What does it mean that Ricochet is a ``data network'' and cellular is a ``voice network''?

Ricochet is a ``data network''

The problem with telephone calls is that, once a voice call is admitted onto the network, the network must always transmit its information immediately. If information takes an unpredictable amount of time to pass through the network, the listener will hear dropouts of silence in the middle of sentences and won't be able to understand the speaker. It's possible to tolerate a transmission delay of 0 - 2 seconds by delaying playback for two seconds, but telephone users don't like this because, well, consider the following conversation, as seen from Alice's perspective:

Alice: says to the baby tomato, ``catchup.''

[...]

Alice: What, don't you get it? see it's Catch Up, or it---

Bob: HAHAHAHAHA!

Alice: Hello?

Alice: Hello?

Bob: Hello.

Alice: Oh, I thought you wer---

Bob: I _said_ ``Hello.''

Bob: Yeah. I'm here!

[...]

Alice: oooo-kay.

[...]

Alice: Hey, do you want to g---

Bob: so, um, what are you do---huh?

Email, SMS, web pages, and ring tones don't have this problem. Telephone biggots will say that packet-switched networks like Ricochet can't handle ``voice'', and are permanently doomed to handle easy information that doesn't care about latency. Only circuit-switched telephone networks can handle voice.

Ricochet is a ``data network,'' but not really.

There is a very simple workaround for Ricochet-style all-packet networks: send all the voice data immediately, and whenever there is any leftover bandwidth, use it to send email, SMS, web pages, ring tones, movie trailers, whatever. This is called QoS: for a fixed amount of information, voice information receives higher Quality Of Service than ring tones and web pages.

3G and GSM's GPRS extension accomodate both types of information by mixing circuit-switched and packet-switched data. I presume the packet-switched data just uses up any bandwidth that isn't already assigned to a voice circuit. This sounds to me almost exactly like priority-based QoS. There is a lot of talk about time slots and circuit-vs.-packet, but intuitively it is basically the same idea: ring tones and email use up whatever bandwidth is ``left over''. There is still a distinction between simple priority-based QoS and circuit-switched, especially when almost all the traffic is voice, but I think it's clear where all this is headed.

Hui Zhang's very old but still under-implemented HFSC algorithm (which is part of ALTQ in the KAME IPv6 stack (no, ALTQ is not a FreeBSD invention. It was imported along with KAME.)) is much more ambitious. It treats bandwidth and low-latency as two network resources that can be allocated independently. One can give bandwidth to movie trailer downloaders, but take low-latency away from them and give it to voice users. HFSC can make and keep latency promises to voice users without resorting to a primitive circuit-based architecture reminiscent of mechanical switches. (Think of Sally Floyd's ``Class-Based Queueing'' as a goal, and Zhang's HFSC as a working implementation.) More importantly, HFSC's extra complexity lets it make and keep a lot more promises on a given network than priority-based QoS, which only works well when the overwhelming majority of traffic is low-quality (high-latency). As we are already seeing with the simpler priority-based QoS voice-IP networks, HFSC is likely to accomodate voice with less expensive switching equipment than the circuit-switched alternative. However, there are still some challenges.

Cellular is a ``voice network''

Cellular is a ``voice network,'' but not really.

If you've ever seen an Asian ex-patriot with a gaudy blue LED capping his or her cheap Nokia celfone, you may have noticed the mesmerizing way that the LED only lights solid while the person is talking. You might also have noticed that sometimes you can only hear a cellular caller's background environment while he or she is talking---when your partner is silent, your partner's room suddenly seems silent, too. The latest voice compression algorithms seem to be ``variable bit rate,'' meaning the algorithms don't use up all the bandwidth reserved for them on their circuit unless they need it. DVD's work the same way, where action-packed scenes can use more space on the disk than still pans. VCD's, in contrast, always spin at the same speed even if nothing is happening in the movie.

Most of the voice codecs I've heard about, including GSM used on most gaudy phones, are constant-bitrate, so the actual variation may well be something simple and lame like ``don't transmit anything below a certain fixed volume,'' but the concept holds even though the bitrate variation might not be as mathematically arcane as video compresion.

Why do celfones do this? If the circuit is reserved for your conversation anyway (and you're being charged by the minute for silence), why not use plain constant-bitrate voice compression that uses up the same bandwidth in your moments of silence?

For one thing, it saves handset battery life.

For another, CDMA networks degrade gracefully, so if one phone isn't transmitting, other phones' signals will arrive at the tower more cleanly.

But, that said, I would be surprised if carriers didn't find some way to transmit ring tones and web pages in the silent intervals of your conversation, even though a circuit is ``allocated'' to your conversation. If they did that, low-quality data isn't just using left-over bandwidth unreserved by the circuit-switched half---it's seeping into the gaps and pauses of your conversation as well. How finely-grained does this bandwidth-stealing have to get before we admit that we're not really using a hybrid ``voice''-circuit/``data''-packet network at all; we're just using a plain old packet-switched network with QoS?

Cellular is a ``voice network,'' but what is ``voice'', really?

Nextel offers a ``direct connect'' feature that emulates half-duplex two-way radios, with a ``push-to-talk'' button. Beeps and burps replace the old tradition of saying ``over'' when you're done talking.

Recalling the fragmented conversation between Bob and Alice, the pair would have had no problems if they'd said ``over'' to yeild the conversation to the other person. Nextel can make direct connect traffic wait. They can add a two-second playback delay without annoying people. Indeed, if you direct-connect to another Nextel subscriber in the same room, you'll notice there is a couple seconds of delay.

Direct-connect is billed like low-quality packet-switched data: all but the cheapest Nextel plans include ``unlimited Direct Connect minutes.''

Traditional cellular networks have a vested interest in making sure we continue communicating mostly with phone calls, because their networks can't accomodate any other medium. This has to change. Sprint's ``wireless web'' looks quite shabby, now that AT&T's and Nextel's WAP equivalents allow unlimited use with no incremental cost. Carriers in Japan are already competing mostly based on the relative merits of their so-called ``data'' services.

Consider the following services:

If only we had such diversity in the US! Suddenly I'm remembering a 5-minute BBC clip about 3G, in which some carrier executive showed off a robotic voice driving-directions application that seemed to me like it would work fine on any celfone with a speakerphone and GPS, nevermind the 3G spectrum license. Cut to a high school amphitheater.

Student: ``Will 3G let people send email from their phones?''

Interviewer: [silence]

so. Given the eight applications above, what do we do with this idea that there are ``data'' services and ``voice'' services? Are Direct Connect and phone calls voice services, while everything else is a data service? I don't think so. Direct Connect makes the same sort of demands on the network as web browsing, which is very different from a phone call (recall Alice and Bob's conversation).

I think there are high-quality services and low-quality services. Phone calls and video conferencing are high-quality, while everything else is low-quality. Once we get past the initial five years of shameless price-gouging, I think that is how carriers will ultimately price the services. They will charge for high-quality bandwidth.

In such an arena, I think phone calls become the least interesting medium. Phone calls are more invasive than the low-quality services, and they're not more efficient than Direct Connect once you get the hang of it. If I have to pay for high QoS, I will want video, even if it's grainy and jerky, even if Alice uses ``video mute'' a lot. People say phone calls are ``more personal'' or something, but ultimately I think we'll find that phone calls represent a middle-ground between videoconferencing and SMS/email that isn't very appealing. What happens then to the so-called ``voice network''? I suspect it will get a new name.

so, stop telling me that Ricochet is a ``data network''! Yes, of course it transmits data. What else is there?

It makes more sense to say that Ricochet is owned by a data-networking company, and cellular carriers are owned by voice-networking companies. They don't have much competence or inclination to work in each others' fields. This observation has much different implications than the just-debunked technical analog.


rants / map / carton's page / Miles Nordin <carton@Ivy.NET>
Last update (UTC timezone): $Id: ricochet-is-data.html,v 1.1 2005/10/30 02:58:57 carton Exp $