Diagnosing Performance Issues in Application-Defined Resources
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
gigiprofiler is a C/C++ tool built on LLVM that combines LLM-based semantic inference (GPT‑4) with static validation and runtime instrumentation to expose application‑defined resources via thread‑local event buffers. It reproduces and diagnoses 15 real‑world performance problems across MySQL, PostgreSQL, MariaDB, Apache HTTPD, and llama.cpp, correctly finding all 15 and uncovering two new MariaDB issues; LLM stage shows ~55% false positives reduced by static (−21.43%) and dynamic (−26.76%) checks.
AI-written summary · about
chapters (9)
- 0:00Diagnosing performance issues
- 1:21Application-defined resources in MySQL
- 2:50Buffer subsystem case study
- 5:20Profiling limitations
- 6:45Empirical study findings
- 8:52GG Profiler design
- 13:29Diagnosis workflow results
- 15:19Model reliability metrics
- 16:49Summary
related talks
47mPractical Performance Analysis
37mAdvanced Performance Profiling Using Perf Tools
52mApplication-Level Debugging and Profiling: Gaps in the Tool Ecosystem
01h 04mProfiling PHP Applications
24mPythia: An Automated, Cross-layer Instrumentation Framework for Diagnosing Performanc...
21mAPMs, Profilers Static Analyzers Who’s who
44mObservability Tools C++: Beyond GDB and printf - Tools to Understand the Behavior of Your Program
41mGain a deeper understanding of your incidents: continuous profiling