Why Did Broadband-ISDN Really Die?


There are positive reasons for the smooth uptake of IP, such as the easy availability of the TCP/IP stack as compared with competing proprietary data protocols, the relative simplicity of the basic Internet architecture, and the prior existence of enterprise multiprotocol routers that could be used directly as Internet routers.
Perhaps more important, though, were the problems with ATM. In the 1980s when ATM was being designed, the dominant usage mode was seen to be the multimedia successor to the phone call—human beings making videophone calls. As we saw above, interactive multimedia is the most challenging application for packet networks, requiring a complex infrastructure of signaling, terminal capability negotiation, and QoS-aware media transport. It was not until the mid-nineties that large ATM switches capable of supporting the required signaling and media adaptation came to market—too expensive and too late.

Even worse, the presumed videophone usage model for B-ISDN was highly connection-oriented, assuming relatively long holding times per call. So ATM was designed as a connection-oriented protocol with substantial call set-up and tear-down signaling required for every call to reserve resources and establish QoS across the network. This required per-call state to be held in each of the transit network switches. For comparison, millions of concurrent calls (sessions or flows) transit a modern Internet core router and that router knows nothing whatsoever about them.

It turned out that critical enabling technologies for the Internet, such as DNS, require brief, time-critical interactions for which a connection-oriented protocol is inappropriate. Even for connection-oriented applications such as file transfer, which use TCP to manage the connection, connection state is held only in the end systems, not in the network routers, which operate in a connectionless fashion. This has allowed the Internet to scale.

So in summary, ATM had too narrow a model for how end-systems would network, and backed the wrong connection-oriented solution that couldn’t scale. Because ATM was designed against a very sophisticated set of anticipated, predicted requirements, it was very complex, which led to equipment delays, expense, and difficulty in getting it to work. The world moved in a direction not anticipated by the framers of B-ISDN and it was stranded, and then discarded.

No comments:

More?