本文将大规模软件动态运行信息映射为随时间变化的调用关系网络，通过采用K 核分解算法提取此静态网络 的层次结构与核心节点信息进行可视化显示；同时基于矩阵二范数方法对动态网络的结构进行频谱分析实现故障定位；对大 规模软件的运行状况进行了可更新的监控，实现了异常的预测，把实验的结果与理论很好地结合在一起，验证了预测模型的有 效性，并为本模型的进一步扩展提供了可能。最后实现了该系统的可视化界面。

This thesis maps the method call relation to a dynamic network, extracts the hierarchical structure and core vertices of the original network based on K-core decomposition, analyzes the dynamic network by 2 norm of the matrix theory, considers the characteristics of the monitoring data and the change of software status. We design a real-time updated visual interface. It achieves good experimental results to verify the validity of the prediction model, and gives the further expansion of this model.