The InfoQ Podcast Frederic Branczyk on Continuous Profiling Leveraging eBPF
Oct 3, 2022
Frederic Branczyk, CEO of PolarSignals and a key figure in the eBPF community, discusses the transformative potential of continuous profiling in cloud-native environments. He delves into innovative technologies like eBPF and Prometheus that enhance observability. Frederic also explores the Parca project’s architecture, the FrostDB database's evolution, and how profiling can extend beyond Kubernetes to offer deeper insights. This conversation uncovers the future of performance optimization and the growing community around continuous profiling tools.
AI Snips
Chapters
Transcript
Episode notes
Branczyk's Origin Story
- Frederic Branczyk shared how continuous profiling became his startup idea after a 2019 keynote at KubeCon.
- His journey included work at CoreOS, Red Hat, and inspiration from Google's profiling paper.
Continuous Profiling Explained
- Continuous profiling collects performance data constantly in production with low overhead using eBPF.
- This continuous sampling reveals rich insights about software behavior across time, beyond short profiling snapshots.
Low Overhead Profiling
- Running Parca continuous profiling in production typically adds about 1% CPU overhead, adjustable by sampling rate.
- This low overhead enables profiling of an entire infrastructure without heavy resource cost.