Autonomic Provisioning in Non-Linear Network Traffic

The real data of the network traffic were analyzed to find out characteristic parameters for autonomic provisioning. An observed strong non-linearity (bursts) leads to heteroskedastic (time dependent conditional variance) model applications. The non-linear time series model and statistical method were applied. It was found that upper limit of burst variations could be quantitatively estimated with the required confidence level. This allows for autonomic provisioning and resource allocation in high speed networks. Depending on the studied case the results may be also interest in stochastic control (leading to control the system in uncertain environment), filtering problems (leading to filter out unwanted effects) or optimal stopping problems (leading to start an action in an optimal time).