Abstract: It often requires an estimation of the number of the nodes with certain characteristics in P2P system.The problem is called size estimation.Since P2P environment tends to be dynamic and large scale，we choose sampling-based size estimation method because it has better performance with local information.Two proved algorithms for P2P environment are proposed which are based on sampling-collision and distribution.The experiments show the algorithms reduce cost greatly while the veracity deteriorates little.The contribution is that we find the algorithm based on distribution has better performance under “big sample” policy.

Key words: Size estimation, sampling method, P2P, capture-recapture

摘要： P2P系统经常需要分布式方法来估计系统中具有某种特征的节点数量，即规模估计。研究了基于抽样理论的规模估方法，该方法具有较好的健壮性和可扩展性。针对P2P应用，对两个基于抽样理论的规模估计算法进行了改进，分别是基于抽样冲突和基于样本分布算法。实验结果指出改进算法牺牲少量的精度而大大减小运行开销。并首次指出当总采样量不变时，基于样本分布的规模估计方法更适合采用“单次大样本”的策略。

关键词: 规模估计, 抽样方法, 对等网络, 捕获再捕获