This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| MVAPICH | |
|---|---|
| Name | MVAPICH |
| Developer | Ohio State University |
| Released | 2005 |
| Programming language | C, C++ |
| Operating system | Linux |
| Platform | x86_64, POWER, ARM |
| License | BSD-style |
MVAPICH
MVAPICH is a high-performance implementation of the Message Passing Interface designed for use with InfiniBand, RDMA over Converged Ethernet, and Omni-Path networks. It was developed at a university research center to accelerate distributed scientific computing on supercomputers and clusters, and it is integrated into production environments at national laboratories, universities, and industry research centers. The project builds on research in parallel computing, networking, and high-performance libraries to provide low-latency, high-throughput MPI for scalable applications.
MVAPICH targets scalable parallel applications and supports the MPI-1, MPI-2, and many MPI-3 features used by large-scale science codes. It emerged in the context of high-performance interconnect advances alongside efforts at institutions such as Ohio State University, Lawrence Livermore National Laboratory, Los Alamos National Laboratory, National Energy Research Scientific Computing Center, and Argonne National Laboratory. Early development responded to performance requirements from codes like LAMMPS, GROMACS, NAMD, Quantum ESPRESSO, and WARPX. The project interacts with standards bodies and consortia including Open MPI Project, Message Passing Interface Forum, and vendor ecosystems like Mellanox Technologies, Intel Corporation, and Cray Inc..
The MVAPICH stack comprises transport layers, collective algorithms, and runtime services that map MPI semantics onto RDMA primitives and network hardware. Core components include the point-to-point engine, collective module, process manager interfaces compatible with SLURM, PBS Professional, and Torque (software), and tuning utilities that leverage hardware features in products from NVIDIA, Broadcom, and AMD. The implementation integrates with system libraries such as libibverbs, rdma-core, and operating system interfaces used on distributions from Red Hat, SUSE, and Ubuntu (operating system). Collective algorithms implement optimized protocols for operations like broadcast and reduce, influenced by work from researchers at Lawrence Berkeley National Laboratory, Stanford University, and University of California, Berkeley.
MVAPICH emphasizes low-latency small-message performance and high-bandwidth large-message transfers through use of zero-copy RDMA, eager/rendezvous protocols, and hardware offloads present in products like InfiniBand Trade Association hardware and Intel Omni-Path Architecture. Benchmarks published by collaborators compare performance on systems such as Frontera (supercomputer), Summit (supercomputer), Perlmutter (supercomputer), and Blue Waters (supercomputer), against implementations from Open MPI Project and vendor MPI releases. Scalability studies evaluate strong-scaling and weak-scaling for applications from the fields represented by Oak Ridge National Laboratory, Argonne National Laboratory, and projects funded by agencies such as National Science Foundation and Department of Energy.
Multiple branches of the project target different hardware ecosystems and feature sets: releases optimized for InfiniBand Trade Association-based fabrics, variants supporting RDMA over Converged Ethernet (RoCE), and editions tuned for Intel Corporation and Hewlett Packard Enterprise platforms. Version histories correspond to evolving MPI standards and hardware features introduced by companies like Mellanox Technologies (now part of NVIDIA), Cray Inc. (now part of Hewlett Packard Enterprise), and Microsoft-partnered solutions. The codebase provides extensions for hybrid programming models blending MPI with threading libraries and accelerator interfaces from CUDA, OpenCL, and SYCL initiatives supported by Khronos Group.
MVAPICH is used in computational domains requiring tightly coupled parallelism, including climate modeling groups at National Center for Atmospheric Research, computational chemistry groups at Pacific Northwest National Laboratory, astrophysics teams at Harvard–Smithsonian Center for Astrophysics, and engineering workflows at General Electric. Representative scientific applications include CESM (Community Earth System Model), WRF (Weather Research and Forecasting Model), LAMMPS, and VASP, which rely on efficient MPI collectives and point-to-point operations. Industrial users in sectors represented by Boeing, Siemens, and Schlumberger employ the stack for design optimization, simulation, and data analysis on platforms provided by cloud vendors collaborating with Amazon Web Services and Microsoft Azure for HPC offerings.
The project is driven by an academic research group collaborating with national labs, vendors, and open-source communities. Contributors include faculty and students from Ohio State University and engineers from companies such as Mellanox Technologies, Intel Corporation, NVIDIA, and Hewlett Packard Enterprise. The development workflow interoperates with tools and services from GitHub, continuous-integration systems used by Jenkins (software), and packaging ecosystems maintained by distributions like Red Hat and Debian. Outreach occurs at conferences including SC (conference), ISC High Performance, and workshops organized by the Message Passing Interface Forum and related research forums.
Reliability features include fault-tolerant collective progress strategies, integration with job schedulers like SLURM, and testing on clusters maintained by facilities such as National Energy Research Scientific Computing Center. Security considerations address vulnerabilities in system libraries such as libibverbs and driver stacks from vendors like Broadcom and Intel Corporation, and the project follows responsible-disclosure practices with industry partners. High-availability deployments have been validated in environments managed by Los Alamos National Laboratory and Lawrence Livermore National Laboratory to meet requirements for robust, repeatable scientific workflows.
Category:Message Passing Interface Category:High-performance computing