Traffic-Predicting A Routing Algorithm Using Time Series Models

A routing algorithm is proposed that analyzes network traffic conditions using time series prediction models and determines the best-effort routing path. To predict network traffic, time series models are developed under the stationary assumption, which is evaluated using the Auto Correlation Function (ACF) and Partial Auto Correlation Function (PACF). Traffic congestion is assumed when the predicted result is larger than the permitted bandwidth. Although the proposed routing algorithm requires additional processing time to predict the number of packets, the packet transmission time is reduced by 5~10% and the amount of packet loss is also reduced by about 3% in comparison to the OSPF routing algorithm. With the proposed routing algorithm, the predicted network traffic allows the routing path to be modified to avoid traffic congestion. Consequently, the traffic predicting and load balancing by modifying the paths avoids path congestion and increases the network performance.