On finding vestiges of Internet backbone failures for optimizing wide area data replication

To design an efficient and effective data replication strategy, we are trying to collect real data on Internet communication failures. Here, we present results of our ongoing effort to find vestiges of Internet backbone failures. Our approach is to apply a statistical outlier detection method on graph-theoretic metrics to pick up suspicious cases within AS-level topology data that have accumulated for years. We report that the number of nodes and average clustering coefficient are promising components.