Am 29.01.2013 um 01:05 schrieb Dave Love: > Reuti <[email protected]> writes: > >> Am 25.01.2013 um 18:37 schrieb Tina Friedrich: >> >>> Don't know if that helps you, but remember that SGE created a temporary >>> directory (something like /tmp/#JOB_NO.#TASK_NO.QUEUE). Available to >>> submitted scripts as $TMPDIR. I believe it even cleans it up automatically >>> once a job finished. >> >> Yep. > > [Except for loosely-integrated parallel jobs?] > >> The job script will be transferred into the spool directory of the >> exechost (unless a binary is submitted with "-b y") by SGE before the >> job starts. If this is local on each exechost (which is often done to >> lower NFS traffic), it's already there (although not the scratch >> area). It can be accessed by $JOB_SCRIPT (besides $0 in Bash and other >> shells). >> >> It's not started in the CWD, but something like >> /var/spool/sge/node09/job_scripts/202092. > > That looks as if it's answering a different question, as I read it. The > job is started in the job's working directory, as defined at submission, > home by default.
I got the impression that the important stuff is in the jobscript (maybe some kind of here-document), hence the jobscript should be transferred to the node to avoid opening it in /home to read all the data. If it's not the case, you are right. -- Reuti _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
