Abstract: In unstructured P2P，the ordinary nodes select a super node by considering synthetically multiple factors such as distance cost，credibility，content similarity，the dynamic processing power of CPU and etc.And the dynamic processing power of CPU exerts the most important factor in all undoubtedly.This paper presents a new approach aiming at how to choose a super node.It predicts the dynamic load of CPU using a Radial Basis Function（RBF） neural network，assesses the dynamic processing power of super node’s CPU by the dynamic loading of CPU and provides the information for ordinary nodes，which helps ordinary select a most suitable super node.Though analysis and simulation experience results indicate that this approach can improve performance of the whole system efficiently，system cost reduction of 96% or more.

Key words: P2P（peer-to-peer）, Radial Basis Function（RBF） neural network, super node selection, dynamic processing power

摘要： 在非结构化P2P网络中，普通节点通常要综合考虑距离、信誉度、内容相似度、CPU动态处理能力等多种因素来选择一个合适的超级节点，其中CPU的动态处理能力无疑是最要的。针对如何选择超级节点的问题提出一种新的方法，该方法通过RBF神经网络预测CPU动态负载，以CPU的动态负载评定超级节点的动态处理能力并将该信息提供给普通节点，帮助普通节点选择一个最合适的超级节点。通过分析和仿真实验表明该方法有效地提高了系统的性能，系统开销减少了96%以上。

关键词: P2P网络, RBF神经网络, 超级节点选取, 动态处理能力