Parallel Solving with HPC Cluster?
 

Hello,
ich just solved the tutorial "breaking of a dam" with the comand "mpirun -np 4 interFoam -parallel > log &" this worked so far fine, but it wasn't faster. I have installed a virtual machine (Oracle VM VirtualBox - as in OpenFoam Homepage documentation described) on a Windows 2008R2 Server - with the HPC-Pack installed on is. A HPC Cluster is installed on it. So my question is: How do I connect from the virtual machine to the HPC cluster? I tried with the command "mpirun --hostfile hostfile.txt -np 2 interFoam -parallel > log &" in the hostfile.txt there is the servername. The resulting message is:
----------------------------------------------------------------------
[1] 5815
Path$ ssh: connect to host SRV-AP04.emg.trasys port 22: Connection timed out
A daemon (pid 5816) died unexpectedly with status 255 while attempting
to launch so we are aborting.
There may be more information reported by the environment (see above).
This may be because the daemon was unable to find all the needed shared
libraries on the remote node. You may set your LD_LIBRARY_PATH to have the
location of the shared libraries on the remote nodes and this will
automatically be forwarded to the remote nodes.
mpirun noticed that the job aborted, but has no info as to the process
that caused that situation.
mpirun: clean termination accomplished
----------------------------------------------------------------------
So at the moment I do not know how it is possible to submit a mpi job from LINUX to the Windows-HPC Cluster. I would be pleased for a solution of this problem

Best regards
BAH

Can't comment on your setup (I have an opinion on HPC and Windows in the same sentence. But that is beside th point)

Don't expect any significant speedups on the damBreak-case. It is to small:
http://en.wikipedia.org/wiki/Amdahl%27s_law

Thanks for your reply!
The point is - whether there is a speed up anyhow. The machine should have 12 Cores 6 at each processor. When running in parallel no speedup at all occured... So in the moment it is a test...At least I would wish to have any speedup. But so far the clock times were 1x337(s) 2x200(s) 3x321(s) 8x2594(s) 16x4659(s) the latest without mpi run was 138(s).
Do you think this is only due to the small size of the problem?

I am new to linux - but I think it should be possible to connect to the server... but how?

The damBreak-case has appro 2300 cells. That means for the 16 CPUs you have approx 140 cells per CPU (recommended is 50000 or more). That means you're only testing the speed of the interconnect

thanks for your hints :)

Greetings to both Bernhard's :)

@bah: Check the report on this thread/post: http://www.cfd-online.com/Forums/ope...tml#post283728 - post#2
The report details a relatively good test case and how it panned out when using multiple machines, namely on chapters 4 and 5.

Best regards,
Bruno

Hello Bruno :-) hello future all :-)*
 

:)
thank you for your post - I have seen you are an expert in Windows Compiling of OpenFOAM :) So the news: on monday I found with google "open foam HPC" the symscape homepage http://www.symscape.com/node/930 I downloaded the test version and started it on the cluster. On the same day i could realize a complete CFD example running on the M$-HPC!!! (thanks to symscape:)) - I am still fascinated.

Just now I read also your post about the history in http://www.cfd-online.com/Forums/ope...hpc-cloud.html
it was quite impressive thanks.

I also found for the scaling with HPC in general in openfoam with the example I also used (damBreak) a comparison for parallelcomputing velocity http://ebookbrowse.com/openfoam-at-scale-pdf-d77932137
and http://www.hpcadvisorycouncil.com/pd..._Practices.pdf

So far my experience of the current state: Symscape seems to be quite useful concerning the physical boundary conditions and automatic meshing.

I am still missing two features: 1.) How I can implement "rotating frames of reference" 2.) and multiple different regions and of course many other things like mesh refining... do you have something adequate in your compilations? Especially I have seen there is an openFOAM fork for turbomachinery but that ist not in the official OpenFOAM distribution nor at symscape. - up to now I don't really know how to use these things.

Concerning physics I may comment - as you mentioned in your profile you are not so familiar with: As far as I can see OpenFOAM is from the physics point of view geniously. The formulation with quasi differential equation language written in physics language and C++ together is making it really powerful now and in future. Both OpenFOAM and C++ are mirrors of the nature representing maximum information syntropy i.e. are of highest scientific value. What ist missing? The same that is missing in physics. The link between particle and continuum physics as occurs in the boundary regions in the nanoworld. I feel that this is my topic in the next decade...
All Love
Bernhard







*Einstein said once a similar in his presentation: "Sehr geehrte An- und Abwesende!" ...dear present and unpresent!

@bah: Hi Bernhard,

You can contact the good people at Symscape and ask them or see directly in Caedium if there is anything in there ready to use.

Like I've said before here:
That version I talk about is the -Extend variant (or fork) of OpenFOAM.

Nonetheless, the official OpenFOAM version does have something for pseudo-moving mesh zones... I believe it's called MRF...

Best regards,
Bruno

Thanks Bruno!