Reproduce “gigiProfiler: Diagnosing Performance Issues by Uncovering Application Resource Bottlenecks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yigong Hu, Haodong Zheng, Yicheng Liu, Dedong Xie, Youliang Huang, Baris Kasikci
gigiProfiler introduces OmniResource Profiling to diagnose performance issues stemming from application-defined resource bottlenecks by integrating application-level resource tracing with traditional system-level profiling. It utilizes a hybrid LLM-static analysis approach to automatically identify custom resources and their operators, accurately diagnosing 12 real-world issues in major applications, including two previously unresolved MariaDB problems, with an average runtime overhead of 5.86%.
Key insights
- Combining Large Language Models (for high-level semantic understanding) with static analysis (for precise validation) enables effective automated identification of custom application resources and their operators.
- Application-defined resource contention often involves threads waiting on custom internal resources, which is invisible to traditional profilers that only track CPU or system call usage.
- Comparing critical variable values and loop execution patterns between different runs can accurately diagnose specific code misbehaviors causing application-level performance bottlenecks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.