The ATM Warren was a demonstration that ATM networks could be built as cheaply as Ethernet networks. The essence of the Warren was to use only third-party set up of calls between simple devices and switches, while maintaining fully-standard ATM networking operation at all interfaces.
The Warren was an ATM sub-network composed of an arbitrary mesh of very simple ATM switches and end stations and which is connected to the outside ATM network at one or more points. The novel aspect was that switches and end stations within the Warren were implementable entirely in hardware - i.e. they required no microprocessors for signalling and management. All of the signalling software for all the devices within the Warren were provided by one or more proxy servers. The proxy servers are software entities situated on one or more general-purpose computers connected to the external ATM network outside the Warren.
A number of prototype devices were built under the sponsorship of Virata. These were intended for the home. The home needs a low-cost network that is multi-media capable. Both software and hardware devices were created.
One of the first devices built was the 6 port Warren switch. This provided six ports of ATM-25. It was designed by Daniel Gordon. Like all Warren devices, it did not contain any microprocessors, but instead was controlled by a master controller situated anywhere on the network. The Warren protocol allowed devices to keep in touch with their controller, and this was implemented in hardware at the device end.
Two versions of the switch were designed. Although similar externally, the second version had two levels of priority for FIFO queuing and a 32 kbyte buffer store for bursty traffic. SPEC.
The very first devices buit were these microphones. They took power over the ATM-25 RJ45 connection and generated a continuous stream of cells.
The physical design, when cased, was meant to resemble a hand-held mic, like an SM58. SPEC.
Warren CD Player
The Warren CD players were created by placing an ATM-25 line card in the back of a commercial CD player.
The services available over the network were sourcing music and control of the device using an ATM encoding of the infra-red remote control. Also, the infra-red received at the device's local receiver could be decoupled from local control and relayed over the network.
Warren Telephone: 1997-98
This was probably the first telephone handset/instrument to directly connect to a packet-switched data network. It was a direct decendent of the Ring Phones connected to the original Cambridge Ring as part of the ISLAND project and a similar project at Xerox PARC. The Ring Phones used a 19 inch rack, whereas the Warren phone, like the later ORL Broadband Phone and CISCO 7941 IP phone, connected to the network using a packet-switched data link. Between the Ring Phone and the Warren Phone there was a whole generation of N-ISDN phones that, while digital, still used circuit switching.
The Warren telephones were created by replacing the PCB inside a standard telephone set with a new digital card. The card supports ATM-25 on unshielded twisted-pair (UTP) or on plastic optical fiber (POF).
The telephones provided all the usual functions of a telephone handset. There was no state or microprocessor in the phone. Every operation was relayed directly over the network interface. Each button push and release caused an ATM cell to be sent. Similarly, off hook status, ringing and tone generation were marked and enabled by a continuous stream of cells, instead of having specific cells to mark the start and stop these states. Such an approach avoided any state needing to be stored inside the phone.
The competition to the Warren was Switched Ethernet. Ethernet could not support QoS but did not suffer from `badput' during congestion where the remains of a datagram were shipped over the network, only to fail in reassembly. Also, the complexity of ATM protocols defined for telecoms use meant that full-featured switches, like the Virata 12+12 shown below, were more complex than Ethernet alternatives for the workgroup. The Warren was primarily a demonstration that this did not have to be the case: the software complexity could be selectively deployed at any point, providing a seamless, interoperable spectrum between none and the full CCITT stacks.
These workgroup ATM-25 switches contained their own microprocessor and a good deal more silicon. The Warren switch pictured above is resting on the Virata 12x12 switch. The Virata switch could control a number of Warren switches. Indeed, many thousands of Warren devices could be interconnected and run off one controller if desired.
In the end, the real competition was switched Ethernet. Today we are learning how to achieve QoS over Ethernet and in the Internet in general. ATM's reduced complexity owing to using fixed-length cells was a chimera that soon became irrelevant with advances in VLSI, and its provision for in-band QoS support mechanisms, such as priority, based on connection-oriented virtual-circuits, is now less favoured than on-the-fly fairness algorithms, implemented in routers that identify the flows for themselves.