Relative traffic gain as a metric for network coding performance evaluation

With more and more network coding methods being put forward, the metric of network coding performance becomes a key issue in network coding application. In this paper, relative traffic gain (RTG) is proposed to measure the performance of network coding; it is defined as the expected value of the ratio of saved traffic flows to the sum of all traffic flows. Considering the diversity of actual network traffic flow, the relationship between relative traffic gain and parameters of different traffic models is analyzed, including normal distribution model, Poisson distribution model, Constant Bit-Rate (CBR) traffic model, exponential distribution model and Pareto distribution model. Independent bidirectional traffic flows and multiple traffic flows are taken as examples to analyze the relative traffic gain of network coding. The results show that: better relative traffic gain can be obtained under larger mean value U and smaller standard deviation σ for normal distribution, larger expected value λ for Poisson distribution, and smaller difference between the traffic flows' rates for CBR distribution, smaller λ for exponentional distribution and lager k for Pareto distribution. The results would help in measuring NC performance and give a reference to implementation and optimization of NC.