Time Stratified Packet Sampling Based on Measurement Buffer for Flow Measurement
流测量中基于测量缓冲区的时间分层分组抽样

Although NetFlow is widely deployed for traffic measurement, the sampling method of Netflow has shortcomings: it consumes excessive router resource during flooding attacks; selecting a suitable static sampling rate is difficult because no single rate gives the right tradeoff between resource consumption and accuracy for all traffic mixes. An easily-implemented packet sampling method is presented in this paper, which samples a fixed number of packets in the constant period with measurement buffer. The method can automatically adapt the sampling rate to traffic variety and provide the controllability of resource consumption. Theoretical analyses demonstrate that the new method can provide unbiased estimation with certain relative standard deviation bound. Experiments are also conducted with the real network traces. Results show that the proposed method can achieve simplicity, adaptability and controllability of resource consumption without sacrificing accuracy compared with other sampling methods.