Explore a conference talk on PCAT, a production change-aware telemetry system designed to handle changes in fast-evolving networks at Facebook. Delve into the challenges of network telemetry in large-scale production environments, focusing on the impact of constant changes in data collection, processing, interpretation, and consumption by applications. Learn about the innovative change cube abstraction for systematically tracking changes and the intent-based layering design for confining and monitoring modifications. Gain insights into real-world incidents demonstrating the magnitude of changes and their effects on multiple components. Discover how PCAT addresses these challenges, including its approach to change attribution and topology derivation. Consider open questions in adaptive telemetry primitives and trustful telemetry as you explore the future of network monitoring in rapidly evolving environments.
Overview
Syllabus
Intro
Network Telemetry is Critical for Network Managem
Key Challenge for Telemetry in Production: Evolvability Network devices and management applications are constantly evolving
Magnitude of Changes
Incident 1: Changes Affect Many Components
Incident 2: Data Misinterpretation
Bringing Changes to First-Class Citizens in Telemetry
PCAT: Production Change-Aware Telemetry System
Change Abstraction: Change Cube
Change Attribution: Three Generations of Telemetry
and Gen2
's Problems
PCAT) Layering Design
Change Exploration: Topology Derivation Creates derived topology from normalized device-level data.
Open Questions: Adaptive Telemetry Primitives
Open Questions: Trustful Telemetry
Summary
Taught by
USENIX
