A scalable and efficient methodology for flow monitoring in the Internet

The large number of small flows bringing very few traffic challenges the scalability of per flow monitoring in the Internet, especially in high-speed links. For several measurement purposes a way to deal with it is to give priority to the monitoring of large “elephant” flows and neglect small ones. We present here a novel methodology (GROWTW RATE) that, through a metric, decides whether to keep the current pool of monitored flows or to admit a new one removing the one currently having the lowest metric value. All the parameters needed to compute the metric are automatically estimated and kept up-to-date while the system is running. We compare our approach with other possible ones (including one already proposed in the literature) and show that it achieves higher efficiency. We conclude the paper with a scalability analysis that confirms the implementability of the approach in a real system.