Parallel computation with FDS

[nick]

Hello everyone,

I am running a large simulation involving a number of meshes in linux
based OS. The simulation approximately takes 3 weeks to finish. So I
want to parallelize it with 2 computers, each consisting of quad
processors (that means 8 processors are available for my task). Since
I am new to FDS, I want to know few things from the user group:

1. How will the input file for parallel computation change from that
for serial one? What are the main features that I need to change in
the input file (e.g., mention no. of processor to be used etc.)? I
have been looking for an example input file for parallel version of
FDS but couldnt find one.

2. How can I assign different meshes to different processors? Do I
need to mention it in the Input file? Is it possible to assign 12
meshes to 8 processors?

3. As the user guide says, I need to install parallel version of FDS
in each of the processors. Could you suggest me how can I do it for
quad processor computers?

4. Regarding post processing, do I have to access each processor
everytime to gather results from a particular mesh? For example, how
can I use fds2ascii_linux to get temperature from a particular mesh
that is assigned to a particular processor?


Thanks in advance and I will be really grateful for your help.


best regards,
Nick

[nick]

sorry, I just found this post which furnishes answer to my question 3.
Thanks.

http://groups.google.com/group/fds-smv/browse_thread/thread/d070e2a6e907208/918a5cf07df005e2?lnk=gst&q=parallel+FDS#918a5cf07df005e2

[rmcdermo]

Nick,

Meshes are assigned processes in the order in which they are set in
the input file. So, if you have

&MESH XYZ... / --> process 0
&MESH XYZ... / --> process 1

For your case I would use two meshes and compile with OpenMP flags
(actually to get the most optimized code, I would run with SVN < 3895
because Kevin overhauled some of the velocity bcs and I don't think
this broke the OpenMP flags for the momentum equation). In this way,
you will utilize shared memory parallelism for the quad core machines
and MPI between meshes. A discussion of setting up multi-mesh
calculations is given in the Users Guide Sec. 3.1.2.

Cheers,
Randy

The way I would set up your case is to use OpenMP for the on your quad
cores and MPI for the different machines.

On May 5, 4:57 am, nick <shankar256...@yahoo.co.in> wrote:
> sorry, I just found this post which furnishes answer to my question 3.
> Thanks.
>

> http://groups.google.com/group/fds-smv/browse_thread/thread/d070e2a6e...