Colby Cosh points to an article by Bob Cringely in which he argues that telcos and cable companies might give higher priority to packets from their own voice-over-IP services, thus "squeezing" the packets used by independent voice-over-IP services. Colby Cosh thinks this makes no sense since telephony over IP takes so few bandwidth and that, unless the Internet service provider dares to strip broadband users from almost all their bandwidth, there will always be enough bandwidth to use voice-over-IP. Telephone calls do indeed take relatively little bandwidth. A standard telephone call has an analog bandwidth of 3 kHz and is transmitted over a 64 kbps channel. A voice-over-IP telephone call will often take less bandwidth; depending on the parameters used for sampling voice, it can take about 20-30 kbps.
But the big challenge with voice-over-IP is not bandwidth, it is to ensure that latency is low enough and reliably so. As a rule of thumb, for a telephone call to be considered of a good quality, its one-way delay (the time elapsed between the moment when the speaker says something and the moment the listener hears it) must be lower than 150 milliseconds:
If delay times are longer than 150 ms, it becomes difficult to know whose turn it is to talk. This adds a level of discomfort or even stress to the conversation. People are more likely to interrupt each other and certainly more likely to cut their call short.After about 200–250 ms of delay, any interactive conversation needs perseverance; people are forced into an awkward, half-duplex conversation, much like talking on a one-way radio.
For VoIP calls, especially ones made with the free services that depend on the Internet from end to end, a network designed for data is being used for voice. While the network may usually have far more than 64 kb/s in available bandwidth to accommodate a new phone call, congestion can arise at any moment and cut the data rate to almost nothing—at least briefly. And the slightest hiccup in the connection, at any point, results in dropped packets and momentary gaps in the conversation.The same hiccups mean nothing in e-mail, where a delay of several seconds is unnoticeable. Although telephony doesn't involve large amounts of data, the time constraint makes it far more demanding than most other Internet applications.
There are three things to worry about in an Internet phone call. Latency occurs when data packets are delivered too slowly—usually because of network congestion. Jitter is a variation in the delays of packets—some arrive on time or only a bit late; others, sent just before or after, arrive much later. Finally, when packets are extremely late, the network drops them, resulting in packet loss. Latency and packet loss can create awkward momentary silences in a phone conversation or make it seem that one party is interrupting the other. These delays can cause echoes and other odd sound effects.
However, these quality-of-service problems are not insurmountable. Indeed, "network engineers have already devised some clever methods to guarantee a minimum bandwidth for a particular application." For example, there are network architectures in which "data packets are assigned labels by specialized routers" and "[t]hese labeled packets are forwarded not by the usual algorithms that best serve the Internet's overall traffic needs, but according to decisions that are tailored to the labels." This allows "[v]oice packets [to be] given special priority that ensures that the congestion preferentially affects applications other than VoIP calls."
Of course, if the telephone or cable company only tags its own voice-over-IP packets as voice packets, while leaving packets from independent voice-over-IP companies tagged as regular data packets, then this is exactly the hypothetical situation Cringely has been describing and this could leave these independent companies with serious quality-of-service issues.