Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ptp-dev] more on RM XML 
----- Albert L. Rossi <arossi@xxxxxxxxxxxxx> wrote:
> 
> ----- Greg Watson <g.watson@xxxxxxxxxxxx> wrote:
> > Al,
> > 
> > Looking good! A couple of things. 
> > 
> > 1. Looking at your PBS XML, I don't think you should be explicitly embedding ssh/scp commands in the job scheduler commands. These should just be the actual PBS commands required for the action. Whether this command is run locally or via an ssh channel will be determined by the RM instance configuration.
> 
> I was worried about this problem because it means we need to specialize the commands.  How else are we do distinguish a copy command from others, so we know to prepend scp, and host name on the target, as opposed to cp? 

Or how it is going to be handled by Jsch? I guess I've been thinking too much in terms of stuff I've written for the workflow engine.  Eclipse uses sftp, if I'm not mistaken.  In any case, I think we need something like this:

<copy>
  <from>
  <to>
</copy>

alongside <command>




> 
> > 
> > 2. I think the mechanism you want for #1 below is org.eclipse.core.variables.dynamicVariables. The syntax would be: ${object:field}. I'd prefer not to use reflection unless absolutely necessary.
> 
> I'll look.
> 
> I noticed already that the "directives" are incomplete.  Instead of -A, for instance, we need to specify the whole prefix # @ PBS -A.  More work to be done!  Had to drop off a bit this afternoon to do Fortran ... back on it tomorrow.
> 
> -Al
> > 
> > Regards,
> > Greg
> > 
> > On Jan 24, 2011, at 3:28 PM, Albert L. Rossi wrote:
> > 
> > > Hello all,
> > > 
> > > I have been updating the posted XSD (http://wiki.eclipse.org/PTP/designs/resource_manager_xsd) as I work through the first example XML, which I have also posted (PBS).
> > > 
> > > My plan of action is as follows:
> > > 
> > > 1.  Write the stub code for marshaling this XML into Java using JAXB.
> > > 2.  Implement prototype RM which extends Greg's abstract class and composes into itself the construction of the data object based on this XML.
> > > 3.  Perhaps try to do the same for OpenMPI (which poses other challenges).
> > > 
> > > Two things are currently on my mind.
> > > 
> > > 1.  I need a way in the replacement process to indicate an object (bean) field.  I don't think the Eclipse mechanism does this.  In the XML I have indicated this by ${object[field]}; this assumes there is a "get[Field]".  Are there overriding objections to using Java reflection?
> > > 
> > > 2.  I am trying to figure out the easiest route to describing the transformation of nodes:ppn into totalCores (to be used on the mpiexec line as -np ${totalCores}) in a general way.  ("for segment in 'nodes:ppn+nodes:ppn+...' do ... totalCores += nodes*ppn").  
> > > 
> > > Maybe I shouldn't bother and just ask the user to overspecify, that is:
> > > 
> > > Nodes:  2:ppn=32
> > > -np:    64
> > > 
> > > ... 
> > > 
> > > Al
> > > _______________________________________________
> > > ptp-dev mailing list
> > > ptp-dev@xxxxxxxxxxx
> > > https://dev.eclipse.org/mailman/listinfo/ptp-dev
> > 
> > _______________________________________________
> > ptp-dev mailing list
> > ptp-dev@xxxxxxxxxxx
> > https://dev.eclipse.org/mailman/listinfo/ptp-dev
> > 
>

Back to the top