[gmx-users] Extending Simulation

[Mark Abraham]

On 10/04/2011 6:04 PM, mohsen ramezanpour wrote:
> Dear Dr.Mark
>
> On Sun, Apr 10, 2011 at 12:20 PM, Mark Abraham 
> <Mark.Abraham at anu.edu.au <mailto:Mark.Abraham at anu.edu.au>> wrote:
>
>     On 10/04/2011 5:40 PM, mohsen ramezanpour wrote:
>
>         Dear All
>         I used the following commands accoring to Extending Simulation
>         in gromacs/Documentation/how-tos/Extending Simulation
>         to extend my simulation.
>         I entered:
>
>         tpbconv  -s  npt-1.tpr    -extend  100  -o npt-1-extend.tpr
>
>         nohup mpirun -np 4 mdrun -s  npt-1-extend.tpr  -cpi  npt-1.cpt
>
>         Ii run this command on a node with 4 cpu.
>         the result was these files:
>         confout.gro , state.cpt   ,md.log  , traj.trr  ,state_prev.cpt
>          ,ener.edr ,
>         #confout.gro.1# ,#confout.gro.2# ,#confout.gro.3# ,#
>         ener.edr.1# ,#ener.edr.2# ,#md.log.1 #,#traj.trr.1#
>         ,#traj.trr.2# ,
>
>         I extended 4 files on 4 distinct nodes!
>
>
>     That's not the behaviour you were looking for. You've run the same
>     .tpr four times, once on each CPU. This is because mdrun was not
>     compiled with MPI. See the installation instructions.
>
> I think I explained bad.let me explain it more:
> Suppose I entered the above commands on just one node(with 4 cpu) for 
> npt-1.tpr (just one time I entered these commands)

You seem to have used mpirun to run four different mdrun processes, each 
using the same .tpr, one on each cpu. You should be trying to run one 
mdrun_mpi using that .tpr, but all four cpus working on the same run.

Look at the top few lines of your .log. That will tell you how many 
nodes mdrun thought it was running on. I think you will see "NNODES=1". 
If so, don't use this mdrun with mpirun.

> besides if you were right it would result the same outputs when I run 
> it on different nodes with the same commands,please see below!

Not strictly true. See 
http://www.gromacs.org/Documentation/Terminology/Reproducibility

>
>         it is excellent,because the number of outputs was different,
>         for example one node produced 3 md.log and 4 confout.gro !
>
>         The main question is:
>         which one of the resulted outputs are the main result? on wich
>         do  I must analyse?
>
>
>     They should all be equivalent, but not necessarily binary identical.
>
> Actually I checked all of them,they are different,with different 
> averages of quantities,
> and all of them show a final text that means the program finished 
> succesfully!

The averages being the same or different merely reflect whether your run 
was reproducible. Assuming there was no output files in your working 
directory to start with, you've run the same irreproducible run four 
times, because you're using a non-MPI mdrun.

Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20110410/26a45502/attachment.html>