A Trick Question

By: Scott Bradner

The Vegas Networld + Interop is almost here and I can predict at least one thing with full assurance: the glitz and occasionally tawdry glamour of Las Vegas will again pale next to the hype over ATM . Booth after booth will be festooned with banners and eager salesdroids trying to convince the attendee that ATM is the future and that the future is now.

ATM is still all the rage. Anything with ATM in the title draws a big audience. (It was suggested that I rename my tutorial to include the character string ATM because, it was reasoned, it would double the attendance. It would be legit since I do talk about ATM but the title is too long already. I'll just hope for the best and consider a name change next time if the ATM hype level continues to be self-sustaining.)

It seems to me that an awful lot of the ATM promise depends on ATM soon becoming ubiquitous as THE networking technology. Not just the data networking technology but also the voice and video networking technology.

ATM is touted as being the only way to provide for many of the high demand applications of the future including desktop video conferencing, multi-media, multi-player interactive games, video on demand and single instrument where your PC is also your phone and TV. At the same time, ATM is also touted as a way to provide for considerable cost savings by being able to combine the enterprise telephone and data network, even if you are old fashioned enough to want to break things out on the desktop to separate phones and computers. Mental pictures are painted of multi-media applications soon requiring many tens of megabits of information per second to each of the corporate desktop machines.

Whether these pictures represent a dream or a nightmare is up to the individual but in order to realize them it would seem that the ATM network would have to extend from to desktop throughout the enterprise and have connections to the ATM networks of collaborating enterprises. Well, at least according to some of the ATM proponents, there are others that maintain that one can support many of the same applications over switched Ethernet, but I guess those people are not on the right team.

I will say that it is a bit hard to see the millions of 10BaseT connections that are being made each year switched overnight, or even overyear, to ATM just after the investment has been made but I guess I'm not quite on the team either.

Part of my problem here revolves around the effect that this assumption of ubiquity has on the designers of future applications. If they assume that 'under all is ATM' then they also make assumptions about the availability of specific resource reservation techniques. They may use an API that knows that ATM is down there and thus closely mirrors ATM's abilities and limitations (I assume there are limitations though ATM proponents speak of any; it is nice to be able to assume perfection in some aspect of life.)

What happens if someone wants to run these applications over some part of the existing multi-billion dollar network infrastructure? Will they work?

One of the long term strengths of TCP/IP is that it makes no assumptions about what it is running over. Everything from barbed wire to satellite links, from phone wires to fiber optic cable are just wires to IP. (As ATM can also be.) When using TCP/IP an application running on a PC attached to a token ring in Tokyo can interact with a supercomputer on a Hyperchannel in Pittsburgh. It would not be a step up if future applications did not support this flexibility.

So I'll leave you with a trick question: Is ATM the last networking technology?

Disclaimer: Harvard has a Folklore and Mythology department. Some of what I've heard about ATM might fit there, but the above are my own worries and opinions.