[gharek]

Hi guys,

I just have a simple question. I have got a code written in Fortran in serial. The maximum matrix size that I can go to is about 2 mil (in 3D). I wish to run more due to the huge size of the problem.

I have been looking around for some MPI tutorials and I still do not know what to do. Are there any better tutorials particularly suited for CFD applications?

I am thinking along the following lines:
1. breaking up my domain to different processors
2. each processor will solve one domain (or mesh) and
3. every processor will communicate with each other to update variable values

From what i know, MPI will help me. But im not sure how or where to start.

Thanks in advance.

[ertan]

Check out the IBM red book. It might be useful for you.
http://www.redbooks.ibm.com/abstracts/sg245380.html

[DoHander]

A simple approach will be to directly parallelize your serial code using OpenMP (check the official site for some good tutorials). Also if you want to be able to use more then 2GB of RAM per application you will need to 64 bits operating system (Windows or Linux both will do as long as you have the 64 bits version and ... plenty of RAM).

Also, the Fortran compiler you use will make a huge difference: Intel Fortran fast, gfortran will do (and it is free) ...

Do

[valgrinda]

Hei,

try these tutorials and lectures notes form the University of Edinburgh. Their material is excellent for starting with MPI.

http://www.epcc.ed.ac.uk/library/doc...tion/training/

While OpenMP seems tempting in the beginning, MPI is definitely the way to go for high performance computing. In the long run, using MPI will definitely pay off.

Regards
Hans