From: Michael Dillon <michael@memra.com>

When we discussed some of these things last year on can.infohighway I pointed out that if they could divide the traffic by type that they could get better trunk utilization by running IP between the CO's. This is the modem in every switch location scenario.

Thinking about this, this is not such a bad idea (I'm from a telco background). But doing it on the IP level, we would need to look into compression for transporting packets back to the nearest "routing center." There are a few ways to compress (from good old STAC and predictor to hardware-level compression like Magnalinks). No doubt that as ports are made available, people will fill them up to create more traffic. Also, a direct switch-port interface needs to be utilized. Without a design right in the central-office switch fabric, all you'll have is just a shorter path to clog up switch ports. Yes, you'll alleviate interswitch trunking, but you'll still end up with no dialtone if you're using twice as many ports to process an "IP call" (one port for the caller and one port for the terminal server).

Additionally, if they segregate voice traffic they could use compression technology to get as much as 4 times the conversations on a trunk. Gandalf makes boxes that will run 4 voice lines down a digital DS0 circuit and the quality is very good. You can tell there is compression there but it's better than long distance line quality often is.

MCI played with voice compression in the late 80's (called ADPCM). and they had pretty bad results. They had equipment which was supposed to detect the initial sequence of modem tones (the first tone "clears" out any echo cancellors on the circuit), but they got so many calls for bad faxes and modems not connecting that they ripped out most of the ADPCM equipment. Some of it is still there, but at more remote sites now. Mfgrs claim to have better detection (and also trying for MPEG compression) but some of the early adoptors (read: telcos) still have healing scars. rob.