Characterizing Spatial Data Traits for Modeling Generic Lossy Rate-Distortion Quality

Modern high-performance computing (HPC) applications generate vast scientific data, straining storage and I/O. Error-bounded lossy compression helps balance data size and fidelity, but different compressors yield varying rate-distortion trade-offs. While rate-distortion curves can guide compressor choice, no efficient runtime solution exists. To address this, we propose X-QUAL, a compressor-agnostic framework that models rate-distortion behavior across compressors to support online decisions. By analyzing spatial data relationship changes, X-QUAL swiftly estimates rate-distortion curves. Evaluation results show that, on average, X-QUAL achieves low estimation error and 15× speedup over related works.