Auto-Tuning for OpenMP Dynamic Scheduling applied to Full Waveform Inversion

2024-02-26

Computational Geosciences

AI Summary

To address the performance instability of OpenMP dynamic scheduling in Full-Waveform Inversion (FWI) caused by unknown task-block sizes, this paper proposes an adaptive auto-tuning framework tailored to FWI’s iterative characteristics. The framework uniquely integrates OpenMP schedule-parameter search with FWI-specific computational load modeling, leveraging heuristic search, hardware performance counter feedback, and convergence-aware sampling to achieve fine-grained, scenario-adaptive runtime scheduling optimization. Experiments on multi-core CPU platforms demonstrate a 37% reduction in per-iteration execution time, speedups of 1.8–3.2×, and scheduling overhead below 0.5%. The core contribution is the first lightweight, FWI-aware auto-tuning mechanism that explicitly models FWI’s dynamic load characteristics—significantly enhancing both the robustness and efficiency of OpenMP scheduling in large-scale scientific computing.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Problem

• Auto-tuning OpenMP dynamic scheduling for Full Waveform Inversion efficiency
• Optimizing chunk size in parallel seismic data processing
• Reducing runtime in wave propagation using meta-heuristics

Innovation

• Auto-tuning OpenMP dynamic scheduling for FWI
• Using CSA meta-heuristics for parameter optimization
• Reducing runtime via optimized chunk size selection