[MITgcm-support] MPI run: unable to open a GM port

Sergey Vinogradov svinogra at aer.com
Wed Mar 22 10:35:34 EST 2006


Ed:
Thank you, I just wanted to make sure it is solely a cluster thing.

Another question: does anyone have an opt file for linux_ppc64_xlf+mpi ?

Thanks
Sergey

Ed Hill wrote:

>On Wed, 2006-03-22 at 09:31 -0500, Sergey Vinogradov wrote:
>  
>
>>Hi:
>>
>>I try to compile/run MITgcm for 4 processors on linux ia32 cluster with 
>>mpif77/mpicc.
>>A compilation doesn't give me any problems, but execution fails with a 
>>following error:
>>
>>Error: Unable to open a GM port ! (4 times)
>>
>>First I thought it is a queueing problem, but I get the same result even 
>>if I start it interactively.
>>Did anyone experience anything similar to that? I'd appreciate any 
>>help/hints/hypotheses.
>>    
>>
>
>Hi Sergey,
>
>The GM device is a driver for Myrinet, a rather common high-speed and
>low-latency connection technology.  So, it looks like you have Myrinet
>problems and these things are by their nature very specific to your
>cluster.  So please talk to your cluster admins and ask them to help you
>sort it out.
>
>Ed
>
>ps - There are plenty of MPI versions of "Hello world" that 
>     you can use to debug these sorts of things.  And there are 
>     programs that can test the MPI connection speed and latency.  
>     It may be a good idea to check with simple MPI tests before 
>     trying to run MITgcm.
>
>  
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20060322/6150576c/attachment.htm>


More information about the MITgcm-support mailing list