Cascading failures often occur in congested networks such as the Internet. A cascading failure can be described as a three-phase process： generation, diffusion, and dissipation of the congestion. In this account, we present a function that represents the extent of congestion on a given node. This approach is different from existing fimctions based on betweenness centrality. By introducing the concept of ＇delay time＇, we designate an intergradation between permanent removal and nonremoval. We also construct an evaluation fimction of network efficiency, based on congestion, which measures the damage caused by cascading failures. Finally, we investigate the effects of network structure and size, delay time, processing ability and packet generation speed on congestion propagation. Also, we uncover the relationship between the cascade dynamics and some properties of the network such as structure and size.
针对数据存储规模的扩大,提出了一种基于融合主成分匹配FPCM(fusion principal components match)的异常检测方法。首先将各子节点数据通过聚类去除孤立点以提高主成分分析的稳定性,将各子节点的聚类中心传送到中心节点,减少节点间传送数据的通信量并且实现求主成分的数据融合;用聚类中心的主成分转换矩阵建立的正常行为模型能够体现整体的数据特征;最后使用决策树方法提高匹配速度。实验结果表明,FPCM方法能保持较高的DOS检测率,在保证整体检测率为97%的同时将误报率控制在10%以下。通过与已有方法比较表明,该方法能使分布式存储的数据在检测结果上达到数据集中存储的检测水平。
