Hi, Joseph,

On 29.04.10, Joseph Bester wrote:
> On Apr 27, 2010, at 4:45 AM, Steve White wrote:
> > Hi!
> > 
> > Problem:  Running a cluster batch job with Globus, and specifying
> > the jobType to be 'multiple', want a convenient way to instruct each
> > process of the job to pick a unique input from a list passed as an
> > argument in the job description.
> > 
> > This seems to me to be a very obvious usage scenario.
> > 
> > I expected each process to get an integer environment variable, maybe
> > something like 'GLOBUS_JOB_PROCESS_INDEX', but I see no such thing.
> > 
> > This is GT 4.0.x.
> 
> The job environment is described in 
> 
> http://www.globus.org/api/c-globus-4.0.3/globus_gram_job_manager/html/globus_gram_job_manager_job_execution_environment.html

Yes, I saw this...
But is one of these variables a solution to the problem?

> 
> > Is some other way provided to get such an index?
> > Or to solve my basic problem of providing different inputs to each 
> > process?
> 
> If you are using pre-ws GRAM in 4.0.x, you can use the gram_myjob library to 
> coordinate tasks within a job for GT4.0.x. It is a pretty minimal API:
> 
> #include "globus_gram_myjob.h"
> 
> int size, rank;
> 
> globus_module_activate(GLOBUS_GRAM_MYJOB_MODULE);
> globus_gram_myjob_size(&size);
> globus_gram_myjob_rank(&rank);
> 
> For WS-GRAM, there's a rendezvous service to do similar things, but is more 
> complicated to use.

Well, I'm doing WS-GRAM here.  Also, I'm using a client's pre-build
binaries...  I guess I could write a C wrapper...

But... are you saying there is no way built into WS-GRAM to get
just one input per process in a 'multiple' jobType job?

That would render it nearly useless, I think, except for applications
that somehow get input from a separate server which keeps track of 
which process gets which input...

Are you saying then that the jobType 'multiple' is in fact such a very 
special-purpose construct?

(Right now I'm experimenting with MultiJobs, but this interface does much
more than I need, on top of being very hard to use...)

> 
> > I can think of various ways to do this myself, but what a mess.
> > 
> > For instance I could have each process increment a counter in a 
> > file, but this will take some careful programming.
> > 
> > And it looks like DUROC would be an extreme overkill solution.
> > 
> > Thanks for your help!
> 

-- 
| -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -
| Steve White                                             +49(331)7499-202
| E-Science                                        Zi. 27  Villa Turbulenz 
| -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -
| Astrophysikalisches Institut Potsdam (AIP)
| An der Sternwarte 16, D-14482 Potsdam
|
| Vorstand: Prof. Dr. Matthias Steinmetz, Peter A. Stolz
|
| Stiftung privaten Rechts, Stiftungsverzeichnis Brandenburg: III/7-71-026
| -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -

Reply via email to