Timing Charts (last update: November 2004)

    o  November 2004 (arps_mpi 5.1.4) MS Excel document

  These charts include results for Alpha EV6.8 (1000 MHz) and AMD Opteron processors.


  Tornado Simulations 2004

  OU Stills

    o  qvtor059
    o  qvtor089

  OU Animations

    o  25x20m_cf
    o  25x20m_cd2

  PSC Animations

    o  Directory Listing


  Scalability

  Notable observation regarding scalability.

    o  ES45
       HP

    o  XT3   
       Cray

    o  Opteron   
       HP


  Construction

  The following compiling systems and settings were used to build the test executables.

    o  ES45 (Compaq)

       F77 = f90 -v -version
       FFLAGS = -c -O4 -fast -convert big_endian -assume byterecl -assume accuracy_sensitive -math_library accurate

    o  XT3 and Opteron PGI (Portland Group)

       F77 = mpif90 -v
       FFLAGS = -c -O2 -Munroll=c:4,n:4 -Mvect=assoc,cachesize:262144,sse -byteswapio


  Miscellaneous

  The ARPS benchmark example defines a fixed problem of modest size. Therefore the slope of
  the curves shown by ARPS scalability charts would ideally be ?????.

  The following computational aspects are measured by the program. Curves reflecting total
  elapsed times and a number of 100% charts have been developed. 

    Initialization

    Computation
      Wind advection
      Scalar advection
      Coriolis force
      Buoyancy term
      Misc Large tstep
      Small time steps
      Radiation
      Soil model
      Surface physics
      Turbulence
      Comput. mixing
      Rayleigh damping
      TKE src terms
      Gridscale precp.
      Kuo cumulus
      Kain-Fritsch
      BMJ cumulus
      Warmrain microph
      Lin ice microph
      NEM ice microph
      Hydrometero fall
      Bound.conditions

    Communications
      Message passing

    Data output

    Miscellaneous

    Entire model
      Without Init/IO

  Testing to date has revealed that the least scalable components of the listed
  cluster system timings are Input/Output and Message Passing which is entirely
  predictable.


  ARPS Web Site

  The Center for Analysis and Prediction of Storms (CAPS) at the University of
  Oklahoma maintains the ARPS website:

    o  CAPS Home Page
    o  ARPS Home Page

  ARPS is the Advanced Regional Prediction System. The Advanced Regional Prediction
  System (ARPS) is a comprehensive regional to stormscale atmospheric modeling and
  prediction system. It includes a real-time data analysis and assimilation system,
  a forward prediction model, and a post-analysis package.