The Evaluation of Poisson-Based Packet Sampling Techniques
Zhenming Lei · Beijing Youdian Xueyuan xuebao · 2005
An event-triggering packet sampling process based on Poisson distribution is proposed. Two types of Poisson sampling techniques in types of timer triggering and event triggering were evaluated on measurement precision in aspects of mean packet length, packet length distribution and whole traffic volume. It shows that event triggering Poisson sampling technique, with O(n) measurement overhead, performs better than time triggering one, and the sampling error of event-triggering Poisson sampling is only 1/2～1/3 of time triggering one. Event triggering Poisson sampling technique is more eligible for high-speed network measurement environment to meet high precision requirement.