Donal K. Fellows wrote:
Szalai Ferenc Attila wrote:
What is the regular way in the current JSDL schema to differentiate jobs which are using MPI, PVM or special checkpoint libraries?
There is no specific construct for what you are wishing to achieve. Within the general JSDL model of the world, such things would occur in a few possible ways:
* Explicitly through invoking some kind of wrapper code that sets this sort of thing up. * Implicitly through the JSDL consumer inspecting the resource set requested and deciding that the use of MPI, etc. is necessary for the running of the job. * Explicitly through the use of some kind of extension to JSDL (which I would expect in these cases to be either done within the Resources or the Application sections).
The first option strikes me as a hacky way to get going on this sort of thing quickly, but the other two are much more robust techniques for the longer term, since they mean that users do not need to understand so much about the actual configuration of the resources that they are using (a Good Thing in my experience).
I prefer last two one as well.
Ultimately we need to understand these sorts of things better so that we can address your sorts of use-cases more fully in future versions. To that end, could you please describe what you're trying to do in more detail? It will really help as we go forward beyond JSDL 1.0.
We are using the JSDL in our system which now provides common interfaces to different kind of local queue systems like Condor, SGE, etc. When the JSDL comes we have to create a queue system specific submit file from that. For example in the Condor system I have to specify explicitly the universe of the job which can be vanilla (= shell script), standard (= using condor checkpoint), MPI, PVM etc. So I need to have something in the JSDL which provides thise kind of information. Is something like a mapping problem. Generally: How to map JSDL to different queue system submit file? -- Regards Ferenc