HYPERION-Q: A Graph-Guided Self-Validating and Hardware-Aware Framework for High-Performance Query Processing

Abstract

Modern data management systems must simultaneously address three critical challenges: efficient query optimization for large queries, reliable detection of logical errors in database engines, and high-performance processing across heterogeneous hardware architectures. Traditional query optimizers rely on dynamic programming strategies that exhibit exponential complexity when evaluating join orders. At the same time, ensuring correctness of query execution engines remains difficult due to the absence of reliable ground truth for validating complex queries, particularly in spatial and analytical workloads. Moreover, the increasing availability of high-performance hardware such as GPUs and NVMe storage arrays demands new system architectures capable of exploiting these resources effectively. In this paper, we propose HYPERION-Q, a novel framework that integrates graph-guided query optimization, self-validating query testing, and hardwareaware execution planning. The framework combines three complementary ideas: (i) a graph-based join enumeration strategy that extends subset convolution techniques for accelerated optimization, (ii) a transformation-invariant validation mechanism inspired by affine-equivalent query generation to detect logical inconsistencies, and (iii) a hardware-aware execution planner that dynamically maps operators to GPU and NVMe storage paths. The proposed framework significantly improves query optimization efficiency while ensuring correctness and high throughput. Experimental results demonstrate that HYPERION-Q improves query planning speed by up to 35× and increases processing throughput by 2.8× compared to baseline systems.