two doubts when installing Materials Studio in a heterogeneus cluster with SunGridEngine

Hi,

I'm Miguel Pan Fidalgo and I'm installing Materials Studio in a heterogeneus cluster with SunGridEngine [http://www.cesga.es/content/view/409/42/lang,en/ ]. The installation works properly and MPI and SGE works perfectly when I run a
job directly from the commandline (RunDiscover.sh for example).

My doubts appear when I'm trying to use the Windows client interface and try to follow the information of the "Info" directory:

"Materials Studio only supports clusters with PBS Pro or SGE queuing when the cluster is homogeneous (that is all machines in the cluster are of the same type and specification and are running the same OS) and the operating system is Linux. On all other configurations, you must ensure that any job submitted to a PBS Pro or SGE queuing system runs only on the local machine. "

well a little hack to do the tests (but this is not the point):
\$ grep mpan dsd_sge.pm
        my \$qsub = "qsub -l arch=64,num_proc=1,s_rt=00:10:00,s_vmem=2G,h_fsize=20G"; # mpan


first doubt: avoid to run a job with an inexistent user in the cluster
----------------------------------------------------------------------------------------------

According to official help: "All jobs submitted to a queuing system by the gateway are run as the user who installed the gateway, so it is necessary for this user to have permission to run jobs on any queue that is exposed through the gateway. However, Materials Studio allows fine-grained control over queue access by respecting the queuing system authorization for the user submitting the job. On password-protected gateways, the user name used to determine queue permissions is the name entered when logging on to the gateway."

\$ ./gwinfo
Getting current gateway parameter information...

gw_archivejobinfo = yes
gw_authentication = password
...

\$ ./gwusers -list
3 user(s):

gatekeeper
mpan
homer


We don't have a homer [simpson ] user in our cluster...

but when I'm running the Windows client interface, I'm able to send jobs to the queue system. The gateway user is "homer" but the propietary of the submitted job is the user that run the gateway daemon. The job must be submitted with the user registered in the gateway (for example mpan) but it is submitted with the user running the daemon.

any idea to avoid this problema? any missing configuration?

this is a problem because the account system doesn't work with this behaviour.


second doubt: work directory for the submitted job
------------------------------------------------------------------------

two submitted jobs with different users:

\$ qstat -j 5094576 | grep cwd
cwd:                        /opt/cesga/MaterialsStudio50/MaterialsStudio50/etc/Gateway/root_default/dsd/jobs/HMFDH
\$ qstat -j 5094591 | grep cwd                                                              
cwd:                        /opt/cesga/MaterialsStudio50/MaterialsStudio50/etc/Gateway/root_default/dsd/jobs/HMGYZ

the tmp directory: /opt/cesga/MaterialsStudio50/MaterialsStudio50/etc/Gateway/root_default/CGItmpfiles

this is another problem because the jobs must run (and write files) in the user space (\$HOME) and not in a shared folder. the temp directory is already define in the queue system (\$TMPDIR).

any idea to fix this?


thank you so much

--

Atentamente,

Miguel Pan Fidalgo (Applications Technician)
mail: mpan at cesga.es
web: http://www.cesga.es
Avda. de Vigo s/n 15705, Santiago de Compostela
Telf.: +34 981 569810 - Fax: 981 594616
-------------------------------------------------------------------------