Kamlesh,
> I have a question about AMBER. When we submit a simulation using
> mpirun....
> it devides the simulation into some number of CPUs. Could you please
> tell
> me how this sander job (one simulation) is devided.
Your question is not overly clear here but I will try to answer what I think
you are asking. The number of cpus the code uses is determined by the
arguments to mpirun, typically '-np X' where x is some power of 2 number, 8,
16 etc. The mpirun command will fire up this many copies of sander.MPI based
on the contents of the MACHINEFILE defined by the mpirun command. This
varies based on the MPI installation but would typically be something like:
node1 cpus=2
node2 cpus=2
In this case if you asked for 8 mpi threads you would get 4 on node1 and 4
on node2. Probably these would be handed out as:
thread0 = node1
thread1 = node1
thread2 = node2
thread3 = node2
thread4 = node1
...
Although this is again highly dependent on the MPI implementation.
Sander.MPI in the simplest explanation will then divide up the work in terms
of atoms giving natom/nthread atoms to each thread in turn. They then
calculate a subset of the interactions and the forces are summed on each
step in order to integrate. Note this is different to the way PMEMD works,
which divides up space rather than atoms.
Note things are more complicated for things like REMD, TI multisander type
runs. Here the individual replicas get divided up amongst cpus and then the
work gets divided up based on the number of threads per group. E.g. if you
ran a 16 replica REMD simulation on 64 cpus you would have 4 cpus doing each
of the 16 MD calculations.
I hope this helps.
All the best
Ross
/\
\/
|\oss Walker
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun May 09 2010 - 20:00:05 PDT
