Router power reduction by active performance control realized with support vector machines

The machine-learning-based dynamic performance control of routers is proposed to reduce router power consumption. In order to achieve fast adaptability to catch the changing traffic characteristics, we introduce two sequential classification measures; normalization with performance thresholds, and periodic staggered use of learning machine sets in combination with Support Vector Machine (SVM). Numerical experiments using several real Internet traffic data sets elucidate that the router power consumption reduction reaches 50-65%.