[gmx-users] grompp -np option
nmoore at physics.umn.edu
Fri Jun 10 18:08:52 CEST 2005
> As the communication uses a ring structure, the number of MPI calls scales
> with the number of processors. This means that the scaling will degrade
> on a large numbers of processors. Summations are also done over the ring.
> I think that (when you don't use PME) for the 'real work' there should be
> number of atoms, but in practise the communication will become the
> at some point. This means the optimum number will grow as the number of
> processors increases.
Can you expond on your ring structure? Does this mean that if a message
is sent from rank=1 to rank=3 the message must go through rank 2 (ie
mpi_send/recv(1->2) then mpi_send/recv(2->3)? I was under the impression
that efficient MPI routing is generally regulated to the vendor MPI
More information about the gromacs.org_gmx-users