Hi Gabriel,
GRAM does not create a JOB-HASH dir under the users home dir. Condor-
G does though, I assume you are using that. Perhaps there is a way to
change that in Condor-G? And then you could export that dir.
GRAM does have a job description variable GLOBUS_SCRATCH_DIR that can
be used in a job description and is intended for something like our
situation. Here is some information on configuring a value for this
variable.
http://www-unix.globus.org/toolkit/docs/4.0/execution/wsgram/admin-index.html#s-wsgram-Interface_Config_Frag-jndiconfig
Here is some info on using it in a job description
http://www.globus.org/toolkit/docs/4.0/execution/wsgram/schemas/gram_job_description.html#SchemaProperties
The default is ${GLOBUS_USER_HOME}/.globus/scratch. Note: GRAM does
not create that dir for the user.
I don't think the scratch dir has gotten a lot of use, but you could
define it to point to a shared dir with the cluster nodes that is not
under the users home dir. You could do that grid-wide and then
clients can start to use GLOBUS_SCRATCH_DIR instead of GLOBUS_USER_HOME.
Regards,
Stu
On Jan 31, 2008, at Jan 31, 7:20 AM, Gabriel Mateescu wrote:
Hello,
I would like to ask a few questions about configuring
the working directory created by WS-GRAM for each
GRAM job.
WS-GRAM creates for each job a directory hierarchy under
${GLOBUS_USER_HOME}/.globus/JOB-HASH
This is problematic when we want the job working directory
to be shared among multiple back-end clusters.
This is because this value of the job working directory
causes a contradiction:
1. ${GLOBUS_USER_HOME}/.globus/ must not be
exported because it holds the user's private key
2. ${GLOBUS_USER_HOME}/.globus/ must be
exported because we want the above working
directory to be available to multiple batch systems
(i.e., not co-located with WS-GRAM).
Because users tend to store their cert/key pair under
${GLOBUS_USER_HOME}/.globus/, requirement (1)
cannot be dropped.
However, requirement (2) could be dropped if one could
define the job's working directory not to be under
${GLOBUS_USER_HOME}/.globus/.
My questions are:
1. Is there a way to define the job's working directory to
be in a configurable location? And a related question:
2. Because the scratch directory looks like a good candidate
for the job's working directory, is there a way to use
as job's working directory
$GLOBUS_SCRATCH_DIR/JOB-HASH
instead of
$GLOBUS_USER_HOME/.globus/JOB-HASH
Thank you.
Gabriel
