Internet telephony revisited

A few days ago, David Weinberger announced a letter to the FCC's Michael Powell, signed by a flock of Net thinkers, entitled "Fail Fast." This document, along with the links that surround it, has got me thinking about how the prospects for computer telephony have changed since I wrote about the subject in a 1996 1994 story.

The subject was on my mind anyway; I have a telephony story coming up in InfoWorld. As part of my research, I solicited and am now using the Vonage IP phone. It's an eye-opener, for sure. A little Cisco box (the ATA 165) jacks into your DHCP-enabled DSL router, you press a button, and away you go. Works right through my fully-locked-down NAT. Most folks I talk to notice the difference -- especially if I clog up my 256-up/256-down circuit with a DSL speed test while I'm talking -- but it's darned impressive. The bottom-line savings on long-distance charges are of course the reason why most folks might be willing to ditch their POTS line for this solution. But I'm equally intrigued by the upside: applications that intercept and work with SIP signalling and digitized voice, integrate phone presence with IM, forward call metadata to Web services, and all the other CTI (computer-telephone integration) fantasies that may finally start to come true.

Central to the "Fail Fast" manifesto is David Reed's famous end-to-end essay. I just reread it in light of a recent InfoWorld piece on reliable messaging, and I see that my article raised but did not resolve a contradiction. If you use message-oriented-middleware to carry SOAP traffic, is the reliability thus gained in the "network" or in the "application"? The same question will apply to technologies deployed in service of higher-quality VoIP.

I think that Alden Hart, whom I quote in my story, points in the right direction. Yes, the network should be dumb and the endpoints smart. But that doesn't mean every application has to reinvent the wheel. His notion of a "smart connector" -- for example, an active intermediary that performs queue control -- is a useful and important construct. As David Reed says in his paper, "Using the end-to-end argument sometimes requires subtlety of analyis." We'll need a lot of that kind of subtlety to work through the convergence of the IP and PSTN networks, as that long-envisoned day draws nearer.

Former URL: