Performance analysis of traffic classification in an OpenFlow switch

Abstract

Traffic classification in an OpenFlow switch will play an important role in large-scale networks along with the advent of IoT. Previous study has evaluated its performance based on actual experiments, however obtained results are not easily reproducible due to dependencies on the experimental conditions. In this paper, we propose a mathematical model to evaluate the performance of an OpenFlow switch incorporated with traffic classification. We consider two representative classification methods, static classification and statistical classification, which can classify a wide range of traffic at high speed. Under some approximation assumptions, the model is analyzed by using queueing theory. We derive the mean and the C.V. (Coefficient of Variation) of setup delay of a flow, which is time from the arrival of a flow whose instruction is not registered in a flow table until the instruction is updated, and the mean number of flows stored for statistical classification as performance measures. Through numerical examples, we find that the analytical results agree well with simulation results as far as the traffic intensity at the switch is within the range of practical use.