Internet traffic volumes characterization and forecasting
MetadataShow full item record
Internet usage increases every year and the need to estimate the growth of the generated traffic has become a major topic. Forecasting actual figures in advance is essential for bandwidth allocation, networking design and investment planning. In this thesis novel mathematical equations are presented to model and to predict long-term Internet traffic in terms of total aggregating volume, globally and more locally. Historical traffic data from consecutive years have revealed hidden numerical patterns as the values progress year over year and this trend can be well represented with appropriate mathematical relations. The proposed formulae have excellent fitting properties over long-history measurements and can indicate forthcoming traffic for the next years with an exceptionally low prediction error. In cases where pending traffic data have already become available, the suggested equations provide more successful results than the respective projections that come from worldwide leading research. The studies also imply that future traffic strongly depends on the past activity and on the growth of Internet users, provided that a big and representative sample of pertinent data exists from large geographical areas. To the best of my knowledge this work is the first to introduce effective prediction methods that exclusively rely on the static attributes and the progression properties of historical values.