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.

A three-layer model of a computer network is used to introduce the concept of interface and of end-to-end protocols. Using a simple interface protocol as example, finite state automaton and Petri nets are introduced. The idea of an interface machine is rejected and the problems related to the transmission medium are approached. End-to end protocols request a global model which includes two local models and a transmission medium model. However, the local model has to be generalized. This eventually leads to the association of the state of the automaton with a context. Petri nets have also to be expanded. Models presented here have been successfully used for modeling transport protocols.

Basic node-to-node protocol of a packet-switched network based on the multiplexing of the link intoNlogical channels is first described. The node structure is then described based on a set of queues, an input routine, a main-task routine, and an output routine. Three protocols which are based on different implementations of the transmission of the acknowledgment (ACK) block are then introduced. The first one is the 1972 version of the Advanced Research Projects Agency Network (ARPANET), the second places the ACK block in regular packets, and the third uses only control packets. The simulation program developed to compare these protocols is then described. It studies the behavior of a link between two nodes, which consists of a full-duplex line and of two sets of queues. The rest of the packet-switched network is simulated by a packet generator which feeds the two sides of the link. First, comparison of protocols is done using symmetrical and asymmetrical loading conditions. In this last case, interesting results are obtained in relation to momentary saturation of theNlogical channels. The influence of the limitation of buffers allocated to an output line is also discussed. Sensitivity of protocols to reverse traffic characteristics is then studied and, based on the results, choice of a protocol is discussed. Based on the dynamic behavior of the queue, some questions are raised in connection with adaptive-routing algorithms.