Burstiness predictions based on rough network traffic measurements

To dimension network links, such that they will not become QoS bottlenecks, the peak rate on these links should be known. To measure these peaks on sufficiently small time scales, special measurement tools are needed. Such tools can be quite expensive and complex. Therefore network operators often rely on more cheap, standard tools, like MRTG, which were designed to measure average traffic rates (m) on time scales such as 5 minutes. For estimating the peak traffic rate (p), operators often use simple rules, such as p = α · m. In this paper we describe measurements that we have performed to investigate how well this rule describes the relation between peak and average traffic rate. In addition, we propose some more advanced rules, and compare these to the simple rule mentioned above. The analyses of our measurements, which have been performed on different kinds of networks, show that our advanced rules more adequately describe the relation between peak and average traffic rate.