[gmx-users] mpirun with remd

[Mark Abraham]

On 15/11/2011 12:22 AM, 杜波 wrote:
> dear teacher,
> when i do remd ,i use the commond :
> mpirun n8,8,8,8,9,10,15,16,16,20,20,22 -np 24 
> /export/software/bin/mdrun_mpi_4.5.5 -multidir ./0/ ./1/ ./2/ ./3/ 
> ./4/ ./5/ ./6/ ./7/ ./8/ ./9/ ./10/ ./11/ -replex 1000 -nice 0 -s 
> pmma.tpr -o md -c after_md -pd -table table.xvg -tableb table.xvg -v 
> >& log1 &
>
> node16:
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 14411 KD 16 0 51528 3552 2728 R 67.6 0.1 6:47.63 mdrun_mpi_4.5.5
> 14409 KD 15 0 51528 3612 2788 S 25.9 0.1 1:22.06 mdrun_mpi_4.5.5
> 14408 KD 15 0 51516 3396 2580 S 22.2 0.1 1:10.81 mdrun_mpi_4.5.5
> 14410 KD 15 0 51508 3212 2408 S 20.6 0.1 1:07.22 mdrun_mpi_4.5.5
>
> and we have 4 cpus on this node16 & there are no other program except 
> the system's:,
> why only one %CPU of thread (14411 ) is big ?

We don't really know enough to say. We don't know that this percentage 
is a valid metric on your system or that your mpirun invocation is 
valid. Set up a run that should take a few minutes and compare the time 
taken in serial with one or two processors, and with REMD systems of 
various numbers of replicas. Now you can start to track down what, if 
any, problem is occurring.

Mark