Storage
HPC and Data for Lattice QCD
Storage
Basic rules for storage
Disk space is available for two different kinds of storage:- pseudo-temporary storage for running jobs - permanent storage
1) Temporary storage
Each apemaster machine has a *local* /data/ disk intended for temporary storage of data for running jobs of the corresponding crates, units or boards:apemaster5:/data/ nc[67] apeamster6:/data/ nc[45] apemaster7:/data/ nc[23] apemaster8:/data/ nc[01] apemaster9:/data/ c[0123] apemaster10:/data/ c[4567] u8[02] b8[13]0
NOTE: This disk space is not in the backup! It will definitely not be appreciated if this disk space is used for other purposes except for storing job input and output data.
NOTE: We expect files on these disks not to be older than 1 monts! Older files may be deleted without further notice in case disks get full!
2) Permanent storage
Each project has its own project file space in the AFS. Please check with the corresponding administrators on how to use this disk space. Note that not all of the AFS space is backuped!Project | Path | Coordinator |
---|---|---|
Alpha | /afs/ifh.de/group/alpha | Rainer Sommer |
ETMC | /afs/ifh.de/group/nic | Karl Jansen |
QCDSF | /afs/ifh.de/group/qcdsf | James Zanotti |
Recommendation
Jobs should be started from the AFS disk space. Input files could be copied by the job script from the AFS space to the local /data file system at the beginning of each job using afscp and vice versa after job execution on APEmille finished:#!/usr/bin/zsh #$ -V #$ -cwd
source /zroot/klogin.zsh
JOBDIR=$PWD TMPDIR=/data/my_name
# Copy files from job directory to temporary run directory for file in *; do afscp $file $TMPDIR || exit 1 done # Change to temporary directory and execute job cd $TMPDIR
APErun -- -o yz my_program.jex || exit 1
# Copy result files back to job directory for file in *; do afscp $file $JOBDIR || exit 1 done cd $JOBDIR # Remove temporary directory rm -rf $TMPDIR