Those steeped in ATM history may remember the 1996 conferences when the young, snotty upstarts from Ipsilon Networks Inc. hauled out a huge stack of books labeled "ATM standards" and compared them with a slim notebook labeled "IP Standards." Then they threw out T-shirts to the befuddled crowds emblazoned with the words "IP: Necessary and Sufficient."

Ipsilon touted a way of switching identified Internet Protocol flows to make IP traffic suited to the traffic bins created for ATM. Yakov Rekhter and others at Cisco devised tag switching, an improvement on Ipsilon's approach. That method later became the basis for multiprotocol label switching. This led to the slow death of Ipsilon, a company remembered only by the IP cognoscenti.

IP proponents never expected carriers would insist, conversely, on keeping a time-division multiplexed (TDM) architecture like Sonet far past its point of maximum utility. At the same time, they demanded an IP service that followed strict rules of jitter and latency. The worlds of Ethernet, IP and MPLS had to collectively take on and defeat the vast ATM and TDM infrastructure globally and were scarcely up to the task.

To protect TDM's flanks, the high priests of Sonet developed link capacity adjustment scheme, virtual concatenation and Generic Framing Procedure, all the better to handle Ethernet traffic. MPLS received the Martini and Kompella extensions for taking on a host of new voice and virtual private network duties. What was once a slim notebook for IPv4 to IPv6 transition became a fat stack of specs for handling complex circuitlike services using Sonet, Ethernet, IP and MPLS.

When the packetheads killed off ATM, they should have recognized something: Getting a packet to behave like a circuit is tough and complex, whether it's a 53-byte cell or a variable-sized Ethernet frame. Users want circuitlike service. Complexity comes with the territory.

Loring Wirbel is Communications editorial director for EE Times and its network publications.