A time window based analytical approach for splitting high-speed network traffic

Traffic splitting plays a very important role in parallel traffic replay since it determines the amount of required PCs and guarantees replay quality for each generated sub-trace. Previously adopted simulation-based method for splitting high-speed network traffic introduces serious performance limitation as it distributes traffic load to the replay PCs in an iterative trial-and-error manner. In this paper we present a time window based analytical approach for splitting high-speed network traffic. In comparison with the existing flow-based simulation methods, our approach has the following two advantages: 1) the number of replay PCs and their corresponding sub-traces are automatically obtained by one execution of the algorithm, making the method computationally efficient; 2) the maximum and average packet delay time in each of the generated sub-trace is theoretically guaranteed prior to replay. Two metrics to evaluate the replay discrepancies are also proposed. Experimental results with actual high-speed traces show that the method outperforms the current flow-based simulation methods in splitting the traffic and the quality of the generated sub-traces are strictly ensured.