Writing speed to the lustre file system
Hi,
we are experiencing a very slow writing speed of the EPOCH .sdf files to the lustre file system. This occurs for all of our small simulations with file sizes of ~ 100s of KB taking tens of minutes to write (!). We are all using different input.decks and therefore it does not depend on the kind of simulation that we are doing. It appear to only affect single core jobs, e.g. jobs that we run with the epoch1d code on the centos7 container directly and we did not use MPI. The one MPI job which we started did not show any I/O issue for the two written .sdf files (up to the node fail). As an example, I attached you the input.deck for the simulation I did. We only log into "virgo-centos7.hpc.gsi.de", load the modulefiles, e.g. "module use /cvmfs/phelix.gsi.de/modulefiles" and load the epoch code "module load epoch/...". after that we locally run epoch1d with the attached input.deck.
Cheers,
Johannesinput.deck