investment
compared to increasing network bandwidth? It depends where one sits. Some facilities-based
operators do indeed find that investing in networking equipment yields a higher return
than putting lots of DSPs at the edge of their networks. But operators that lease
bandwidth find that investing in equipment (gateways with low bandwidth voice coders and
routers with link level compression) to reduce their bandwidth needs yield a good return
in reduced bandwidth costs. In addition, DSP price/performance continues to improve at a
very rapid pace. Hence, equipment vendors need to deploy on a flexible platform as
technology economics play out on several planes at once.
REARCHITECTING THE NETWORKS
Those of us that were promoting IP telephony two years ago received all kind and manner
of counter arguments. Those of us that stayed away from the "cost driver is
compression" argument more typically used the "cost driver is converging
networks and reducing deployment and maintenance expense" argument. The argument is
familiar now: Instead of deploying and maintaining two networks (one for voice, one for
data), putting voice on data networks allows operators to focus their energy and expenses
at deploying and maintaining a single network. Two years ago the simplest and
hardest-to-refute argument was: "If putting voice on data networks is more cost
effective than running parallel networks, then the big carriers would be doing it. How
come they are not?" In hindsight, the answer is obvious: "They are."
Announcements and deployments from Qwest, Level 3, Sprint, and others make clear that many
carriers have been planning and investing in converged networks for some time.
The Surprise Early Entrants
The early money was on the Internet Service Providers (ISPs) being among the first IP
telephony service providers. In 1996 and 1997, there were at least a half dozen
conferences aimed at "Internet Telephony for ISPs." Two years later, however,
almost all the service providers are from other areas:
- Entrepreneurs: Delta 3, ITXC, Access Power, Planet Telecom, and dozens of others.
- Facilities-based operators: All the big RFP action now is at the major telco level.
- Telecom resellers: IDT, USA Global Link, and dozens of others.
Upon reflection, the last one should not be a surprise: ISPs have some of the right
equipment, but mostly the wrong customers. ISPs have customers with discretionary money to
spend - not exactly the right profile. Telecom resellers already have customer and billing
relationships with companies and individuals looking for ways to reduce their telecom
bills.
THE PATENT PROBLEMS MUST BE WORKED OUT
I have to admit that I was at the VoIP/IMTC meeting in Newark in March, 1997 as certain
representatives of companies that claimed patents for G.723.1 assured the industry,
basically, "If you vote to approve G.723.1 as a standard, we will work out the patent
issues. Trust us." Well over a year later, after the IMTC did, indeed, approve
G.723.1 as the baseline coder for IP telephony, the patent situation for G.723.1
specifically and voice coders more broadly is hardly clear or reasonable. Make no mistake
- the patent situation is slowing industry growth, and will continue to slow industry
growth.
WE CAN MAKE A DIFFERENCE
Viewed from a regulatory perspective, IP telephony would seem to be in a perilous
position. Shifting paradigms. Changing economics. Disruptive technology. Upstart
competitors. Very rich, very entrenched, very politically savvy incumbents. The only thing
going for IP telephony is the power of right. That, as they say, and $3.45, will buy you a
venti mocha at Starbucks. In this environment, some true believers formed the VON
Coalition in 1996 as a grass roots effort. And, VON has enjoyed enormous leverage for a
relatively small effort. It proves the power of right, with diligent and cogent follow
through, can have an impact with policy makers, particularly when it aligns with the
agenda of certain agencies or administrations or legislators. Thanks to the VON Coalition,
for example, in the U.S., both the FCC and the National Telecommunications and Information
Administration (the telecommunication policy advisor to the Clinton administration) are
aware of IP telephony, conversant on many of the regulatory issues, and engaged in dialog
with the industry. The VON Coalition had a major impact in avoiding - for now - a negative
regulatory position from the FCC in their April 10 report to Congress on universal
service. (The VON Coalition has become more formal and is now accepting new members. Visit
www.von.org for more information.)
SUMMARY
In August 1996, VocalTec commercially shipped the world's first IP telephony gateway. It
supported two phone lines, and was based on a Windows NT computer with computer telephony
boards originally designed for something altogether different.
Two years later, we have seen an explosion of client products converge to a small
handful; dozens of vendors of gateways bring product to market; dozens of service
providers; multi-billion dollar startup next generation carriers; and multi-billion dollar
mergers between telecom equipment manufacturers and networking companies. One shudders to
project the world two years hence. Hmmm, perhaps in a future column
Laurence J. Fromm is vice president, new business development for Dialogic
Corporation. Dialogic is a leading manufacturer of high-performance, standards-based
computer telephony components. Dialogic products are used in voice, fax, data, voice
recognition, speech synthesis, and call center management CT applications. The company is
headquartered in Parsippany, New Jersey, with regional headquarters in Tokyo and Brussels,
and sales offices worldwide. For more information, visit the Dialogic Web site at www.dialogic.com.
|