Network Management System Dimensioning with Performance Data

Abstract

The goal of this work was to develop a new solution for dimensioning Nokia's network management system by utilizing performance data to achieve more accurate predictions for capacity usage. An algorithm for generating predictive models for different system resources and application performance attributes was implemented. Multiple linear regression with an exhaustive branch-and-bound search through the space of possible predictor variables was utilized to predict the usage of different resources. Solutions for the internal and external validation, outlier detection, and the visualization of the models were also specified. The algorithm was evaluated with a sample of data from real network management system environment and the results were promising, because fairly accurate predictions could be made. This solution is still in progress: The model maintenance and specification, as well as handling nonlinear relationships must be considered.