Analysis and performance of priority and switching in local area networks
Amr Elsaadany · OhioLink ETD Center (Ohio Library and Information Network) · 1995
Due to the widespread use of distributed computing and the emergence of new demanding applications, local area networks (LANs) are experiencing higher traffic loads. These emerging applications require higher network transfer rates. Moreover they require timely delivery which put stringent requirements on the network delay. The design of high-performance LANs is a very important research topic. The use of LANs is continuing to grow as the use of distributed systems grows and as computers become more and more powerful and affordable. However, because LAN access is based on sharing the communication bandwidth among users, LANs have variable end-to-end delay. Moreover, most existing LANs have limited bandwidth and because of the overhead of LAN protocols, the throughput of the network is typically low, and the utilization of the physical medium is usually low. This dissertation contributes to the design of high-performance local area networks (LANs). We investigate how priority communication and packet switching can be used in the design of high-speed LANs. Local area networks were originally designed to handle data traffic which are bursty in nature and for which variable delay is acceptable. However, a large number of emerging applications require constant delay in addition to fast delivery. For example, multimedia video packets must be received at the user site with very small variation in delay such that disruption in the user's presentation is minimized. We make several contributions to LAN design in order to enable them to satisfy the needs of these emerging applications with various data types. First, we introduce priority-based communication schemes that provide low end-to-end delay for high priority traffic and compare the performance of these schemes. Second, we present hub-network configurations in which multiple distributed switches are fully-connected such that only one hop is needed to get traffic from one LAN segment to another. Hub-networks improve the performance as well as the reliability of the network. They reduce the overall end-to-end delay and eliminate the (traffic) bottleneck. Third, we study various switch architectures and compare their performance. We show that parallel switches work very well under high traffic loads and at light-to-medium load, they give close to optimal performance. Parallel switches provide much better overall performance than traditional switches. Finally, we study buffering within switches and their effect on the performance of the network. In particular, we evaluate the effect of buffer size on the packet loss rate and the percentage of delayed packets.