Hi Hai,
There is no difference in the way pmemd.MPI works on a single node compared to multiple nodes. The executable doesn't even look at if the MPI tasks are running on the same node or not. Thus this to me looks more like a problem with your MPI installation rather than AMBER per se. Have you tried running the test cases that come with the MPI installation you are using to see if they work properly across multiple nodes? I've seen these problems before when people use an MPI installation that is incompatible with their interconnect. E.g. using a MPI configured for infiniband when you have ethernet.
All the best
Ross
> On Oct 3, 2018, at 22:57, Hai Long <spaceshiptoo.gmail.com> wrote:
>
> Hi David and Ross,
>
> Thank you for your kindly replies and sorry I didn't update the status for
> this problem. I had made it working by compiling using parallel MKL. But it
> only works for one node. When running on multiple nodes it results in
> segmentation fault. If I choose not to use MKL, I need to recompile FFTW3
> by using F77=mpiifort, and again, it only works for one node. Any ideas on
> this problem? Thanks!
>
> Best
>
> On Thu, May 24, 2018 at 4:08 PM Ross Walker <ross.rosswalker.co.uk> wrote:
>
>> Hi Hai,
>>
>> I've seen this type of behavior when the mpirun command being used to run
>> pmemd.MPI is different from the mpi installation that was used to build the
>> executable.
>>
>> I would check that which mpif90 matches which mpirun. I'd then try
>> rebuilding from scratch with your current environment and see if the
>> problem persists.
>>
>> All the best
>> Ross
>>
>>> On May 24, 2018, at 5:44 PM, Hai Long <spaceshiptoo.gmail.com> wrote:
>>>
>>> Hi David,
>>>
>>> It won't work with any pmemd tests in the test suite. The following is
>> from
>>> the 4096wat test output. The same thing happens. It seems to me that
>>> pmemd.MPI somehow messed up electrostatic interactions. The vdwaals
>>> (6028.9517) is identical with the sander output though.
>>>
>>>
>>> NSTEP = 1 TIME(PS) = 1.001 TEMP(K) = NaN PRESS =
>>> 0.0
>>>
>>> Etot = NaN EKtot = NaN EPtot =
>>> NaN
>>>
>>> BOND = 0.0000 ANGLE = 0.0000 DIHED =
>>> 0.0000
>>>
>>> 1-4 NB = 0.0000 1-4 EEL = 0.0000 VDWAALS =
>>> 6028.9517
>>>
>>> EELEC = NaN EHBOND = 0.0000 RESTRAINT =
>>> 0.0000
>>>
>>> Ewald error estimate: NaN
>>>
>>>
>>> On Thu, May 24, 2018 at 5:38 AM, David A Case <david.case.rutgers.edu>
>>> wrote:
>>>
>>>> On Wed, May 23, 2018, Hai Long wrote:
>>>>>
>>>>> We recently upgraded from Amber14 to Amber18. The compiler, MPI, and
>> MKL
>>>>> that I used for Amber 18 are all intel 2017.0.5. The sander,
>> sander.MPI,
>>>>> and serial pmemd are all working properly. However, pmemd.MPI produces
>>>>> errors. The following output is from a simulation of a 5X5X5nm
>>>> equilibrated
>>>>> water box, showing something wrong with EELEC and VIRIAL:
>>>>
>>>> Thanks for the report. We would need to see the actual inputs, plus the
>>>> command-line options you chose, in order to try to track down the
>>>> problem. It looks like there is something wrong with your input
>>>> coordinates.
>>>>
>>>> (I'm assuming that you have run the pmemd test suite in parallel, and
>>>> that things look OK there. If you haven't done this, please do so,
>>>> since that can remove the possibility of a bad input file.)
>>>>
>>>> ....dac
>>>>
>>>>
>>>> _______________________________________________
>>>> AMBER mailing list
>>>> AMBER.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Oct 04 2018 - 06:00:03 PDT
