Parallel Programming on SGI Systems

The objective of parallel processing is to reduce the wall clock time associated with the execution of program code by distributing the code across multiple processors.  Note, however, that the total CPU time for the program will increase as a result of the additional communication between processors.  Thus, typically, only long running programs which are to be run multiple times are written or modified to be executed as parallel applications.

SGI information about parallel computing is available in its Performance Tuning Guide.

Four approaches are commonly utilized to enable parallel processing and all of them are available on the SGI systems:

• Using the auto-parallel option with the Intel Fortran and C Compilers
• Using parallel subroutine libraries
• Using OpenMP
• Using MPI (Message Passing Interface)

Note:  The first two of the above approaches are implemeted via OpenMP on SGI systems.

You can use the parallel capabilities of the SGI systems by compiling your programs using Intel "ifort" and "icc" compilers.  Message passing parallel codes may also be compiled using the "gcc" and "g++" compilers.  You can obtain information about these compilers and display a listing of the available compiler options by entering:

    ifort -help
    icc   -help
    man   gcc

Here is a listing of some web sites which provide information about parallel computing:

• Google Parallel Computing Categories & Web Pages
• LLNL Introduction to Parallel Computing
• Maui HPC Center -- Introduction to Parallel Computing
• Internet Parallel Computing Archive
• Ohio Supercomputer Center
• Cornell Education Information
• SDSC References   
• National Energy Research Scientific Computing Center
• Linux Parallel How To