Performance Monitoring and Root Cause Analysis for Cloud-hosted Web Applications

In this paper, we describe Roots {a system for automatically identifying the "root cause" of performance anomalies in web applications deployed in Platform-as-a-Service (PaaS) clouds. Roots does not require application-level instrumentation. Instead, it tracks events within the PaaS cloud that are triggered by application requests using a combination of metadata injection and platform-level instrumentation. We describe the extensible architecture of Roots, a prototype implementation of the system, and a statistical methodology for performance anomaly detection and diagnosis. We evaluate the efficacy of Roots using a set of PaaS-hosted web applications, and detail the performance overhead and scalability of the implementation.