Hi All,
We are using a 10 nodes cluster (each 16 CPUs with a total of 64 GB
RAM; 4 GB per node) for a relion 3d classification. The process is
taking more memory from SWAP when we run as below (please see the
attachment). Every iteration takes about 3 hrs on 80 k particle dataset
(5 classes).
mpirun -np 150 -machinefile .openmpi-default-hostfile `which
relion_refine_mpi` ... --j 6 --memory_per_thread 4
or
mpirun -np 150 -machinefile .openmpi-default-hostfile `which
relion_refine_mpi` ... --j 4 --memory_per_thread 2
or
mpirun -np 150 -machinefile .openmpi-default-hostfile `which
relion_refine_mpi` ... --j 1 --memory_per_thread 4
Is it normal or do we miss something on the command line?
Please provide your suggestions
with kind regards
Mani.
|