diff --git a/docs/source/introduction.rst b/docs/source/introduction.rst index 4d8b1625b..8ed4be097 100644 --- a/docs/source/introduction.rst +++ b/docs/source/introduction.rst @@ -42,12 +42,10 @@ Suren Byna, Bin Dong, Houjun Tang, Quincey Koziol, Jingqing Mu, Jerome Soumagne, **1.2.** Installation --------------------- -PDC offers the following methods for installing core dependencies: - -1. :ref:`Spack ` -2. :ref:`PDC Source ` - -PDC offers the following installation targets: +PDC offers several installation targets, +including C and Python APIs, as well as support +for use as an HDF5 VOL connector. These options are listed below, +and the accompanying figure illustrates the installation path for each: 1. `C API` 2. :ref:`Python API (PDCpy) ` @@ -57,7 +55,13 @@ PDC offers the following installation targets: :alt: PDC Installation Diagram :align: center - Installation workflow to install the client targets offered by PDC. +Installation workflow to install the client targets offered by PDC. + +As seen in the figure above, + PDC offers the following methods for installing core dependencies: + +1. :ref:`Spack ` +2. :ref:`PDC Source ` .. note:: @@ -102,12 +106,35 @@ PDC and its dependencies can be installed with spack: PDC Source ~~~~~~~~~~ -We recommend using GCC version 7 or later. Intel and Cray compilers also work. - When building PDC from source, either MPICH or OpenMPI can be used as the MPI library, if your system doesn't have one installed, follow `MPICH Installers Guide `_ or `Installing Open MPI `_ +We list the required compilers, MPI implementations, supporting libraries, +and optional dependencies along with their recommended versions below: + +* **Compilers** (one of the following): + + - GCC version 7 or later. + - Intel compiler (version stable) + - Cray compiler (version stable) + +* **MPI implementations**: + + - OpenMPI (version stable) + - MPICH (version stable) + - Intel MPI or Cray MPI (version stable) + +* **Supporting libraries**: + + - libfabric (version 1.18.0) + - Mercury (version 2.2.0) + +* **Optional dependencies**: + + - Python (required if using the PDC Python API) — version 3.x + - HDF5 (required if using the PDC HDF5 API) — version 1.12.1 + We provide detailed instructions for installing libfabric, Mercury, and the PDC library below. .. attention:: @@ -596,7 +623,6 @@ This section offers the following examples for different PDC target installation .. note:: All examples omit detailed error checking for clarity. In practice, always check the return values of PDC API calls. - See the section TODO_FIX_REFERENCE for more information on detecting and handling PDC errors. .. _c-api-first-program: diff --git a/docs/source/roadmap.rst b/docs/source/roadmap.rst index 3c5aff6a0..245d909ca 100644 --- a/docs/source/roadmap.rst +++ b/docs/source/roadmap.rst @@ -22,32 +22,26 @@ vision for PDC. - Expand documentation and example workflows for common HPC use cases. - **Interoperability** - - Strengthen integration with HDF5, ADIOS, and other I/O frameworks. - - Improve support for various network transports through Mercury (``ofi``, ``tcp``, ``verbs``, etc.). + - Strengthen integration with HDF5 and other I/O frameworks. **7.3.** Short-Term Goals (Next 6-12 Months) -------------------------------------------- -- Implement enhanced **data caching and eviction policies** to reduce I/O latency. -- Improve **burst buffer management** for hybrid memory/storage architectures. -- Extend **metadata indexing** and search capabilities. -- Add more comprehensive **unit and integration tests** in CI pipelines. -- Expand **PDC client APIs** in C and Python with better documentation and examples. +- Implement enhanced client-side data caching and eviction + policies to automatically exchange data from other MPI ranks and from multiple nodes. **7.4.** Medium-Term Goals (1-2 Years) -------------------------------------- -- Introduce **multi-tier data management** (memory, SSD, disk, object storage). -- Develop **asynchronous data movement and prefetching** mechanisms. -- Enhance **seurity and authentication** (e.g., Cray DRC, token-based access). -- Support **federated PDC deployments** across distributed sites. +- Integrate data movement between multiple HPC systems and between HPC and Cloud object storage systems. +- Compound data type support for regions. **7.5.** Long-Term Vision (Beyond 2 Years) ------------------------------------------ -- Enable **self-optimizing data placement** based on access patterns and system telemetry. -- Integrate with **workflow and data provenance frameworks** for end-to-end data lifecycle management. -- Expand **AI-driven data management features**, such as automated cache tuning and prediction-based prefetching. +- Integrate with workflow and data provenance frameworks for end-to-end data lifecycle management. +- Expand AI-driven data management features, such as automated cache tuning and prediction-based prefetching. - Achieve production-grade stability and adoption across DOE and large HPC facilities. +- Integrate PDC in DOE applications and workflows. *Last updated: November 2025*