[Wien] w2web -> cluster

[Peter Blaha]

In principle such things can quite easily be configured, but a few facts 
are still unclear.

What does it mean: you are not allowed to run w2web on the cluster ?
Once you can login to the cluster frontend, you can run a program there 
? Where's the difference between a "top" command which often runs for 
days on such frontends and w2web ?
If there is a firewall between, you can create an ssh-tunnel and use 
this tunnel.
--------------------
If you don't like this you need to know how the filesystems are set up.
You said, the users are using their local PCs, thus most likely their 
data are NOT available on your local cluster (over a NFS system ?)?

If this is the case, first (and last) thing you must do is use 
"migrate_lapw" to get the data to the cluster (and to get it back 
afterwards). Please check its options (migrate_lapw -h)
Then you must submit via    ssh cluster "bsub [options] -n N prun ./job" 
   your job.
A template for the job file would be the example you mentioned on the 
web, but most likely you have to adjust it.

In w2web you can define "execution-types" (in configure).

Define  types like
cluster8=cluster8.csh < %f   # for a 8 node queue
cluster16=cluster16.csh < %f   # for a 16 node queue   and so on ...

And write your own cluster8.csh script (or use bash,perl,...)
This script must read the command which should be executed (and is piped 
via   < %f   into your script).

This script must perform:

  "migrate_lapw -put ..."

create a job8 file (containing lsf-commands for a queue of 8 nodes, the 
code for generating the .machines file "on the fly" (from PBS_NODES or 
whatever the name on your queuing system is) and has the execution line 
(from %f) as last line. Note: You don't need to know in advance which 
nodes you will get from lsf.

ssh cluster "bsub [options] -n N prun ./job8"

check with qstat if your job is still running (check/ sleep 60/check/..) 
and when its finished

migrate -get    bring the files back.
-----------
If you have NFS, then everything is much easier and eventually only the
the middle steps are necessary.

Hope this helps

Marc Petitmermet schrieb:
> Dear List Members
> 
> Imaging that you have users who only know how to use wien2k using the  
> w2web gui. They are using it on there local Linux clients.
> 
> I'm now challenged to configure their setup so that they can submit  
> parallel jobs to a cluster (288 dual-processor compute nodes based on  
> AMD Opteron 250 processors, 8 GB memory connected by means of a  
> Quadrics QsNet II interconnect). Installation of wien2k on the  
> cluster and password-less ssh connection from client to cluster are  
> no problem. But how do I get the job correctly submitted to the  
> queueing system (LSF, bsub [options] -n N prun ./program) on the  
> cluster from the w2web gui on the client (we are not allowed to run  
> w2web on the cluster itself)? Furthermore, I don't know the node  
> names which will be used in advance. I have found some information on  
> http://www.wien2k.at/reg_user/faq/pbs.html but I still don't know yet  
> how to "glue" all this information together. Is this possible at all  
> using the w2web gui? Thanks in advance for any pointers.
> 
> Regards,
> Marc
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien