Hi,
So CPPTRAJ has parallelization at 2 different levels: 1) across
trajectory/ensemble, which is done with MPI, and 2) at the level of
individual time-consuming actions, which is done with either OpenMP or
CUDA. Since all you are doing here is trajectory processing and no
actions, you will only potentially benefit from using an MPI-enabled
cpptraj variant (cpptraj.MPI etc). Note that in order to write NetCDF
trajectory files in parallel, cpptraj.MPI needs to have been compiled
with parallel NetCDF; you can check this by ensuring -DHAS_PNETCDF
shows up when you run 'cpptraj.MPI --defines'.
When processing trajectories in parallel, you should do some
benchmarking first, especially if you're running on a single node;
you'll be limited by the maximum I/O throughput on your system (i.e.
there comes a point at which adding more processes results in little
to no speedup because you're limited by how fast your system can read
from the disk or over NFS etc). I recommend taking a segment of the
trajectory that takes about 10 seconds to process in serial, then run
with 2 processes, 4 processes, etc and see what kind of speedup you
get.
Hopefully this makes things more clear. Let me know if you have any
more questions.
-Dan
On Tue, Oct 4, 2022 at 5:30 AM Giulia Sormani via AMBER
<amber.ambermd.org> wrote:
>
>
> Hello to everyone
> I have been running a long simulation, which is divided in four parts and
> I would like now to have a unique nc trajectory keeping all the frames.
> I tried to use the following input file for cpptraj:
> """""
> parm dfo_Zr_2Cl.prmtop
> trajin NVT_0_400.nc 1 last 1
> trajin NVT_200_400.nc 1 last 1
> trajin NVT_400_600.nc 1 last 1
> trajin NVT_600_1000.nc 1 last 1
>
> trajout traj_1microsec_ts4ps.nc
> """""
>
> But it takes forever to run the command. To what I read here
> https://amberhub.chpc.utah.edu/parallelization/ , things are be faster
> using the MPI version of cpptraj, however I using Amber on Marconi100
> where cpptraj has been parallelized with with OpenMP.
> I am not really familiar with the concepts of parallelization, can someone
> help me in how to concatenate traj in a fast way using the openMP version
> of cpptraj?
>
> Many thanks
> Giulia
>
>
>
> _______________________________________________
> 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 Tue Oct 04 2022 - 06:00:02 PDT
