Detecting and Diagnosing Network Performance Degradations with Statistical Methods

Nick Feamster
Georgia Institute of Technology

Users and network operators often need to efficiently and accurately diagnose the source of network performance problems. Conventionally, users and operators have relied on domain expertise to locate the source of a network disruption; unfortunately, performance problems often result from unique causes, making it difficult, if not impossible, to enumerate all possible such causes. In this talk, we examine the ability of statistical methods to help users and operators diagnose network disruptions and performance degradations. We focus on two specific cases of performance degradations within an Internet service provider network: (1) detecting and locating causes of routing disruptions; and (2) identifying degradations in application-level performance.

Presentation (PowerPoint File)

Back to Workshops IV: New Mathematical Frontiers in Network Multi-Resolution Analysis