A STUDY ON WEEKLY JPIX TRAFFIC VOLUME PREDICTION CONSIDERING DIFFERENCES IN DAILY TRAFFIC CHARACTERISTICS

Internet has been established as one of infrastructures indispensable for global social and economic activities. According to the investigation of Ministry of Internal Affairs and Communications, penetration rate of Internet is 84.9% per one household at the year-end 2014 in Japan Estimated Internet traffic volume reaches at 2.89 Tbps (May 2014) which showing 10.7% increase in a year. The internet traffic prediction is getting more and more important for appropriate Internet traffic management. This paper proposes a method for weekly internet traffic on JPIX (JaPan Internet eXchange), one of the largest Internet backbone in Japan. Singular value decomposition is applied to the daily JPIX traffic data updating every 5 minutes. The data were taken from graphical data presented at JPIX’s web page. Two orthogonal vectors are found to be useful for efficient traffic prediction. The prediction is made based on monthly trend of weight vectors. The trend has been examined in detail and the best prediction scheme has been proposed. Three classes, i.e. weekday, weekend and holiday season showed characteristic patterns useful for prediction. Adaptive weight update algorithm has been also proposed to improve the short term prediction accuracy. The method will be useful for network routing management in daily bases