CRII: SHF: RUI: A Unified LLM-Empowered Framework for Systematic Software Performance Issue Testing, Localization and Optimization

Software performance is a critical quality attribute influencing system responsiveness, resource utilization, and user satisfaction. However, modern software systems face increasing performance challenges due to their complexity, making traditional performance testing and optimization approaches insufficient. This proposal introduces a unified, Large Language Model (LLM)-empowered framework to systematically address these challenges.

The project will improve software performance engineering by advancing automation in performance testing, issue localization, and optimization. Additionally, the project’s research results will be integrated into computer science curriculum and provide online training modules to equip future software engineers with state-of-the-art performance engineering skills.

The project will focus on automatic generation and prioritization of performance-centric test cases, leveraging Large Language Models (LLMs) to generate diverse test cases that specifically target performance issues. It will also introduce a multi-granularity monitoring framework, integrating dynamic profiling, static analysis, and Graph Neural Networks to accurately detect and localize performance bottlenecks across different system levels.

Finally, the project will develop a dependency-aware, graph-based optimization recommendation system, employing LLMs with Retrieval-Augmented Generation to provide adaptive performance enhancement strategies, from fine-grained code improvements to architectural refactoring.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.