Parameter inference of queueing models for IT systems using end-to-end measurements

Abstract

An optimization-based inference technique to tackle the parameter inference of queueing network models for Information Technology (IT) system was discussed using end-to-end measurements. The method was shown to be superior to other available methods for obtaining parameter values from multiple, multi-dimensional experiments. The method was found to have the ability to self-adjust, with minimal cost, to significantly noisy data, outlying experiments, by monitoring the residual error after each level of the nesting procedure. The method was found to be robust to both noise and insufficient data.