MVAPICH Frequently Asked Questions

  1. General
    1. What is MVAPICH?
    2. Why the name "MVAPICH"?
    3. Where can I get MVAPICH?
    4. Whom do I contact for support?
    5. Is there any more information on MVAPICH?
    6. Is there a user guide available for MVAPICH?
    7. Which paper should I cite for the MVAPICH project?
  2. Technical Questions
    1. What to do in case of problems?
    2. How can I build MVAPICH?
    3. MVAPICH says I need a certail version of autotools, how do I install that?
    4. Where can I get enhancements and bug fixes for MVAPICH?
    5. How do I submit a patch to the MVAPICH Group?
    6. I am getting linking errors when building MVAPICH2-GDR (-lcudart not found)
    7. The previous solution did not fix the linking errors

A. General

  1. What is MVAPICH?

    MVAPICH is a high performance implementation of MPI-1 over InfiniBand and RDMAoE based on MPICH1. MVAPICH2 is a high performance MPI-2 implementation for InfiniBand, Omni-Path, Ethernet/iWARP, and RoCE based on MPICH2. High performance, scalable and fault-tolerance support are provided in these packages. These packages provide multiple underlying support (such as OpenFabrics/Gen2, OpenFabrics/Gen2-Hybrid, OpenFabrics-RDMAoE, QLogic-InfiniPath, and the standard TCP/IP) for portability across multiple networks.

  2. Why the name "MVAPICH"?

    The name `MVAPICH' was chosen to reflect the fact that the original implementation was an MPI implementation over the InfiniBand VAPI interface based on the MPICH implementation. MVAPICH is pronounced as ``em-vah-pich''.

  3. Where can I get MVAPICH?

    MVAPICH source is available in various forms. Depending upon your need, you may choose to download the release tarball (stable) or if you want to be on the bleeding-edge of our developments, you may choose to check out from our anonymous SVN repository.

    Both software packages are available from our Download tab.

  4. Whom do I contact for support?

    Bug reports and suggestions can be posted to mvapich-discuss mailing list.

  5. Is there any more information available on MVAPICH?

    Yes. Please refer to the MPI over InfiniBand, Omni-Path, Ethernet/iWARP, and RoCE Project web site. This web page is frequently updated with latest releases, publications, benchmarks, installation guides and performance tuning guidelines.

    If you would like to get information about future updates, releases, publications, etc. related to the MPI over InfiniBand, Omni-Path, Ethernet/iWARP, and RoCE project, you may subscribe to the mvapich mailing list . This is an announcement-only list. All release announcements related to MVAPICH and MVAPICH2 are posted to this list. There is also a mvapich-discuss mailing list associated with this project. You may subscribe to this list to participate in the discussions related to MVAPICH and MVAPICH2.

  6. Is there a user guide available for MVAPICH?

    Yes. Please look under the `Support' tab.

  7. Which paper should I cite for the MVAPICH project?

    The following paper can be cited.

    D. K. Panda, H. Subramoni, C. Chu, and M. Bayatpour, The MVAPICH project: Transforming Research into High-Performance MPI library for HPC Community, Journal of Computational Science (JOCS), Special Issue on Translational Computer Science, Oct 2020.

    An electronic copy of the paper can be obtained from here

    The Bibtex entry for the paper is as follows:

    @article{PANDA2021101208,
    title = {The MVAPICH project: Transforming research into high-performance MPI library for HPC community},
    journal = {Journal of Computational Science},
    volume = {52},
    pages = {101208},
    year = {2021},
    note = {Case Studies in Translational Computer Science},
    issn = {1877-7503},
    doi = {https://doi.org/10.1016/j.jocs.2020.101208},
    url = {https://www.sciencedirect.com/science/article/pii/S1877750320305093},
    author = {Dhabaleswar Kumar Panda and Hari Subramoni and Ching-Hsiang Chu and Mohammadreza Bayatpour},
    keywords = {HPC, MPI, MVAPICH, RDMA},
    abstract = {High-Performance Computing (HPC) research, from hardware and software to the end applications, provides remarkable computing power to help scientists solve complex problems in science, engineering, or even daily business. Over the last decades, Message Passing Interface (MPI) libraries have been powering numerous scientific applications, such as weather forecasting, earthquake simulations, physic, and chemical simulations, to conduct ever-large-scale experiments. However, it is always challenging to practice HPC research. In this article, we use the well-established MVAPICH project to discuss how researchers can transform the HPC research into a high-performance MPI library and translate it to the HPC community.} }

B. Technical Questions

  1. What to do in case of problems

    MVAPICH provide detailed and very easy to follow guides to help troubleshoot your installation. Please refer to the user guides under the `Support' tab.

    The troubleshooting sections in our user guides include solutions to many of the common problems encountered.

    If you still cannot figure out what went wrong, do not hesitate to post a note to the MVAPICH user community by sending an email (with attached logs) to the mvapich-discuss mailing list.

  2. How can I build MVAPICH?

    Please refer to the appropriate sections on building and installing MVAPICH in their respective user guides. The user guides are available under the `Support' tab.

  3. MVAPICH says I need a certail version of autotools, how do I install that?

    You can use the this script to build and install the necessary version of autotools.

    If you invoke it as ./build_autotools.sh $PWD, it will install everything under $PWD/Tools/install-dir. You can then add $PWD/Tools/install-dir/bin to $PATH and $PWD/Tools/install-dir/lib to LD_LIBRARY_PATH and run autogen.sh

  4. Where can I get enhancements and bug fixes for MVAPICH?

    Based on the feedback obtained from current users, we provide general improvements and bug fixes to the current version of MVAPICH. These fixes can be obtained directly from our anonymous SVN or as minor releases following a major release. Please use the `Download' tab on the top menu to access the source code for MVAPICH.

  5. How do I submit a patch to the MVAPICH Group?

    The MVAPICH Group encourages patches from outside developers. There are two main "types" of patches: small bugfixes and general improvements.

    To submit your patch, please send an email to mvapich-discuss mailing list with your patch.

  6. I am getting linking errors when building MVAPICH2-GDR (-lcudart not found)

    When building MVAPICH2 especially GDR and binaries that link to it, it is necessary to link against several cuda runtime libraries including libcudart. This on most systems this library is only accessible on GPU enabled nodes and requires an active allocation. Please make sure that a CUDA module is loaded and an allocation is active. Please try adding the containing directories of libcuda.so and libcudart.so object to your LD_LIBRARY_PATH

  7. The previous solution did not fix the linking errors

    On some Linux distributions, most commonly SUSE, the previous step is insufficient. If adding the containing directories of libcudart.so and libcuda.so to your LD_LIBRARY_PATH did not solve the issue, please try adding them to the LIBRARY_PATH environment variable as well.