We study the impact of the selection of flow control functions in high-speed transport protocols on the performance for both nonbursty client-server traffic and bulk-data communication such as file transfer in a local high-speed environment. Therefore, we consider a workstation cluster based on a FDDI network. Although there are some differences in the behaviour all flow control and retransmission schemes studied (Go back N, Selective Repeat, Selective Retransmission) are able to provide the same performance in the local environment. Here an additionally used rate control scheme is able to reduce a high buffer utilization of a receiving station for a bulk-data transfer without any impact on the transfer delay and throughput. In the conclusion we discuss both functionality and implementational aspects of transport protocols.>

We study the throughput characteristics of a FDDI workstation cluster installed at the University of Paderborn. We compare theoretical results to the performance really available to the user at the service access point to TCP and UDP. Furthermore, we compare the performance for two difficult operating systems: we consider the SUN-OS based on BSD and the new Solaris 2.3 developed for multiprocessor environments. The measurement results showed a maximum throughput of less than 60 Mbps which is still considerably less than the bandwidth promised by FDDI vendors. Moreover, it becomes clear that high throughput absolutely requires the transmission of large frames. We also briefly address the current discussion on higher speed LANs within IEEE 802.>

Contention control strategies for high speed backbone networks are studied. Packet loss is a serious problem in a backbone scenario due to the speed mismatch of the different networks. The authors' new mechanism tries to reduce the packet loss ratio for fiber distributed data interface (FDDI) operating as a backbone. When one buffer is overloaded and packets would be lost, the new schemes try to store those packets at another station. The proposed mechanism provides significant reduction in packet loss and prevents congestion in other parts of an interconnected network.

The authors study packet loss in high speed networks interconnecting conventional local area networks (LANs). They consider two different techniques to control contention at the station where packets leave the high speed network and must wait for service by the LAN. One technique is based on a backoff strategy where packet transmission to a temporary overloaded station is interrupted for a randomly chosen time. The other technique studied uses rate-based flow control. A discussion of results for both fiber distributed data interface (FDDI) and distributed queue dual bus (DQDB) shows protocol-specific peculiarities requiring different solutions to different problems. Packet loss characteristics obtained for FDDI and DQDB are very close to each other.>

Recently, a discussion came up dealing with the frame copied indicator in the frame status sequence of FDDI. This indicator shows whether or not the destination has copied the frame and this way allows the sender to react in a proper way. However, in interconnected networks the destination node may represent a large number of destinations with different addresses making it difficult to check within a very short time whether a certain packet is to be copied and the indicator is to be set. The paper discusses the effect of using or not using the frame copied indicator. Additionally, it proposes a class of new congestion control strategies for FDDI.>