> Hello
>
> I have a few questions:
>
> I have found on the ibm homepage (I think it was the ibm homepage) a
> tutorial that describes how to submit a parallel job but unfortunately I
> cant find it any more! I think there were two mathematical functions that
> ran parallel and when they were finished a final job combined the results
>
> I am thankful for every link etc that has some java sample code for the
> java api to test a few things(e.g wsrf-query with java );
Have a look at wsrf-* commands in $GLOBUS_LOCATION/bin. They are
all shell scripts and launch Java programs which are part of
Java WS Core. Look for the line EXEC=... in these scripts to see
which Java class is launched.
Download the sources of wsrf (cvs co -r globus_4_0_branch wsrf) and
have a look at these classes. They are quite small and easy to read
because all the commandline handling stuff is done in a superclass
BaseClient.
> I have access to a grid environment where a container runs on the front
> end node. If I submit a multijob where in every subjob i execute
> /bin/hostname I always get the hostname of the front end node. But isnt it
> the case that I should get the name of the node where the subjobs are
> executed e.g. node1 node2 (to the nodes where gram distributes the
> subjobs?) If I submit a job to a gatekeeper I get the name of the node e.g
> node1.
It should be the case, although it's the job of the local resource
manager (LRM) like Condor to dispatch the job to the nodes. I checked
that with a multijob (5 jobs) and the subjobs ran on 2 different cluster
nodes.
> Am I right that ws-gram also distributes the jobs/subjobs to the grid
> nodes?
It does not submit them to nodes directly but to an LRM. The LRM then
submits them to the resources under its control.
I wonder if you might have specified "Fork" in the EPRs of the
factoryEndpoints of the jobs of your multijob? In this case no job
would be submitted to an LRM. All jobs would be executed on the headnode
directly. You must specify the cluster-specific label here if you
want to run the jobs on the cluster.
> How can I see where the subjobs are running?
You can use programs of the LRM (condor: condor_q, condor_status,
pbs: qstat) to check what is running where.
> Is it also possible to submit a multijob to a gatekeeper? Because in every
> subjob is the epr defined?
No, there are no multijobs in Pre-WS GRAM.
Btw: you could answer on an email to our list if you want:
[EMAIL PROTECTED] had a question about multijobs; i think you
have some code for that and could help him out, right?
Thanks, Martin
> thx
>
>
> --
> Der GMX SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!
> Ideal für Modem und ISDN: http://www.gmx.net/de/go/smartsurfer
>
>
