Profiling the HotSpot JVM Using eBPF

on YouTube ↗45m

" summary

Shows how to build a JVM profiler that collects runtime information and produces flame graphs by using eBPF (Linux extended Berkeley Packet Filter) instead of the JVM TI interface. Covers using eBPF to observe HotSpot JVM execution and gather profiling data for performance optimization of Java applications.

AI-written summary · about

" chapters (14)

1. 0:00Introduction
2. 1:33What is eBPF
3. 3:00Java profiling issues
4. 8:02eBPF basics
5. 10:51How eBPF profiler works
6. 14:51Running eBPF code
7. 17:19Naive profiling approach
8. 18:45eBPF kernel limits
9. 21:59Profiler architecture
10. 22:40Comparison with profilers
11. 23:23eBPF development challenges
12. 28:19Kernel version issues
13. 30:45Why eBPF profiler
14. 33:17Q&A

" related talks

• 46mCPU Profiling on the JVM and Flame GraphsJerónimo López · Codemotion Madrid 2017
• 01h 07meBPF: Fueling New Flame Graphs & moreBrendan Gregg · YOW! Melbourne 2022
• 27mLow-Overhead Tracing Using eBPF for Observability into Kubernetes Apps and ServicesGaurav Gupta · KubeCon + CloudNativeCon Europe 2018
• 50mSet Your Code on Fire with FlameGraphsFrancesco Nigro · Voxxed Days Ticino 2018
• 36mObservability and performance analysis with BPFDavid Calavera · ScaleConf Colombia 2019
• 24meBPF Application Instrumentation for Java: Challenges, Design, and Real...Endre Sara, Stephen Lang · Observability Summit North America 2026
• 59mAn Introduction to JVM PerformanceRafael Winterhalter · JavaDay Lviv 2016
• 40mhello-ebpf: Writing eBPF Programs Directly in JavaJohannes Bechberger · JCON EUROPE 2025