The effect of bursty traffic on the performance of local area networks

LAN performance under typical network configurations and realistic workloads indicative of today´s computing environments is examined. Hub topologies are common configurations for Ethernet and token ring networks due to their flexibility and their support for network management and diagnostics. Moreover, traffic generated within these networks tends to be much burstier than represented by typical load models This increase in traffic burstiness is due to the widespread use of distributed applications (e.g., distributed file systems and distributed databases) and high-speed computers capable of transmitting large amounts of data in a very short period of time. The Markov modulated Poisson process, (MMPP) is used to model the burstiness of network loads. The effect of bursty traffic on the application-to-application performance is studied, and the effect of mixing different types of traffic on the same network is shown