DEVELOPMENT SITE : FORUM and SUPPORT

In 2001 I extend the CIVx to the third dimension. To accelerate the computation I used the MPI librairies which permit to dispatch the whole 3D data on a cluster of PC. Acquisitions during an internal wave experiment permit to validate the code in real case.

The source code is available and you can download here the binaries and the data acquired during this experiment to test the technique on your own lab.

Instructions:

• Download the following files: volume1(60Mo) volume2(60Mo) parametres-definitions
• Run the executable under MPI environment. The calculus distribution runs rock stable on the Coriolis heterogeneous cluster .
• If you run the process with the files provided , you will obtain this 2 AVS files file1 file2 , if you plot them you can visualize figure 1
• 

Figure 2 is a vertical slice extracted from the volume on which you can appreciate the vertical resolution of the method.

Requirement & Performance:

The algorithm is compile for windows x86 platform using MPI libraries ( I suggest MPICH (executable) )

To process a 93*1024*1024 voxels volumes without swapping you need 586Mo of RAM for the first process and 11Mo for each additional process (minimum 1), as example to process the example data on a 35700 points grid using the following arrangement of computers : 4 * 1.4 Ghz + 1 * 1 Ghz + 7*550Mhz the global computation time is 4 minutes including data loading and output. The correlation boxes size was 13*13*7 with a search box of 31*21*15

Input :

The file format for the images/volume is PNG , if the volume is composed by nz images of size nx by ny , we save it as an image of size nx*nz in width and ny in height. If it's not crystal clear for you, please visualize on of the image file given in the example using the great SuperPNG plugins.

For the parameter selection, please download the parametres.conf file provide in the example

Output :

The results use the proprietary AVS .fld format; a netcdf format will be soon supported too.