Delegates to a recent US conference might have worn black armbands to signify IP switching was killing off ATM, but according to at least one independent consultant, rumours of ATM's death have been greatly exaggerated.
The president of strategic planning consulting firm CIMI Corporation, Tom Nolle, speaking at an IBM-sponsored seminar for government IT users, scorned the claims of many router supporters and was almost evangelical about the importance of ATM.
After start-up company Ipsilon Networks unveiled a switch that delivers full 155Mbit/sec ATM throughput and which can route TCP/IP protocols at five times the speed of current devices earlier this year, delegates to one US conference wore black armbands to signal the death of ATM. Transparent rubbish, according to Nolle.
"The Ipsilon product that is going to "kill" ATM is called the ATM 1600. The ATM 1600 has no port options except ATM ports. Do you know what the ATM 1600 does when it recognises an IP flow it wants to switch? It sets up an ATM virtual circuit. And that's the truth," Nolle said.
"And whatever a vendor tells you an IP switching architecture is, there's only two things it can really be: thing number one is a way to apply virtual circuit technology to IP routing. Thing number two is something that doesn't work."
And building a network on a virtual circuit topology, Nolle says, is the best way to simplify the network. Over time, the use of virtual circuits and a static routing table can eliminate the need to use routers at all.
Nolle was equally scathing of router vendor claims that the solution to SNA/LAN integration was datalink switching. Acknowledging the sleazy reputation of some consultants, and promising faithfully that "everything I tell you here, everything, is the truth to the best of my ability to learn it", Nolle said he had watched "this architecture of datalink switching create the worst SNA network failure that I ever saw happen - 63 hospitals, one full day outage. VTAM would not come up, nothing ran for one day because of problems in this area."
Nolle continued: "The problem with data link switching is fundamental. Connectionless architectures can't control performance and response time. SNA networks were written for sub-second response time.
"There are many factors that make SNA applications time-critical. Router networks are not time-critical networks. I can make them time-critical by oversupplying with resources, but the fundamental operation of router networks is not to support time-critical applications. Most organisations who try to do this over time run into problems," Nolle said.
A fervent believer in the value of a carefully planned evolution to ATM as the ideal way to solve both global network problems and day-to- day tactical issues, Nolle said ATM offered the best investment protection around.
"You in the government sector know, better than many commercial organisations know, you do not have the latitude to go back and do things over all of the time. You really have to get it right, you have to plan a long-term technology investment trend line that's going to suit your application requirements for some reasonable period of time so that your investment is pro-tected and so the taxpayer doesn't turn you out at the next election."
A carefully planned evolution to ATM provided that capability, he said.
"ATM has a property that is really important: ATM is the only architecture that we have conceived as a technical community that allows us to create cost against Quality of Service on an application by application basis. And the most that you can ask your network to do for you is to be able to do that.
"That is what makes it a superior overall architecture. Whatever it is that we can reasonably expect networks to do for us, ATM can do. We can't propose a technology that has that same flexibility other than ATM."
Today's LANs with router-based internetworking, and consuming traditional leased-line or virtual circuit Internet connections, should evolve to ATM hub LANs combining legacy stations and ATM stations, and consuming ATM Internet services, Nolle says.
The main problem facing most SNA organisations today is the effect of unruly LAN growth and unplanned internetworking. This has created a separate network with its own traffic flows, own sources, own destination and own costs, meaning organisations must find a way to harmonise SNA and LANs. They must be able to work together in a single network infrastructure, economically and in the interests of economy. Phased implementation of ATM is the answer, he says.
Nolle added: "One of the important things about ATM is to understand that ATM to the desktop is important not necessarily because it has to touch every human in Australia, but because it has to touch the people who are most critical to the operation.
"Organisations that are strong candidates for ATM to the desktop are organisations who have a high unit value of labour per employee. In other words, you want to enhance the productivity of somebody who is expensive as a resource and also somebody who has a high level of computer literacy.
"Even if we were only going to penetrate a quarter of the desktops with ATM, if those are the 25 per cent of the people who make the company's decisions, plan the company's activities, that is enough ATM in itself to be important," Nolle said.
"I am not telling you to throw everything out just because 20 per cent or 50 per cent or 80 per cent of your users are going to use something like this. What I am saying is the benefit of ATM for the other users will be there and the benefit of supporting these elite users will be there."