HP-MDR: High-performance and Portable Data Refactoring and Progressive Retrieval with Advanced GPUs

TL;DR

HP-MDR tackles the data deluge in scientific computing by delivering a GPU-accelerated end-to-end pipeline for data refactoring and progressive retrieval with QoI error control. It introduces optimized bitplane encoding designs, a hybrid lossless compressor, and a pipeline that overlaps CPU-GPU operations to boost throughput while preserving portability across GPU architectures via HPDR. On real-world datasets, HP-MDR achieves up to six-point-six times throughput in data refactoring and progressive retrieval, ten-point-four times throughput for QoI based data recomposition, and four-point-two times end-to-end retrieval compared to state-of-the-art. These results demonstrate practical impact for exascale systems by reducing data movement and enabling flexible accuracy controls in downstream analytics, including quantities of interest such as $V_{total}$.

Abstract

Scientific applications produce vast amounts of data, posing grand challenges in the underlying data management and analytic tasks. Progressive compression is a promising way to address this problem, as it allows for on-demand data retrieval with significantly reduced data movement cost. However, most existing progressive methods are designed for CPUs, leaving a gap for them to unleash the power of today's heterogeneous computing systems with GPUs. In this work, we propose HP-MDR, a high-performance and portable data refactoring and progressive retrieval framework for GPUs. Our contributions are three-fold: (1) We carefully optimize the bitplane encoding and lossless encoding, two key stages in progressive methods, to achieve high performance on GPUs; (2) We propose pipeline optimization and incorporate it with data refactoring and progressive retrieval workflows to further enhance the performance for large data process; (3) We leverage our framework to enable high-performance data retrieval with guaranteed error control for common Quantities of Interest; (4) We evaluate HP-MDR and compare it with state of the arts using five real-world datasets. Experimental results demonstrate that HP-MDR delivers up to 6.6x throughput in data refactoring and progressive retrieval tasks. It also leads to 10.4x throughput for recomposing required data representations under Quantity-of-Interest error control and 4.2x performance for the corresponding end-to-end data retrieval, when compared with state-of-the-art solutions.

Figures (14)

• Figure 1: High-perf. portable multi-precision data refactoring framework (HP-MDR)
• Figure 2: Three optimized designs for bitplane encoding
• Figure 3: Four encoding designs using register shuffling
• Figure 4: Optimized refactoring and reconstruction pipeline represented as DAGs that enable data transfer latency hiding. Green: host-to-device copy; Red: device-to-host copy; Blue: compute; Yellow: copy-compute mixed
• Figure 5: Illustration of the three error bound estimation methods: CPU Porting (CP), Minimal Augmentation (MA), and Minimal Augmentation with Proportional Estimation (MAPE). A lower number of bitplanes indicates a smaller retrieval size, and fewer iterations indicate higher throughput.