Pythia: Diagnosing Performance Problems in Wide Area Providers.

Performance problem diagnosis is a critical part of network operations in ISPs. Service providers typically deploy monitoring nodes at several vantage points in their network, to record end-to-end measurements of network performance. Network operators use these measurements offline; for example, to troubleshoot customer complaints. In this work, we leverage such monitoring infrastructure deployments in ISPs to build a system for near real time performance problem detection and root cause diagnosis. Our system works with wide area interdomain monitoring, unlike approaches that require data sources from network devices (SNMP, Netflow, router logs, table dumps, etc.). Operators can input operational and domain knowledge of performance problems to the system to add diagnosis functionality. We have deployed the system on existing monitoring infrastructure in the US, diagnosing over 300 inter-domain paths. We study the extent and nature of performance problems that manifest in edge and core networks on the Internet.