In times of Internet access being a popular consumer application even for "normal" residential users, some telephone exchanges are congested by customers using modem or ISDN dial-up connections to their Internet Service Providers (ISPs). In order to estimate the number of additional lines and switching capacity required in an exchange or a trunk group, Internet access traffic must be characterized in terms of holding time and call interarrival time distributions. In this paper, we analyse log files tracing the usage of the central ISDN access line pool at University of Stuttgart for a period of six months. Mathematical distributions are fitted to the measured data and the fit quality is evaluated with respect to the blocking probability caused by the synthetic traffic in a multiple server loss system. We show how the synthetic traffic model scales with the number of subscribers and how the model could be applied to compute economy of scale results for Internet access trunks or access servers. Keywords: Internet traffic modelling, ISDN access, trace evaluation

Abstract-In this work the authors show how the behaviour of Web users strongly affects the dynamics of TCP connections in Internet. Analysing actual and systematically generated HTTP traces, it is proved that the time between the download of two pages is critical to determine the re -utilisation of TCP connections. On the other hand, the study also shows that the utilisation of 1.1 version of the HTTP standard does not significantly affect the traffic generated by HTTP 1.0. In this sense, the heavy-tailed nature of the size of HTTP connections can be considered as an invariant property.

This paper presents a comprehensive characterization of the DAS-2 1 workloads using twelve-month scientific traces. Metrics that we characterize include system utilization, job arrival rate and interarrival time, job size (degree of parallelism), job run time, memory usage, and job queue wait time. Differences with previous reported workloads are recognized and statistical distributions are fitted for generating synthetic workloads with the same characteristics. This study provides a realistic basis for experiments in resource management and evaluations of different scheduling strategies in a multi-cluster environment. 1