CKO


Parallel Implementation of MICOM

Version: 2.7

This version of PMICOM version is based on the physical model implemented in MICOM version 2.6. This means that this version of PMICOM has the same limitations as MICOM 2.6. Most important, it is not suited for areas of the oceans near the north or south poles.

Download
You can download a 1.7 Mb gzipped tar file containing the source code of PMICOM and a set of data files from here: pmicom-2.7.tar.gz. This version has been tested on a parallel SGI64 computer.

Note that this version requires that the Message Passing Toolkit is installed on the computer. This toolkit is availabe from SGI.

Installation
To install the package you should create a directory PMICOM and copy the downloaded file into it. After changing into the PMICOM directory you can unpack the tar file with the command 

gunzip -c pmicom-2.7.tar.gz | tar xvf -
Next, you should compile the partit program that computes how the work is divided amongst a number of processors. To do this, you should change directory to source and type 
make partit
to compile this program. Copy the program to the PMICOM/bin directory with the command 
cp partit ../bin
and change back to the PMICOM directory.

Configuration and Compilation
Information on configuring the PMICOM program is avialable in the README file in the PMICOM/source directory. However, it is much easier to configure and compile the code using the partition program that is located in the PMICOM/bin directory. This program should be executed in the directory containing the input data files, i.e., the PMICOM/data directory. You should provide the width and height of the processors grid as arguments to the program. For example, to configure and compile the PMICOM executable to run on two by three processor grid using the input data contained in the distribution, you enter the following commands 

cd data
../bin/partition 2 3
Upon completion, the executable micom-2x3 can be found in the directory PMICOM/bin.

Running
You can perform a demo run of PMICOM by changing to the PMICOM/data directory and typing the command 

mpirun -np 6 ../bin/micom-2x3 < limits > output
We note that the argument value supplied to the -np option must match the total numbers of processors allocated to the grid. The program requires about XX Mb of memory to run in this configuration. It completes in approximately 25 min on a Silicon Graphics parallel computer provided that sufficient processors are available.

Performance Considerations
The actual performance improvement of the parallel implementation compared to the standard implementation of MICOM depends on many factors. To ensure a good performance, one should at least ensure that the number of processors allocated to the grid are really available while PMICOM is running. If this is not the case, a possibly severe loss in performance will occur. In fact, PMICOM might require several times more time to complete the simulation than the standard implementation MICOM.

Support
PMICOM is supported at the S1 level (see also the page on support levels). For more information about PMICOM, you can contact the author.


Camiel Severijns
Last modified: Thu Nov 4 14:37:03 CET

Return to the CKO MICOM page