CNET también está disponible en español.

Ir a español

Don't show this again


VoIP needs a reality check

Consultant Fred Goldstein believes conventional wisdom on voice over Internet Protocol overtaking traditional phone networks needs a rethinking.

    Conventional wisdom has it that the telephone network will evolve away from old-fashioned circuit switching and be replaced by voice over IP.

    Investors have ponied up to the bar to fund VoIP companies; most major telecom manufacturers have largely abandoned development of new circuit-switching and time division multiplexing (TDM) products. But before we nail the coffin shut on the legacy of the digital telephone technology that was so widely hailed two decades ago, let's examine the strengths and weaknesses of VoIP.

    In looking at VoIP as a technology, it's important to distinguish between the cost of building or running a network and the price a service provider charges. VoIP benefits from the Internet's pricing model. There is rarely a charge for usage or for distance. The incremental price of a call thus seems to be free. That's not quite true, of course; additional loading can lead to degraded performance, or require a fatter, costlier access pipe.

    VoIP benefits from the Internet's pricing model.
    Telephone companies, on the other hand, are masters at charging for calls, often with widely varying rates depending on who calls whom, when, where and how. That annoys customers, but at least it's profitable. Services like Vonage bypass this, provided that their subscribers have already paid for broadband access. Regulators should recognize that the 1930s-era long-distance payment system is thus obsolete, but not by arbitrarily favoring one technology over another.

    VoIP is actually far more complicated than TDM, and thus more costly to implement. With VoIP, every packet needs header processing and buffering. The protocol stack is several layers deep. With TDM switches, on the other hand, the switching fabric is trivially simple. Bits are simply clocked on and off a time-division bus, or switched by clocking them through a memorylike crosspoint chip. Zarlink Semiconductor, for instance, makes a TDM switch chip for less than $100 that can connect 32,768 calls in to any of 32,768 outputs. Of course hardware and chip manufacturers would rather sell a VoIP solution--it's worth more to them.

    Get Up to Speed on...
    Get the latest headlines and
    company-specific news in our
    expanded GUTS section.

    Sometimes, though, VoIP can be cheaper to implement than TDM. One reason is the cost of legacy TDM network interfaces. TDM components are often made in lower volumes, or may also need extra hardware. But even then, high-capacity TDM switch ports aren't all that expensive, given their capacity. Amortized on a per-minute basis, TDM switching cost is a small fraction of a penny.

    TDM's lack of packet headers also makes it more bandwidth efficient than VoIP. Sometimes VoIP is compressed to make up for that, but compression reduces voice quality. If IP bandwidth supposedly is too cheap to meter, then why compress voice from its native 64 kbps down to about 9 kbps? The best-sounding VoIP systems use uncompressed voice, but need about 100 kbps per call, taking into account both header overhead and IP's buffering requirements.

    Another purported advantage of VoIP is the ability to have one network carry both voice and data. It's clear that having a single network below the IP layer makes sense. But packet switching was invented precisely because data has very different traffic requirements (bursty, loss-tolerant) from voice (steady, loss-intolerant).

    IP data is most efficient on a network that is not good enough for voice.
    IP data is most efficient on a network that is not good enough for voice. TCP's (Transmission Control Protocol is one of the main protocols in TCP/IP networks) flow control is dependent on packet loss--networks will discard packets, they're supposed to, but that does impair voice. Economy of specialization often trumps economy of scale. ATM (asynchronous transfer mode) and MPLS (multiprotocol label switching) are both ways to address this issue above the TDM layer, but below IP.

    But those details aside, VoIP developers have made real contributions to telephone network technology. Modern "softswitch" architecture, for instance, was largely developed for VoIP. Softswitch software is capable of routing phone calls and other voice features typically handled by massive and expensive hardware.

    Network control protocols such as SIP (Session Initiation Protocol), MGCP (media gateway control protocol) and Megaco (H.248) are real innovations, providing network operators with real flexibility, much more than earlier Signaling System 7 networks provided. And the best part is that modern switches don't all depend on having the voice encapsulated inside IP. Megaco can be used with VoIP, ATM or even TDM networks, on a flexible mix-and-match basis.

    It's not fair to compare VoIP implementations with the predominant TDM systems found in telephone networks. The Lucent 5ESS and Nortel DMS switches used by the regional Bell companies are basically legacies of the VAX-11 era, decades-old technology that survives because of its huge base of embedded software. Modern switches, built with up-to-date hardware technology and softswitch internal architecture, can often support both TDM and VoIP, and sometimes ATM as well, at a fraction of the size and price of these legacy behemoths. But their lower cost is not a result of using VoIP; rather, it's the difference between a design from 1980 versus one based on today's components. Does anyone remember what a 1980-era packet switch looked like?

    There are still good VoIP applications, both in the enterprise and in carrier networks. But don't write off TDM or ATM yet; they're looking better than ever, and are still the better choice for many applications.