MPI Forum

MPI Forum
Name	MPI Forum
Formation	1992
Type	Standards organization
Purpose	Message Passing Interface standardization
Headquarters	International
Region served	Worldwide

MPI Forum The MPI Forum is a community-driven standards group that develops the Message Passing Interface specification for high-performance computing. It originated from collaborations among researchers at Argonne National Laboratory, Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, Los Alamos National Laboratory, and industry partners such as IBM and HP. The Forum’s work influences software used in projects at National Energy Research Scientific Computing Center, NERSC, PRACE, and deployments on systems like Summit (supercomputer), Fugaku, and Sierra (supercomputer).

History

The group began during the early 1990s amid discussions at meetings of ACM SIGPLAN, IEEE, and workshops associated with International Conference on Parallel Processing. Early contributors included teams from Cray Research, Intel, DEC, Microsoft Research, and university groups at University of Tennessee, University of Illinois Urbana–Champaign, Stanford University, Massachusetts Institute of Technology, and University of Cambridge. Initial drafts were debated alongside standards efforts such as POSIX, OpenMP, PVM (Parallel Virtual Machine), and initiatives by X Consortium. Major milestones include the release of MPI-1, MPI-2, MPI-3, and subsequent technical specifications discussed at symposia like SC Conference and International Supercomputing Conference. The Forum’s evolution intersected with programs at Department of Energy (United States), European High Performance Computing Joint Undertaking, and funding from agencies such as National Science Foundation and DARPA.

Purpose and Activities

The Forum’s purpose is to produce portable, performant, and portable application programming interfaces for message passing on distributed-memory systems. Activities include drafting specifications, maintaining errata, coordinating conformance tests, and promoting adoption through presentations at venues like Usenix, IEEE International Conference on Cluster Computing, and EuroMPI. The Forum liaises with implementers associated with Open MPI, MPICH, IBM Spectrum MPI, Cray MPI, and research projects from Lawrence Berkeley National Laboratory and Sandia National Laboratories. It has relationships with consortia such as The Open Group, Linux Foundation, ECP (Exascale Computing Project), and collaborations with vendors like NVIDIA, AMD, Arm Ltd., and Broadcom.

Membership and Organization

Membership comprises representatives from national laboratories, universities, and corporations including Google, Amazon Web Services, Microsoft Azure, Intel Corporation, HPE, and academic groups from California Institute of Technology, Princeton University, ETH Zurich, University of Oxford, and Tsinghua University. Organizational roles are often filled by engineers who have worked on projects like MPICH, LAM/MPI, FLAME GPU, and middleware stacks for platforms such as Kubernetes, Slurm Workload Manager, and OpenStack. The Forum coordinates with standards bodies like ISO and IEC and interacts with scholarship venues including ACM, SIAM, and IEEE Computer Society.

Standardization Process

The Forum adopts a consensus-driven process with drafts, technical proposals, and ballots circulated among member organizations. Design discussions reference previous standards such as MPI-1, MPI-2, MPI-3, and extensions inspired by protocols like InfiniBand, RDMA over Converged Ethernet, and technologies from Mellanox Technologies. Proposals are vetted through interoperability tests at facilities like Oak Ridge Leadership Computing Facility and Argonne Leadership Computing Facility. The process has produced features for collective communication, one-sided communication, and remote memory operations used in codes like LAMMPS, GROMACS, NAMD, and Quantum ESPRESSO.

Implementations and Impact

Implementations of the specification appear in open-source and commercial products including Open MPI, MPICH, MVAPICH, Intel MPI, IBM Spectrum MPI, and vendor-tuned stacks for systems such as Tianhe-2, Blue Waters, and HPC Wales. The Forum’s standards enabled scalable science in projects like Human Genome Project-related computation, climate modeling at ECMWF, and simulations at CERN’s Large Hadron Collider. MPI has been cited in literature across institutions like Los Alamos National Laboratory, Brookhaven National Laboratory, and universities including Cornell University and University of California, Berkeley. It influences curricula at Carnegie Mellon University and University of Toronto and underpins commercial offerings from Oracle Corporation and cloud HPC services from Google Cloud Platform.

Meetings and Working Groups

Meetings are held at conferences such as SC Conference, EuroMPI, International Symposium on High-Performance Computer Architecture, and at host sites including Argonne National Laboratory, Oak Ridge National Laboratory, and industry campuses owned by IBM and Intel. Working groups focus on areas like point-to-point communication, collective operations, fault tolerance, and parallel I/O, and coordinate with projects including HDF5, NetCDF, MPI Forum Tools, and performance tools developed at National Institute of Standards and Technology. Specialized subgroups have addressed GPU-aware communication with CUDA and integration with runtimes like OpenMP and PGAS languages such as UPC.

Category:Computer standards Category:High performance computing Category:Computer networking