Samuel Genheden, 2012
ln -s /pfs/nobackup$HOME $HOME/pfs
This will create a symbolic link to the pfs-filesystem in your home director. So now you can easily goto the place were you should store your files:
cd ~/pfs
Also, on your desktop computer you can mount this directory on Abisko directly:
sshfs user@abisko.hpc2n.umu.se:/home/u/user/pfs Abisko
user is your user name, and u is the first letter in your user name.
There is another queue-system on Abisko, so all the commands are different if you compare to Platon. Check out this page for a comparison.
/home/g/genheden/pfs/Amber11, i.e., in my directory, but everybody in the same project as me can read it.Executing with 8 cores
#!/bin/bash
#SBATCH -N 1
#SBATCH -n 8
#SBATCH -t 24:00:00
#SBATCH -A SNIC020-11-20
module add intel-fortran
module add openmpi/intel
export AMBERHOME=/home/g/genheden/pfs/Amber11
export PATH=$AMBERHOME/exe:$PATH
srun --cpu_bind=rank -n 8 sander.MPI ...
Change 24:00:00 to your wall time and ... to the regular input things to sander. Note that sander.MPI could be changed to pmemd.MPI to execute pmemd
Executing with more cores
#!/bin/bash
#SBATCH -N 4
#SBATCH -n 32
#SBATCH -t 24:00:00
#SBATCH -A SNIC020-11-20
module add intel-fortran
module add openmpi/intel
export AMBERHOME=/home/g/genheden/pfs/Amber11
export PATH=$AMBERHOME/exe:$PATH
srun --cpu_bind=rank -n 32 sander.MPI ...
Here I have used 32 cores, and I have not tested anything else at the moment. You will get a speed up of about 2.7, compared to 8 cores. Note that Amber would like to have cores that have 2 as a base, i.e., 2, 4, 8, 16, 32, 64, etc. Note also that I have spread out the job on several nodes, -N 4. If you attempt to use this on a single node, -N 1, the execution will be about 3 times slower.