The VuNet is a gigabit-per-second desk/local-area ATM network which interconnects general-purpose workstations, network-based multimedia devices and bridges to other networks.
During the course of an application, media streams are exchanged between the workstations, and devices in a seamless manner. This architecture has several advantages over traditional workstation-centric systems including the ability to share multimedia devices and to reduce the burden of multimedia tasks on the workstation.
We believe the next step in the evolution of multimedia systems involves allowing multimedia information to reach the application. This means channeling the multimedia information to the workstation processor rather than bypassing it with specialized hardware. This will allow a greater variety of multimedia applications that go beyond teleconferencing, allowing the computer extraction of information from the workplace.
The VuNet is based on a modified ATM standard, which uses a 5 byte header and a 48 byte payload. The philosophy motivating the design of the VuNet hardware has been simplicity. Our goal is to build network hardware that is easy to design and test from off-the-shelf VLSI components. We believe that a local environment such as the ViewStation does not need sophisticated network functions. It can be effectively served by a switch fabric having a limited number of access ports and an internal speed that is greater than that of the clients.
The VuNet interface between client and switch is designed to be simple, allowing clients to be easily built. This follows the sofware intensive philosophy, which emphasizes a simple, flexible hardware substrate, and pushes complex functionality into workstation-based software.
Within a desk-area context, an office is equipped with a network switch which interconnects workstations and multimedia devices. Offices are then networked by connecting individual switches with links.
The architecture of the VuNet switch consists of bidirectional ports with first-in first-out (FIFO) buffers which feed a crossbar matrix. Clients connect to the switch through the ports, through which they control the reading, writing and routing of ATM cells. The switch is passive in that it simply executes the commands of the clients.
The current version of the switch has four ports and has been operating reliably at a speed of 700 Mb/s. In addition, the switch has tested reliably at an internal data rate of 1.5 Gb/s.
An important aspect of the switch is that the ports are easy to interface to in many respects. Since the timing of the client side of the port is decoupled from the internal timing of the switch through the FIFOs, there is no need to synchronize operation with the internal switch clock as in traditional designs. Furthermore, both the transmit and receive blocks can be operated concurrently. The FIFOs also serve to buffer cell bursts. Finally, the data bus width to the port is selectable, either 32 or 64 bits, allowing a simple mapping to workstation and processor buses. All these factors make it easy to design devices which connect directly to the switch.
In order to interconnect the switches, high speed optical links are designed. In keeping with the software-oriented connection setup, the links include in-band remapping of the VCIs.
Connection management in the VuNet is done by special cells on a particular ATM VCI. Cells received by the link on these special VCIs (Control Cells) can cause the link to alter its header lookup table, which contains information on mapping VCIs to VCIs and Ports. They can also cause the link to emit a cell containing table entries, so link tables can be read back. Thus cells in the VuNet are routed hop-by-hop as they pass through the links.
Each workstation in the VuNet is responsible for opening, maintaining, and closing its own connections. This can be done in a ``wormhole'' fashion by way of ATM Control Cells embedded in the data stream. The allocation algorithms in the connection daemon prevent nodes from stealing other nodes' connections. Processes are also run in the background which verify link tables and refresh connections if necessary, such as in the case where a switch has gone down.
The Vidboard is a video capture and processing board which connects directly to the VuNet. It is capable of generating video streams having a wide variety of characteristics relating to the presentation of video (e.g. picture size, color space, frame rate) and transport of video across a network (e.g. protocols for traffic and transport). Devices within the system use a set of ATM protocols for requesting video from the Vidboard.
The VuNet interface (known as the VudBoard) is a simple DMA interface, allowing programmable length DMA bursts both to and from main memory into the board's FIFOs, which can hold one cell in each direction. The bus from the interface to the network is shared by the receive and transmit circuitry, so arbitration is done with priority given to incoming cells.
The interface is capable of bursting at the full rate of the Turbochannel, which is nominally 800 Mbs. However, due to the small size of the DMA transactions and the DMA latency, the maximum achievable inbound DMA rate is 400 Mbs. This is comparable to the Bellcore Osiris interface, which can achieve a maximum input throughput of 463 Mbs.
Since the ViewStation project serves as an end-user LAN for the Aurora wide-area gigabit network project, a VuNet-Aurora (SONET) interface has also been fabricated.
The Aurora-VuNet link (AVlink) allows the VuNet to interoperate with the Aurora wide area facilities. It converts cells between the modified ATM format used in the VuNet into standard ATM cells used in Aurora. This includes generating the Header Error Check (HEC) and modifying the length of the header. The AVlink also bridges the physical layers of the two networks.
The AVlink demonstrates that one of the advantages of using ATM in both the local and wide-area networks is the simplicity of interconnecting them. The AVlink only needs to make minor cell changes at the ATM and physical layers, which can be done as the cells pass from VuNet to Aurora.
The design of the AVlink allows the LAN/WAN boundary to appear seamless to clients on either side of the interface.
The Vid40 System is a multiprocessor system for video processing that connects directly to the VuNet. The system is based on the C40 DSP from Texas Instruments which has special communication ports for building multiprocessor architectures. The system is composed of individual boards which are connected together through the C40 communication ports.
The purpose of the Vid40 System is the real-time execution of computation-intensive video processing operations. Examples of such operations are filtering and compression. We intend to draw upon established practices in the fields of parallel processing and image processing for specific implementation choices, such as multiprocessor topology and compression algorithms.
There are currently 4 deployed VuNet nodes in the Telemedia, Networks, and Systems group, and one VuNet node in New Jersey. The two networks are connected over the Aurora wide-area facilities using the Aurora-VuNet link. The students have collectively designed and built the hardware and are actively extending the system's scope.