使用 Homebrew, Nix 安装 hdf5-mpi

项目链接

• 主页
• 仓库
• 上游文档
• 软件包管理器页面

命令和别名

• h5pcc
• h5perf
• h5pfc
• ph5diff

项目历史

HDF began at NCSA in 1987 as an architecture-independent scientific data library and file format. By 1996, HDF's maintainers considered the older line insufficient for larger data volumes and newer computing systems, and in 1998 HDF5 was implemented as the more scalable successor to HDF4.

Parallel HDF5 was designed around compatibility with serial HDF5 files, a single shared file image across processes, and a portable standard parallel I/O interface. The HDF Group documentation says the initial target was MPI programming rather than shared-memory programming, matching the MPI-enabled package role.

采用历史

The MPI-enabled package exists because HPC users often need the same HDF5 data model but with collective access from multiple processes. Upstream installation guidance for Parallel HDF5 calls out MPI compilers with MPI-IO support and POSIX-compliant parallel file systems, which maps directly to cluster module and package-manager variants.

For Homebrew and Nix users, hdf5-mpi separates the parallel build from the default serial package so dependent software can link against the expected ABI and compiler wrappers. The package role is about build configuration and dependency selection, not a separate upstream project.

使用方式

Parallel HDF5 is used by MPI applications that need many processes to create, open, read, write, and close HDF5 files while preserving serial-file compatibility. Upstream CMake guidance enables this path with HDF5_ENABLE_PARALLEL=ON and MPI compilers such as mpicc or mpif90.

The package exposes MPI-aware compiler wrappers and tools recorded in the batch input, including h5pcc, h5pfc, h5perf, and ph5diff. Those names signal to build systems and users that this package is for parallel HDF5 development and testing.

为什么软件包爱好者会关心

hdf5-mpi is a classic package-manager split: same upstream, different feature matrix. It matters because mixing serial HDF5, MPI HDF5, C, Fortran, C++, thread-safety, filters, and compiler wrappers can break downstream builds in subtle ways.

Package users care about this variant when building scientific Python, Julia, Fortran, C, and C++ software against cluster MPI stacks. The package is less about command-line novelty and more about getting the right library compiled with the right parallel I/O assumptions.

时间线

• 1987: NCSA began work on the original architecture-independent HDF technology.
• 1998: HDF5 was implemented as a more scalable successor to HDF4.
• 1998: Parallel HDF5 design work targeted MPI programming and portable parallel I/O.
• 2006: The HDF Group began operating as a separate nonprofit steward.
• 2020: The official HDF5 source repository was created on GitHub.
• 2026: Batch input records Homebrew and Nix packaging for the MPI-enabled package variant.

Related projects

• HDF5 is the upstream library and file format project.
• MPI and MPI-IO provide the parallel programming and I/O substrate for Parallel HDF5.
• h5py and HDF5.jl are examples of higher-level ecosystems that distinguish serial and MPI-enabled HDF5 builds.
• netCDF is a neighboring scientific-data ecosystem commonly tested with HDF5 by upstream integration workflows.

来源

• https://formulae.brew.sh/formula/hdf5-mpi
• https://github.com/HDFGroup/hdf5
• https://github.com/HDFGroup/hdf5/blob/develop/docs/README_HPC.md
• https://support.hdfgroup.org/documentation/hdf5/latest/_intro_par_h_d_f5.html
• https://www.hdfgroup.org/2015/03/04/hdf-group/
• https://www.hdfgroup.org/about-us/
• https://www.hdfgroup.org/solutions/hdf5/