[gmx-users] Assistance needed running gromacs 4.6.3 on Blue Gene/P

Prentice Bisbal prentice.bisbal at rutgers.edu
Wed Aug 7 17:04:30 CEST 2013


On 08/07/2013 10:35 AM, Prentice Bisbal wrote:
> Bogdan,
>
> Prentice
>
> On 08/07/2013 05:51 AM, Bogdan Costescu wrote:
>> Hi Prentice,
>>
>> The subject and first phrase mention version 4.6.3, but all scripts
>> refer to 4.6.2. Are you sure you are not mixing versions ? Do you have
>> another 4.6.x version working already ?
>
> Your correct - I'm working with version 4.6.2. the 4.6.3 in the 
> subject line was a mistake.
>>
>>> -DCMAKE_C_FLAGS="-O3 -qarch=450d -qtune=450" \
>> Could you try with a lower optimization level ?
>
> -O3 should be safe, but yes, I can try it with a lower optimization.
>>
>>> <Aug 06 17:25:55.303940> BE_MPI (ERROR): "killed with signal 6"
>> What is signal 6 on this platform ?
>
> I couldn't find anything explicit in the documentation, The compute 
> node kernel (CNK) is proprietary, but is said to be very Linux-like, 
> so a signal 6 should be the same as on Linux, which would be an abort 
> signal.

Actually, after some googling, I found references indicating that signal 
6 means the executable is too big:

http://www.neuron.yale.edu/phpbb/viewtopic.php?f=31&t=2721
http://corsi.cineca.it/courses/hpcUserDay/20110305/bluegene.pdf‎


Personally, I'm skeptical that this correct, especially in this 
instance. On my BG/P, in VN mode, each process has 512 MB of RAM, and in 
SMP mode, each process has access to 2 GB of RAM. Further, the 
executable is only 17 MB. I have tried running the program in both VN 
and SMP modes with the same result. Also, the Gromacs clearly starts 
running, and then encounters an error. If the executable was too big to 
fit in RAM, it would fail immediately. The executable is statically 
linked, so there's no possibility of issues being caused by loading 
additional libraries dynamically.

>>
>> Cheers,
>> Bogdan
>




More information about the gromacs.org_gmx-users mailing list