A value of ${GLOBUS_SCRATCH_DIR} is set in etc/gram-service-<Local Resource 
Manager>/jndi-config.xml. By default it is ${GLOBUS_USER_HOME}/.globus/scratch. 
It doesn't mean that this directory exists. Usually user home directories are 
over NFS that is not too fast. Bigger clusters usually have so called scratch 
directory that is over Lustre and mounted as the directory /scratch with 
separated directories /scratch/${GLOBUS_USER_NAME} for every user.
So if you use a default WS GRAM configuration you need to create 
${GLOBUS_USER_HOME}.globus/scratch before you submit a job. If you have a 
scratch directory different from filesystem with user home directories and you 
want to use this scratch directory you need to modify etc/gram-service-<Local 
Resource Manager>/jndi-config.xml. You can also change 
${GLOBUS_USER_HOME}/.globus/scratch to ${GLOBUS_USER_HOME}.

Regards,
Lukasz


On Mar 22, 2010, at 3:59 PM, Dougal Ballantyne wrote:

> Hi,
> 
> I am trying to work out a way to have job specific directories using
> GT4.2.1. Below is the XML job I would *like* to run.
> 
> <job>
> <executable>/bin/ksh</executable>
> <directory>${GLOBUS_SCRATCH_DIR}/${GLOBUS_JOB_ID}</directory>
> <argument>job_script.ksh</argument>
> <stdout>job.log</stdout>
> <project>test</project>
> <fileStageIn>
> <transfer>
> <sourceUrl>gsiftp://globus.example.com/home/testuser/job_script.ksh</sourceUrl>
> <destinationUrl>file://${GLOBUS_SCRATCH_DIR}/${GLOBUS_JOB_ID}/job_script.ksh</destinationUrl>
> </transfer>
> </fileStageIn>
> <fileStageOut>
> <transfer>
> <sourceUrl>file://${GLOBUS_SCRATCH_DIR}/${GLOBUS_JOB_ID}/job.log</sourceUrl>
> <destinationUrl>gsiftp://amsdc1-n-s00114.europe.shell.com/glb/home/nldbae/job.log</destinationUrl>
> </transfer>
> </fileStageOut>
> <fileCleanUp>
> <deletion>
> <file>file://${GLOBUS_SCRATCH_DIR}/${GLOBUS_JOB_ID}/job_script.ksh</file>
> </deletion>
> <deletion>
> <file>file://${GLOBUS_SCRATCH_DIR}/${GLOBUS_JOB_ID}/job.log</file>
> </deletion>
> </fileCleanUp>
> </job>
> 
> However there seems to be way no to get Globus to create the
> ${GLOBUS_SCRATCH_DIR}/${GLOBUS_JOB_ID} directory.
> 
> Is there any advice on how to achieve this?
> 
> Kind regards,
> 
> Dougal

Reply via email to