Replication Data for: Parallel I/O analysis in Distributed Deep Learning applications on High-Performance Computing
Edixon Párraga, Betzabeth León · CORA.Repositori de Dades de Recerca · 2026
Not cached yet. This record was fetched live for this request and is stored after the lookup. Reopening the page should serve the cached copy.
CORA.Repositori de Dades de Recerca · Consorci de Serveis Universitaris de Catalunya · dataset · 2026
Abstract
This reproducibility dataset contains the experimental evidence, processed data products, analysis outputs, figures, metadata, validation products, and software artifacts associated with the article “Parallel I/O analysis in distributed deep learning applications on high-performance computing”. The deposited material documents parallel I/O experiments performed with the DLIO and DeepGalaxy workloads on an HPC system using Lustre storage.
It preserves the provenance chain linking HPC executions and JobIDs with Darshan traces, DXT access-pattern information, Parser and performance products, execution indexes, consolidated results, generated figures, Figure Registry mappings, Golden Reference information, and Article Validation products. The dataset supports reconstruction and verification of Figures 2-16 of the associated publication. Figures 2-5 and Figure 10 characterize I/O access patterns from DXT-derived evidence; Figures 6-9 and 11-16 contain I/O performance results.
Figure 10 and its underlying evidence are retained as scientifically valid; partial automatic panel resolution in the deposited validation snapshot is documented as an automation/provenance limitation rather than a failure of the experimental result. The physical synthetic DLIO input datasets and the original DeepGalaxy input file output_bw_512.hdf5 are not redistributed. Their provenance, software context, and relevant reproduction information are documented separately.
Open the resolved locator · 10.34810/data3618
Metadata licence: OpenAlex metadata (CC0)