Intelligent network management using graph differential anomaly visualization

Managing large-scale networks involving users and applications is challenging due to the complexity and dynamic nature of the heterogeneous graphs. How to quickly identify the meaningful changes and hidden anomalous activities in the spatiotemporally dynamic network graphs is essential in many aspects of network management, such as security, performance and troubleshooting. In this paper, we explore the viability and efficacy of a novel graph differential anomaly visualization (DAV) model in the area of network management. Our approach combines algorithmic graph analysis methods and visualization technologies by taking advantages from both computer and human intelligence. We focus on DAV at various levels, i.e., nodes, links and communities. Specifically, a novel community-based DAV scheme is proposed that can help understand the managed networks with a right balance of granularity and complexity. More importantly, the community-based DAV algorithm is less susceptible to network dynamics and high churn. The developed visual analytic tool can not only detect but more importantly find the root causes of anomalies in a time efficient manner.