[gmx-users] g_tune_pme error in blue gene

[Mark Abraham]

On 1/04/2012 9:13 AM, Mark Abraham wrote:
> On 1/04/2012 4:50 AM, Albert wrote:
>> Hello:
>>
>>   I am trying to run g_tune_pme in blue gene with following script:
>>
>> # @ job_name = bm
>> # @ class = kdm-large
>> # @ account_no = G07-13
>> # @ error = gromacs.info
>> # @ output = gromacs.out
>> # @ environment = COPY_ALL
>> # @ wall_clock_limit = 160:00:00
>> # @ notification = error
>> # @ job_type = bluegene
>> # @ bg_size = 64
>> # @ queue
>> mpirun -exe /opt/gromacs/4.5.5/bin/g_tune_pme -args "-v -s md.tpr -o 
>> bm.trr -cpo bm.cpt -g bm.log -launch" -mode VN -np 256
>>
>> but I've got the following messages as soon as I submit jobs and it 
>> terminate soon:
>>
>> ---gromacs.info--------------
>> <Mar 31 20:45:57.677742> BE_MPI (ERROR): Job execution failed
>> <Mar 31 20:45:57.677803> BE_MPI (ERROR): Job 10969 is in state ERROR 
>> ('E')
>> <Mar 31 20:44:58.476985> FE_MPI (ERROR): Job execution failed (error 
>> code - 50)
>> <Mar 31 20:44:58.477065> FE_MPI (ERROR):  - Job execution failed - 
>> job switched to an error state
>> <Mar 31 20:45:57.714358> BE_MPI (ERROR): The error message in the job 
>> record is as follows:
>> <Mar 31 20:45:57.714376> BE_MPI (ERROR):   "Load failed on 
>> 192.168.101.49: Executable file is not a 32-bit ELF file"
>
> This means the executable is unsuitable for the hardware to run. Front 
> and back ends of BlueGene are different hardware, of course.
>
>> <Mar 31 20:44:58.691897> FE_MPI (ERROR): Failure list:
>> <Mar 31 20:44:58.691923> FE_MPI (ERROR):   - 1. Job execution failed 
>> - job switched to an error state (failure #50)
>
> g_tune_pme relies on being able to spawn mpirun processes and measure 
> their performance. Back-end BlueGene/L jobs cannot spawn new 
> processes, and I'm skeptical that BlueGene/P would be able to do this 
> either (but P is less restrictive). So you will need to run g_tune_pme 
> compiled for the front end in your job script, and consult g_tune_pme 
> -h for clues on how to set up your job script so that g_tune_pme can 
> correctly call mpirun to invoke mdrun_mpi on the back end.
>

Posting your successful job script would be a welcome contribution for 
those in the community who will face this problem in the future.

Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20120401/ae933c8a/attachment.html>