Very few measurements have been done on wide area ATM networks. In this paper, we first look at how a compliant traffic can be generated with standard ATM equipment. We show that in our environment, even with a 23.42 Mbps VP out of 34 Mbps links, it is necessary to use some spacing in the ATM adapters to generate a compliant traffic. Our measurements show that with a large window TCP is able to achieve a throughput of 19 Mbps when the ATM level traffic is compliant with the traffic contract. However, if the ATM level traffic is not entirely compliant the segment loss rate increases quickly, and the TCP throughput collapses. A 3% change in the cell rate of the ATM level traffic can cause the TCP throughput to drop from 19 Mbps down to 1 Mbps.

The Internet TCP/IP protocol suite is the first higher-level protocol stack to be used on ATM based networks. High-speed networks reinstate the end-system as the communication path bottleneck. We show how the host architecture and host network interface are crucial for memory-to-memory TCP throughput. In addition, configurable parameters like the TCP maximum window size and the user data size in the send and receive system calls influence the segment flow and throughput performance. We present measurements done on Sparc2 and Sparc10 based machines for both generations of ATM-cards from FORE Systems. The first generation cards are based on programmed I/O. The second generation cards on DMA. To explain the variations in the throughput characteristics, we put small optimized probes in the network driver to log the segment flow on the TCP connections.>