Hierarchical Data Format Version Research Articles

A global four‐dimensional gridded dataset of ocean dissolved oxygen concentration retrieval from Argo profiles

AbstractLack of a long‐term time series of dataset with a high spatiotemporal resolution at a global scale poses a great challenge to clarify the characteristics of DOC in space and depth, and its trend in time. Thus, there is an urgent need for the development of a global DOC gridded dataset in space, time and depth. The Biogeochemical Argo (BGC‐Argo) provides an important data source for obtaining global DOC, but is limited by irregular spatial sampling locations. Besides, BGC‐Argo has shorter time series coverage and fewer profiles compared to Core‐Argo. Thus, this manuscript aims at reconstructing the DOC profiles based on the Core‐Argo and BGC‐Argo profiles and then developing a spatial, temporal and depth‐specific gridded DOC dataset, named G4D‐DOC. Validation results demonstrate that G4D‐DOC has a good overall consistency with WOA18 and GLODAPv2 datasets, particularly at depths of 10 dbar and 1000 dbar, where it surpasses consistency at other standard depths. In addition, compared to WOA18, G4D‐DOC has achieved a breakthrough in a temporal resolution from a climatological monthly to monthly, and compared to GLODAPv2, G4D‐DOC has achieved a breakthrough in space from irregular discrete locations to regular grids. Further, G4D‐DOC can be widely used to conduct the characteristics of DOC in space and depth and its trends at global and regional scales. The metadata of G4D‐DOC is as follows: four dimensions mean two dimensions in space (longitude and latitude), one in time and one in depth; data format is standard Hierarchical Data Format Version 4 (HDF4) with a spatial resolution of 1 degree and temporal resolutions of annual, seasonal and monthly intervals at 26 standard layers above 2000 dbar in depth; the spatial coverage is global and the time period is from 2005 to 2022.

H5bench: A unified benchmark suite for evaluating HDF5 I/O performance on pre‐exascale platforms

SummaryParallel I/O is a critical technique for moving data between compute and storage subsystems of supercomputers. With massive amounts of data produced or consumed by compute nodes, high‐performant parallel I/O is essential. I/O benchmarks play an important role in this process; however, there is a scarcity of I/O benchmarks representative of current workloads on HPC systems. Toward creating representative I/O kernels from real‐world applications, we have created h5bench , a set of I/O kernels that exercise hierarchical data format version 5 (HDF5) I/O on parallel file systems in numerous dimensions. Our focus on HDF5 is due to the parallel I/O library's heavy usage in various scientific applications running on supercomputing systems. The various tests benchmarked in the h5bench suite include I/O operations (read and write), data locality (arrays of basic data types and arrays of structures), array dimensionality (one‐dimensional arrays, two‐dimensional meshes, three‐dimensional cubes), I/O modes (synchronous and asynchronous). In this paper, we present the observed performance of h5bench executed along several of these dimensions on existing supercomputers (Cori and Summit) and pre‐exascale platforms (Perlmutter, Theta, and Polaris). h5bench measurements can be used to identify performance bottlenecks and their root causes and evaluate I/O optimizations. As the I/O patterns of h5bench are diverse and capture the I/O behaviors of various HPC applications, this study will be helpful to the broader supercomputing and I/O community.

Real-time data processing for ultrafast X-ray computed tomography using modular CUDA based pipelines

In this article, a new version of the Real-time Image Stream Algorithms (RISA) data processing suite is introduced. It now features online detector data acquisition, high-throughput data dumping and enhanced real-time data processing capabilities. The achieved low-latency real-time data processing extends the application of ultrafast electron beam X-ray computed tomography (UFXCT) scanners to real-time scanner control and process control. We implemented high performance data packet reception based on data plane development kit (DPDK) and high-throughput data storing using both hierarchical data format version 5 (HDF5) as well as the adaptable input/output system version 2 (ADIOS2). Furthermore, we extended RISA's underlying pipelining framework to support the fork-join paradigm. This allows for more complex workflows as it is necessary, e.g. for online data processing. Also, the pipeline configuration is moved from compile-time to runtime, i.e. processing stages and their interconnections can now be configured using a configuration file. In several benchmarks, RISA is profiled regarding data acquisition performance, data storage throughput and overall processing latency. We found that using direct IO mode significantly improves data writing performance on the local data storage. We could further prove that RISA is now capable of concurrently receiving, processing and storing data from up to 768 detector channels (3072 MB/s) at 8000 fps on a single-GPU computer in real-time. Program summaryProgram Title: GLADOS/RISACPC Library link to program files:https://doi.org/10.17632/65sx747rvm.2Developer's repository link:https://codebase.helmholtz.cloud/risaLicensing provisions: Apache-2.0Programming language: C++Journal reference of previous version: Comput. Phys. Commun. 219 (2017) 353-360 [1]Does the new version supersede the previous version?: Yes.Reasons for the new version: Extended capabilities for real-time operation with latest UFXCT hardware.Summary of revisions: (i) Add forking and joining of processing pipeline branches(ii) Add runtime (re-)configuration of pipeline stages and connections(iii) Add UDP receiver stage to acquire detector data in real-time(iv) Add high-throughput data dumpingNature of problem: Ultrafast electron beam X-ray computed tomography scanners stream multiple Gigabytes of raw data per second via Ethernet to a control computer. Receiving the data with low latency, real-time image-based control would become possible. For this, data need to be captured from the network, stored on disk, reconstructed and post-processed concurrently. The current total data rate of up to 3072 MB/s requires high-throughput solutions for each of these tasks.Solution method: Using a pipeline scheme, RISA processes incoming raw data in distinct stages (sources, processors, sinks). These are implemented in GPU kernels and are executed concurrently to exploit data parallelism as well as task parallelism. To capture detector data, we implemented a UDP packet capturing stage based on DPDK [2] which acts as a source stage. By allowing the pipeline to fork up into multiple branches, we concurrently acquire, store and process the data. For storing these data, we use the HDF5 format [3]. We achieve the required data rates by writing in direct IO mode onto an SSD array in RAID 0 configuration.Additional comments including restrictions and unusual features: RISA provides a set of general-purpose processing stages which are suitable for generic image stream processing.

A Robust and Fast Data Management System for Machine-Learning Research of Tokamaks

In recent years, machine-learning (ML) research methods have received increasing attention in the tokamak community. The conventional database (i.e., MDSplus for tokamak) of experimental data has been designed for small group consumption and is mainly aimed at simultaneous visualization of a small amount of data. The ML data access patterns fundamentally differ from traditional data access patterns. The typical MDSplus database is increasingly showing its limitations. We developed a new data management system suitable for tokamak ML research based on experimental advanced superconducting tokamak (EAST) data. The data management system is based on MongoDB and hierarchical data format version 5 (HDF5). Currently, the entire data management has more than 3000 channels of data. The system can provide highly reliable concurrent access. The system includes error correction, MDSplus original data conversion, and high-performance sequence data output. Furthermore, some valuable functions are implemented to accelerate ML model training of fusion, such as a bucketing generator, the concatenating buffer, and distributed sequence generation. This data management system is more suitable for fusion ML model research and development than MDSplus, but it cannot replace the MDSplus database. The MDSplus database is still the backend for EAST tokamak data acquisition and storage.

QuakeLabeler: A Fast Seismic Data Set Creation and Annotation Toolbox for AI Applications

Abstract The production and preparation of data sets are essential steps in machine learning (ML) applications. With the increasing volume and scale of available ML techniques in seismology, annotating seismograms or seismic features has become time consuming and tedious for many researchers. Furthermore, most methods train and validate on unique data subsets, which hampers independent performance evaluation and comparison. To address this problem, we have developed the software QuakeLabeler, an open-source Python package to customize, build, and manage earthquake training data sets, including processing and visualization. QuakeLabeler has tight pipeline functions, which include retrieving seismograms from multiple online data centers, querying online human-reviewed catalogs, signal processing, annotating (labeling), and analyzing data distribution. In addition, relevant statistical graphics and human-readable output files can be generated. Various file export formats are supported, such as Seismic Analysis Code (*.sac), mini Standard for Exchange of Earthquake Data (*.mseed), NumPy (*.npz), MATLAB (*.mat), and the Hierarchical Data Format version 5 (*.hdf5). This toolbox is packaged with an interactive command-line interface. Three alternative running modes (beginner, advanced, and benchmark) are implemented, intended to offer specific data set solutions for different types of applications, that is, quick-start recipes for simple ML solutions, advanced design for customized project training, and benchmark bulletins for model comparison.

Automated Log Data Analytics Workflow – The Value of Data Access and Management to Reduced Turnaround Time for Log Analysis

The oil and gas industry of today is undergoing rapid digitalization. This implies a massive effort to transform standard work procedures and workflows into more efficient practices and implementations using machine learning (ML) and automation. This will enable geoscientists to explore and exploit vast amounts of data quickly and efficiently. To address these current industry challenges, we propose a pilot well-log database in HDF5 (Hierarchical Data Format version 5) format that can be continuously extended if new data become available. It also provides versatility for data preparation for further analysis. We show an alternative way to store and use log files in a hierarchical structure that is easy to understand and handle by research institutes, companies, and academia. We also touch upon well-log depth matching, a long-standing industry challenge, to synchronize data from different logging passes to a single depth reference. Having a robust automated solution for depth matching is important to facilitate the use of all available data in a depth interval for analysis by ML. We propose an automatic well-log depth-matching workflow capable of handling multiple log types simultaneously and its integration with the database. The updated depth-matched logs are added to the database with their corresponding metadata, giving the geoscientist full control. We implemented two algorithms—classical cross correlation combined with a scaling factor to simulate stretch-squeeze effects and a constrained dynamic time warping (DTW). Our results indicate that the classical cross correlation outperforms the warping for both robustness and speed when the DTW is constrained to avoid excessive signal distortion and when the number of processed curves increases, respectively. Some limitations of our approach are related to large changes in the log patterns between the runs, as well as the assumption of negligible depth shift between log types within the same run. The cross correlation also allows a consistent application of depth matching to the metadata. This prototype workflow is tested using two wells from the Norwegian North Sea. We see the potential for extending this automatic database-processing workflow to give geoscientists access to all the data to improve interpretation.

ScDIOR: single cell RNA-seq data IO software

BackgroundSingle-cell RNA sequencing is becoming a powerful tool to identify cell states, reconstruct developmental trajectories, and deconvolute spatial expression. The rapid development of computational methods promotes the insight of heterogeneous single-cell data. An increasing number of tools have been provided for biological analysts, of which two programming languages- R and Python are widely used among researchers. R and Python are complementary, as many methods are implemented specifically in R or Python. However, the different platforms immediately caused the data sharing and transformation problem, especially for Scanpy, Seurat, and SingleCellExperiemnt. Currently, there is no efficient and user-friendly software to perform data transformation of single-cell omics between platforms, which makes users spend unbearable time on data Input and Output (IO), significantly reducing the efficiency of data analysis.ResultsWe developed scDIOR for single-cell data transformation between platforms of R and Python based on Hierarchical Data Format Version 5 (HDF5). We have created a data IO ecosystem between three R packages (Seurat, SingleCellExperiment, Monocle) and a Python package (Scanpy). Importantly, scDIOR accommodates a variety of data types across programming languages and platforms in an ultrafast way, including single-cell RNA-seq and spatial resolved transcriptomics data, using only a few codes in IDE or command line interface. For large scale datasets, users can partially load the needed information, e.g., cell annotation without the gene expression matrices. scDIOR connects the analytical tasks of different platforms, which makes it easy to compare the performance of algorithms between them.ConclusionsscDIOR contains two modules, dior in R and diopy in Python. scDIOR is a versatile and user-friendly tool that implements single-cell data transformation between R and Python rapidly and stably. The software is freely accessible at https://github.com/JiekaiLab/scDIOR.

The critical care data exchange format: a proposed flexible data standard for combining clinical and high-frequency physiologic data in critical care

Objective. To develop a standardized format for exchanging clinical and physiologic data generated in the intensive care unit. Our goal was to develop a format that would accommodate the data collection pipelines of various sites but would not require dataset-specific schemas or ad-hoc tools for decoding and analysis. Approach. A number of centers had independently developed solutions for storing clinical and physiologic data using Hierarchical Data Format—Version 5 (HDF5), a well-supported standard already in use in multiple other fields. These individual solutions involved design choices that made the data difficult to share despite the underlying common framework. A collaborative process was used to form the basis of a proposed standard that would allow for interoperability and data sharing with common analysis tools. Main Results. We developed the HDF5-based critical care data exchange format which stores multiparametric data in an efficient, self-describing, hierarchical structure and supports real-time streaming and compression. In addition to cardiorespiratory and laboratory data, the format can, in future, accommodate other large datasets such as imaging and genomics. We demonstated the feasibility of a standardized format by converting data from three sites as well as the MIMIC III dataset. Significance. Individual approaches to the storage of multiparametric clinical data are proliferating, representing both a duplication of effort and a missed opportunity for collaboration. Adoption of a standardized format for clinical data exchange will enable the development of a digital biobank, facilitate the external validation of machine learning models and be a powerful tool for sharing multiparametric, high frequency patient level data in multisite clinical trials. Our proposed solution focuses on supporting standardized ontologies such as LOINC allowing easy reading of data regardless of the source and in so doing provides a useful method to integrate large amounts of existing data.

HiSS-Cube: A scalable framework for Hierarchical Semi-Sparse Cubes preserving uncertainties

A wide variety of approaches are available for big data cube visualization and analysis. However, few exploit the power of array databases and none preserve the scientific uncertainties in measurements when constructing lower resolutions. In machine learning applications, we often need to rapidly search data for regions of interest and then focus on these areas, but without having to retrain the model every time we change the resolution. However, the reliable verification of these areas also requires details of the accuracy of the measured values. In this study, we developed a new software infrastructure called Hierarchical Semi-Sparse Cube (HiSS-Cube) based on Hierarchical Data Format version 5. HiSS-Cube enables visualization and machine learning using combined heterogeneous data and it was designed to be scalable for big data. HiSS-Cube allows data from multiple domains (imaging, spectral, and time-series data) to be combined and the construction of a multi-resolution semi-sparse data cube that preserves the uncertainties of scientific measurement at all resolutions. The functionality of HiSS-Cube was verified based on a subset of the Sloan Digital Sky Survey Stripe 82 survey. We compared the times and volumes for visualizations and machine learning data exported to HiSS-Cube and the original format (FITS). Using these data, we demonstrated that HiSS-Cube is faster by several orders of magnitude. HiSS-Cube supports export to the VOTable format and it is compatible with common Virtual Observatory tools. The source code for our prototype HiSS-Cube is available from GitHub and the data are available from Zenodo.

BD5: An open HDF5-based data format to represent quantitative biological dynamics data.

BD5 is a new binary data format based on HDF5 (hierarchical data format version 5). It can be used for representing quantitative biological dynamics data obtained from bioimage informatics techniques and mechanobiological simulations. Biological Dynamics Markup Language (BDML) is an XML (Extensible Markup Language)-based open format that is also used to represent such data; however, it becomes difficult to access quantitative data in BDML files when the file size is large because parsing XML-based files requires large computational resources to first read the whole file sequentially into computer memory. BD5 enables fast random (i.e., direct) access to quantitative data on disk without parsing the entire file. Therefore, it allows practical reuse of data for understanding biological mechanisms underlying the dynamics.

ExaHDF5: Delivering Efficient Parallel I/O on Exascale Computing Systems

Scientific applications at exascale generate and analyze massive amounts of data. A critical requirement of these applications is the capability to access and manage this data efficiently on exascale systems. Parallel I/O, the key technology enables moving data between compute nodes and storage, faces monumental challenges from new applications, memory, and storage architectures considered in the designs of exascale systems. As the storage hierarchy is expanding to include node-local persistent memory, burst buffers, etc., as well as disk-based storage, data movement among these layers must be efficient. Parallel I/O libraries of the future should be capable of handling file sizes of many terabytes and beyond. In this paper, we describe new capabilities we have developed in Hierarchical Data Format version 5 (HDF5), the most popular parallel I/O library for scientific applications. HDF5 is one of the most used libraries at the leadership computing facilities for performing parallel I/O on existing HPC systems. The state-of-the-art features we describe include: Virtual Object Layer (VOL), Data Elevator, asynchronous I/O, full-featured single-writer and multiple-reader (Full SWMR), and parallel querying. In this paper, we introduce these features, their implementations, and the performance and feature benefits to applications and other libraries.

NeuroChaT: A toolbox to analyse the dynamics of neuronal encoding in freely-behaving rodents in vivo

There is a dearth of freely-available, standardised open source analysis tools available for the analysis of neuronal signals recorded in vivo in the freely-behaving animal. In response, we have developed a freely-available, open-source toolbox, NeuroChaT ( Neuron Characterisation Toolbox), specifically addressing this lacuna. Although we have particularly emphasised single unit analyses for spatial coding, NeuroChaT also characterises rhythmic properties of units and their dynamics associated with local field potential signals. NeuroChaT was developed using Python and facilitates a complete pipeline from automation of analysis to producing and managing publication-quality figures. Additionally, we have adopted a platform-independent format (Hierarchical Data Format version 5) for storing recorded and analysed data. By providing an easy-to-use software package, we aim to simplify the adoption of standardised analyses for behavioural neurophysiology and facilitate open data sharing and collaboration between laboratories.

NeuroChaT: A toolbox to analyse the dynamics of neuronal encoding in freely-behaving rodents in vivo.

There is a dearth of freely-available, standardised open source analysis tools available for the analysis of neuronal signals recorded in vivo in the freely-behaving animal. In response, we have developed a freely-available, open-source toolbox, NeuroChaT ( Neuron Characterisation Toolbox), specifically addressing this lacuna. Although we have particularly emphasised single unit analyses for spatial coding, NeuroChaT also characterises rhythmic properties of units and their dynamics associated with local field potential signals. NeuroChaT was developed using Python and facilitates a complete pipeline from automation of analysis to producing and managing publication-quality figures. Additionally, we have adopted a platform-independent format (Hierarchical Data Format version 5) for storing recorded and analysed data. By providing an easy-to-use software package, we aim to simplify the adoption of standardised analyses for behavioural neurophysiology and facilitate open data sharing and collaboration between laboratories.

Scientist and data architect collaborate to curate and archive an inner ear electrophysiology data collection.

In the past scientists reported summaries of their findings; they did not provide their original data collections. Many stakeholders (e.g., funding agencies) are now requesting that such data be made publicly available. This mandate is being adopted to facilitate further discovery, and to mitigate waste and deficits in the research process. At the same time, the necessary infrastructure for data curation (e.g., repositories) has been evolving. The current target is to make research products FAIR (Findable, Accessible, Interoperable, Reusable), resulting in data that are curated and archived to be both human and machine compatible. However, most scientists have little training in data curation. Specifically, they are ill-equipped to annotate their data collections at a level that facilitates discoverability, aggregation, and broad reuse in a context separate from their creation or sub-field. To circumvent these deficits data architects may collaborate with scientists to transform and curate data. This paper’s example of a data collection describes the electrical properties of outer hair cells isolated from the mammalian cochlea. The data is expressed with a variant of The Ontology for Biomedical Investigations (OBI), mirrored to provide the metadata and nested data architecture used within the Hierarchical Data Format version 5 (HDF5) format. Each digital specimen is displayed in a tree configuration (like directories in a computer) and consists of six main branches based on the ontology classes. The data collections, scripts, and ontological OWL file (OBI based Inner Ear Electrophysiology (OBI_IEE)) are deposited in three repositories. We discuss the impediments to producing such data collections for public use, and the tools and processes required for effective implementation. This work illustrates the impact that small collaborations can have on the curation of our publicly-funded collections, and is particularly salient for fields where data is sparse, throughput is low, and sacrifice of animals is required for discovery.

A SPARQL query engine for binary-formatted IFC building models

Abstract To date, widely implemented and full-featured query languages for building models in their native exchange formats do not exist. While interesting proposals exist for querying Industry Foundation Classes (IFC) models, their functionality is often incomplete and their semantics not precisely defined. With the introduction of the ifcOWL ontology as an equivalent to the IFC schema in the Web Ontology Language (OWL), an option to represent such models in RDF (Resource Description Framework, a general information modeling method) is provided, and such models can be queried using SPARQL (SPARQL Protocol and RDF Query Language). The size of data sets in complex building projects, however, renders the use of clear-text encoded RDF infeasible in many cases. A SPARQL implementation, compatible with ifcOWL, is proposed, directly atop a standardized binary serialization format for IFC building models. This novel format is the binary equivalent of traditional IFC serialization formats but with more compact storage and less overhead than the graph serialization in RDF. The format is based on ISO 10303-26 and relies on an open standard for organizing large amounts of data: Hierarchical Data Format version 5 (HDF5). Due to hierarchical partitioning and fixed-length records, only small subsets of the data are read to answer queries, improving efficiency. A prototypical implementation of the query engine is provided in the Python programming language. In several realistic use cases, the proposed system performs equivalent to or better than the state of the art in SPARQL querying on building models. For large datasets, the proposed storage format results in files that are 2–3 times smaller than the current, most concise, RDF databases while offering a platform-neutral, containerized exchange file.

Design and optimisation of an efficient HDF5 I/O Kernel for massive parallel fluid flow simulations

SummaryMore and more massive parallel codes running on several hundreds of thousands of cores are entering the computational science and engineering domain, allowing high‐fidelity computations on up to trillions of unknowns for very detailed analyses of the underlying problems. Such runs typically produce gigabytes of data, hindering both efficient storage and (interactive) data exploration. Advanced approaches based on inherently distributed data formats such as hierarchical data format version 5 become necessary here to avoid long latencies when storing the data and to support fast (random) access when retrieving the data for visual processing. This paper shows considerations and implementation aspects of an I/O kernel based on hierarchical data format version 5 that supports fast checkpointing, restarting, and selective visualisation using a single shared output file for an existing computational fluid dynamics framework. This functionality is achieved by including the framework's hierarchical data structure in the file, which also opens the door for additional steering functionality. Finally, the performance of the kernel's write routines are presented. Bandwidths close to the theoretical peak on modern supercomputing clusters were achieved by avoiding file‐locking and using collective buffering.

Applying n-bit floating point numbers and integers, and the n-bit filter of HDF5 to reduce file sizes of remote sensing products in memory-sensitive environments

Memory sensitive applications for remote sensing data require memory-optimized data types in remote sensing products. Hierarchical Data Format version 5 (HDF5) offers user defined floating point numbers and integers and the n-bit filter to create data types optimized for memory consumption. The European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) applies a compaction scheme to the disseminated products of the Day and Night Band (DNB) data of Suomi National Polar-orbiting Partnership (S-NPP) satellite’s instrument Visible Infrared Imager Radiometer Suite (VIIRS) through the EUMETSAT Advanced Retransmission Service, converting the original 32 bits floating point numbers to user defined floating point numbers in combination with the n-bit filter for the radiance dataset of the product. The radiance dataset requires a floating point representation due to the high dynamic range of the DNB. A compression factor of 1.96 is reached by using an automatically determined exponent size and an 8 bits trailing significand and thus reducing the bandwidth requirements for dissemination. It is shown how the parameters needed for user defined floating point numbers are derived or determined automatically based on the data present in a product.

Enabling Scientific Collaboration and Discovery Through the Use of Data Standardization

The Railgun Data Format Standard (RDFS) provides a flexible format for the archival of electromagnetic railgun (EMRG) laboratory test data. This effort utilizes the hierarchical data format version 5 for the data model and file format, as well as extensible markup language for data structure and data type encapsulation. The RDFS was developed through a collaborative effort within the Office of Naval Research (ONR) EMRG program, including members from the Naval Surface Warfare Center Dahlgren, the Naval Research Laboratory, and the Institute for Advanced Technology. The intent behind the RDFS is to facilitate collaborative data sharing among EMRG researchers through the standardization of EMRG test data archives. The structure of the RDFS is discussed, along with details on its implementation and use within the ONR EMRG community.

Reimplementing the Hierarchical Data System using HDF5

The Starlink Hierarchical Data System has been a very successful niche astronomy file format and library for over 30 years. Development of the library was frozen ten years ago when funding for Starlink was stopped and almost no-one remains who understands the implementation details. To ensure the long-term sustainability of the Starlink application software and to make the extensible N-Dimensional Data Format accessible to a broader range of users, we propose to re-implement the HDS library application interface as a layer on top of the Hierarchical Data Format version 5. We present an overview of the new implementation of version 5 of the HDS file format and describe differences between the expectations of the HDS and HDF5 library interfaces. We finish by comparing the old and new HDS implementations by looking at a comparison of file sizes and by comparing performance benchmarks.

HDF5 based parallel I/O technology for structure grid CFD applications

Large scale parallel Computational Fluid Dynamics( CFD) simulation has posed high demand on the data I /O capabilities. Hierarchical Data Format version 5( HDF5) can effectively manage large scale scientific data and has a good support to parallel I / O. The HDF5 data storage model for a parallel CFD application was designed. The parallel I / O method for the application's data was implemented based on the HDF5 parallel I / O API. Performance experiments were performed on a parallel computer system. The results show that the data writing speed of this HDF5 based parallel I / O method outperforms the parallel I / O methods( i. e. each process writes an ordinary data file independently) by 6. 9 to 16. 1 times, when 4 to 32processes are used. The data reading speed of this HDF5 based parallel I / O method is slower than the parallel I / O methods,with a speed of 20% to 70% times of the latter. However, the cost of data reading time is much smaller than the cost of data writing time, hence has minor effect on the total performance.