Verne and Kraken do not currently share any file systems except users' home directories. However, data can be transferred between the two systems using any of the standard data transfer tools. For automated, scheduled data transfers between the two systems, NICS recommends that users stage their data through HPSS. This has the added benefit of also creating a backup copy of the data in the HPSS archive.
This process is slightly complicated by the fact that Verne does not have shared scratch space at the moment, so it is currently necessary to stage data both in and out of a Verne node in the context of a single job.
Staging Data from Kraken to Verne
First, push the data from Kraken into HPSS:
#PBS -N hpss-in-kraken #PBS -q hpss@nid00016 #PBS -l walltime=1:00:00 #PBS -j oe cd $PBS_O_WORKDIR hsi put file.dat
qsub hpss-in-kraken.pbs
Retain the jobid of this job, as it will be used for a job dependency on Verne.
Second, pull the data from HPSS to Verne, process it, and push the results back to HPSS:
#PBS -N hpss-out-verne #PBS -q analysis@verne #PBS -l walltime=1:00:00 #PBS -j oe cd $TG_NODE_SCRATCH hsi get file.dat [do work] hsi put new-file.dat
qsub -W depend=afterok:$krakenjob hpss-out-verne.pbs
This second job will not run until the first job completes successfully.
Staging Data from Verne to Kraken
First, push the data from Kraken into HPSS:
#PBS -N hpss-in-verne #PBS -q hpss@verne #PBS -l walltime=1:00:00 #PBS -j oe cd $PBS_O_WORKDIR hsi put file.dat
qsub hpss-in-verne.pbs
Retain the jobid of this job, as it will be used for a job dependency on Kraken.
Second, pull the data from HPSS to Kraken:
#PBS -N hpss-out-kraken #PBS -q hpss@nid00016 #PBS -l walltime=1:00:00 #PBS -j oe cd $TG_CLUSTER_SCRATCH hsi get file.dat
qsub -W depend=afterok:$vernejob hpss-out-kraken.pbs
This second job will not run until the first job completes successfully.
