Search for a command to run...
•• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • Flow labelled IP over ATM was both an architecture and a technology to solve a number of problems that emerged from the train-crash that was the collision between Broadband-ISDN and the Internet. It was not merely a hack to work around the mismatch in semantics between a virtual circuit switched underlay and a stateless datagram routed overlay. The intention was to take advantage of facilities offered by the underlay technology such as flow state maintenance and flowisolation and protection in switches, to offer QoS at various levels of granularity, whether for unicast or multicast. The paper was one of the seminal influences for what is now the burgeoning MPLS industry, although in the architecture described we see something simpler, more elegant, and, perhaps, more aligned to the traditional Internet design philosphy [Clark88], than to the telecommunications mindset in the hardcore Broadband Integrated Services Digital Networks (B-ISDN) Asynchronous Transfer Mode (ATM) proponents1. Of course there has been a long and valiant history of tweaking IP (and other datagram protocols) to run over virtual circuit (VC) networks, dating back to the need to run IP over X.25 in the 1970s-1980s. Typically, there was no advantage taken of the underlying circuit technology (indeed there was a disadvantage in having to use it), and so the main techniques developed involved heuristics for triggering circuit setup and tear-down from a shim module that sit between IP and the VC layer and monitored IP level activity. Alongside, if you were lucky, there might be a management knob to allow you to adjust the laxity or aggressiveness of the shim, depending on potential cost/performance tradeoff decisions: X.25, and later IP over dial-up typically incurred monetary costs by time and/or volume. B-ISDN (and, guilty by association, ATM) was originally intended as an overarching system for networking, replacing IP. It was too complex and too late to survive in that role, although the underlying cell-switching technology is still a vital component in today’s; network landscape (many xDSL lines use ATM as a multiplexing layer to allow low latency voice circuits to run without over-provisioning, alongside IP and video traffic - at lower speeds, such as the typical xDSL deployment of a handful of megabits a second, this is fairly crucial). However, in the panoply of B-ISDN there was a whole armoury of signalling protocols (UNI, NNI etc). The authors have strong systems credentials as well as networking, and this shows in their interpretation of the problem, and the architecting of solutions including the details. They rightly ignored the complex smorgasbord of B-ISDN and seized on what ATM cell switching could offer2. Take-home messages from this paper are: • You can separate routing from forwarding. If forwarding can all be done flexibly enough in hardware but cheaply and faster, then make sense of it. At the time, ATM switches did this nicely.