A Survey of MPI Implementations

William Saphir
Lawrence Berkeley National Laboratory
University of California
Berkeley, CA 94720

Abstract:

The Message Passing Interface (MPI) standard has enabled the creation of portable and efficient programs for distributed memory parallel computers. Since the first version of the standard was completed in 1994, a large number of MPI implementations have become available. These include several portable implementations as well as optimized implementations from every major parallel computer manufacturer. The ubiquity and high quality of MPI implementations has been a key to the success of MPI. This review describes and evaluates a number of these implementations.

 
• Introduction and History
• Relation of MPI to PVM and HPF
• Review Criteria
  • Compliance.
  • Unspecified Behavior.
  • Integration with the environment.
  • Performance.
  • Tools.
• MPICH
  • General comments on all devices
    • Goodies.
    • Building MPICH.
    • Debugging.
    • Problems.
  • The ch_p4 device
  • The ch_shmem and ch_lfshmem devices
  • The ch_nexus Device
  • MPP devices: ch_cenju3, ch_mpl, ch_nx, nx, t3d
• Other Freely Available Implementations
  • LAM -- Local Area Multicomputer
    • Usability.
    • Tools.
  • CHIMP -- Common High-level Interface to Message Passing
  • Other implementations
    • NT
    • Win32
    • Active Messages
    • Fast Messages
    • Multithreaded (MT) Device
    • TransMPI
• Vendor Implementations
  • IBM
    • Usability.
    • Tools.
  • HP
    • Usability
    • Tools
  • Sun
  • Digital
  • SGI
    • SGI/PVP
    • SGI/T3E
    • SGI/Irix
      • Usability
      • Tools
  • NEC SX-4
  • Others
    • Mercury Race
    • Hitachi
    • NEC Cenju-3
    • Alpha Data
    • Fujitsu
    • PVMPI
• Portability Issues
  • Buffering
  • Non-standard Send Modes
  • Standard I/O
  • Performance Considerations with User-Defined Datatypes
  • Thread-safe MPI
• MPI-2
• Conclusions
• Acknowledgements
• References
• About this document ...