| 摘要 |
<p>This invention is concerned with packet switching techniques, protocols, networks, and equipment for use on a local area or wide area scale. Packets which include a fixed-length data field and memory offset field are employed so that packeting, block data transfer to contiguous memory and depacketing can be effected at the destination by hardware-mediated processes with minimal processor involvement. Multiple sessions can be readily handled by assigning addresses to multiple memory buffers at the destination. Thus, each packet carrying computer data provides three levels of addressing for its data segment corresponding to the destination node or device, the memory buffer for the data block being transfered and the memory location within that buffer into which the segment of data carried by the packet is to be placed. The equipment includes loop-based, collision-free packet switches in which packets are modified and re-circulated when a destination is busy. This allows a high-speed loop to transfer packets to and from devices running at various (lower) speeds. Loop latency is minimized and data transfer rates maximized by the use of fully-buffering parallel forms of the switch. A method for handling broadcasts is also disclosed, together with a technique for preempting low priority packets and exploiting short-cuts on a loop. These improvements allow application of the technique to a high-speed voice/data packet-switching in large wide area networks as well as in small local area networks. By using the methods and equipment disclosed, an extensive, powerful and extremely fast integrated communications subnet can be largely built in hardware with little need for processor intervention.</p> |
