Compression of data traffic in packet-based LANs

Abstract only given; substantially as follows. Data compression is very commonly used to improve the performance of telecommunications systems. In local-area networks compression technology can help reduce transmission bottlenecks if the network transmits data slower than the computers can generate it. Network designers need to have accurate models of the network traffic in order to plan network capacity. When designing networks, one must take into account not only the amount of traffic, but also the nature of the traffic. The authors are particularly concerned with communication systems that are packet-based, and with how compression changes the statistical properties of packet sizes. They also discuss how adaptive compression can be used with connectionless protocols, which pose serious synchronization difficulties. They have collected large quantities of data from a live network, and have simulated the effect of compressing the data using several different techniques. Relevant statistics of the simulated system have been calculated, allowing to characterize data compression as a stochastic transformation of teletraffic. Compression can improve throughput in packet-based networks by decreasing the size and number of packets. When applied to individual packets, not only is the mean packet size reduced, but the variance also decreases. Compression has the potential to give significant performance improvements.