Many people seem to consider me an ATM opponent, but I consider myself an ATM realist. Wearing whichever hat you place on me, I'd like to engage in a bit of an ATM (Asynchronous Transfer Mode) post-mortem.
Once upon a time, the future was to be all ATM, all the time. For years I saw projections of $X billion worth of ATM sales N years in the future. X and N varied depending on the consulting firm that was putting out the projection, but the values of X and N seemed to stay consistent year after year from any individual consulting firm. In absolute terms, ATM has grown into a reasonable-sized business, but it remains a very small part of the overall network market.
It is a bit of a cheap shot to talk about the fate of desktop ATM, so I'll let dead dogs lie and move on to two places where ATM was expected to take over the market: backbone and access networks.
With all the speed they can muster, telephone companies have been rolling out ATM-based services at a glacial pace. In many parts of the US, you can get ATM virtual circuit-based connections. Some offerings even have distance-insensitive pricing, making them very attractive. Many companies are signing up for these services, but this will be a short-lived phenomenon.
I expect that IP services, including IP-based virtual private networks, will take over this market. In addition, I don't see much future for ATM in supporting IP backbone services. ATM is good at splitting up links into smaller pipes, but this does not seem to be all that useful when ISPs are growing to the point where they need full OC-48 and OC-192 links between point-of-presence locations.
I expect that Multiprotocol Label Switching will be used instead of ATM where links still need to be subdivided. That leaves access networks. I think ATM features are a good match for the requirements of access networks. Links to individual locations tend to be small, and if there is a desire to multiplex multiple services over those links, ATM seems to be a better technology than IP. For example, I expect to see a lot of ATM-supported integrated access devices being deployed as part of telco-based digital subscriber line services. This is partly because much of the telco world hungers for any way to maintain the circuit-based philosophy of the telco networks. ATM will not give the telco world much satisfaction in this case, however, because the ATM virtual circuit will only span the access link and not be end to end.
Proponents of ATM irrationally assumed that a single technology would meet all network requirements, and a lot of money was lost in that assumption. I wonder if some IP proponents are guilty of the same hubris.