Advanced Performance Profiling Using Perf Tools
summary
Shows a set of perf-based tools that map PMU memory samples to DWARF to attribute accesses to specific structs, fields and cacheline offsets, with new perf map/record/report wrappers and three new sort keys (type, type+offset, type cacheline). Introduces a latency profiling method that weights samples by runtime parallelism using scheduler info to surface serial bottlenecks (example: python shows 33% latency vs 16% CPU in a parallel build). Adds a production-focused kernel lock contention tool using tracepoints+BPF that reports spin/sleep wait times, top waiters, and best-effort symbolization via slab metadata or callstacks.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:13Performance profiling overview
- 6:23Profiling tools and platforms
- 12:47Data and memory analysis
- 18:10CPU performance optimization
- 25:59Production profiling
- 34:22Trace vs lockdep
related talks
45mPerf tools updates and beyond
44mObservability Tools C++: Beyond GDB and printf - Tools to Understand the Behavior of Your Program
45mFast and Safe Production Monitoring of JVM Applications with BPF Magic
29mPerf Support in Xen
38mAdvanced BPF profiling techniques: how to escape the curse of NMI and fetch ELF...
36mDeveloping Continuous eBPF Profiler to look Beneath the Kernel to Beyond the Clouds
26mProfiling Mastery: Decode Your Data and Elevate Observability
44mThink Outside the VM: Unobtrusive Performance Measurement