Modelling Throughput Prediction Errors as Gaussian Random Walks

Abstract

One of the most critical aspects of anticipatory networking is assuming that future system conditions can be estimated. In this paper we address how accurate the current state of the art predictors are in providing a forecast of short term throughput. We propose a simple model for the short term prediction error based on Gaussian Random Walks that allows for mathematical analysis of the impact of imperfect future knowledge on network optimization.