[GE users] lam/mpi sge and msc nastran integration

Reuti reuti at staff.uni-marburg.de
Tue Oct 21 16:44:42 BST 2008


Hi Pedro,

Am 21.10.2008 um 16:27 schrieb Pedro G:

> Hello I am new in this list. I was asking this on another list and  
> Reuti told me to use this since it is the correct one.
>
> Also I have done what he said; I redirected LAM/MPI to /bin/true as  
> suggested

Welcome!

To be clear: the usual LAM/MPI startup and other LAM/MPI programs  
should still work in your setup. Only inside the jobscript where  
Nastran will start the daemons on its own it was supposed that  
lamboot points to a dummy program.

The problem might be: which version of LAM/MPI was used to compile  
Nastran? Only since 7.1.1 it's SGE aware. Maybe MSC used an older  
version.

Sometimes you can get this information by:

$ strings my_binary

which will print all sequences of at least 4 readable ASCII  
characters in the file.

> and now I get "It seems that there is no lamd running on the host "  
> message.
>
> I have been trying to get rid of it but I couldn't.
>
> I attach the files nastran uses. I think the more important file is  
> ref-1.T8677_13. Take a look to the end of the file where it  
> launches mpirun. I think the problem is there, but I couldn't solve  
> it.

Yep. They are setting LAM_MPI_SOCKET_SUFFIX/-SESSION_PREFIX. As LAM/ 
MPI is SGE aware, these already point to a specific location in  
$TMPDIR by default for other LAM/MPI runs when running inside SGE.  
You can try to comment these lines and also remove the additional  
LAM_MPI_SESSION_PREFIX from the mpirun line.

As you have the jobscript, you could also comment the line with  
"lamboot" instead of doing tricks with the PATH.

-- Reuti


>
> Thank you very much.
>
>
> Am 20.10.2008 um 14:17 schrieb Pedro G:
>
>
> On 10/20/08, Reuti <reuti at staff.uni-marburg.de> wrote: Hi,
>
> Am 20.10.2008 um 12:18 schrieb Pedro G:
>
> I would like to know if Msc Nastran supports openmpi.
>
> I have been searching in google about Nastran and Openmpi but I  
> couldn't find if it works or not.
>
> Now I'm using lam/mpi for nastran parallel jobs, but I have some  
> problems with lam, nastran and sge, so I'd like to upgrade to openmpi.
>
> do you have the source code of the application? If you have only  
> the binary compiled for LAM, then you can't do anything to change.  
> You checked the LAM/MPI Howto at the SGE website?
>
> No, I don't have the source code, the application seems to be able  
> to work with openmpi since it has an option openmpi=yes in the  
> command line, but I couldn't make it to work. Anyway, I think it is  
> not fully supported yet since there is nothing about that option in  
> the user manual
>
> About LAM/MPI I have already read the howto and did a tight  
> integration. The problem is that Msc Nastran in parallel jobs start  
> a new lam environment getting out of control of sge.
>
> I contacted with Nastran and they told that was a LAM/MPI or SGE  
> problem.
>
> No, IMO it's not. When they start a new LAM/MPI environment, they  
> are violating the granted slot allocation. What about the following  
> (although it would be more a discussion on the [GE suers] list:
>
> - Suppose you have a tight LAM/MPI integration for other MPI programs.
> - In your jobscript, change the $PATH, so that the found "lamboot"  
> points to e.g. /bin/true - i.e. doing nothing.
> - When they then call "mpirun C", they should get the already  
> started LAM daemons.
> - If mpiexec is not working, maybe the found mpiexec must also  
> point to a script to supply the proper "-np ..." values.
> - When they call "mpiexec" in one-shot-mode, it must also be mapped  
> to a script to execute just the program, but not to do a "lamboot".
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list