HIGH-SPEED INTERCONNECTION OF WORKSTATIONS - CONCEPTS, PROBLEMS AND EXPERIENCES

High speed networking technologies with transmission capacities in the range of hundreds of megabits per second (or even gigabits per second) together with very low error rates are difficult to operate due to a performance gap between the amount of raw data capacity and the end-to-end performance which remains available for applications. One of the reasons for that effect is the limited processing power of endstations. Although the performance of those stations is rapidly increasing there is a mismatch between end station capacities and network capacities. This is due to network I/O interfaces, system architectures, host interprocess communication and - last but not least - to inadequate functionality and implementation of transport protocols. In this contribution we describe different methods which may be applied in order to increase the end-to-end-performance of high speed networks. After illustrating some of the bottlenecks which may limit the performance we propose different approaches for high performance network interfaces. The results are illustrated by a comparison of implementations for TCP/IP as well as for the lightweight XTP protocol within a SUN OS kernel.