Graeme Stewart wrote:
>>> 1. Job visibility: this suggests a scenario where users are not allowed
>>> to run their own executables, but instead are only allowed to manage
>>> parameters/configuration, and input/output data. Executable software is
>>> all deployed by sites or through the VO-specific software manager.
>>
>> I'm afraid that this model won't work for VOs like pheno, where each
>> one of our 20 odd users runs their own software. I would have to make
>> them all SW managers. Alternatively, we would need to give up on the
>> idea of a common theory VO, and splinter into project-specific VOs
>> (most of which would have 1 or 2 members).
>
> Let me echo David's point. We're now working with several research
> groups at Glasgow for whom the standard EGEE software deployment model
> dose not work at all.
>
> For the moment we support shell access to our own cluster to get around
> this issue (including for the Durham pheno people) - but clearly this
> does not scale at all.
As I see it, the "bigger" issue for a general purpose computational grid
is that I can imagine many sites which would be happy to allow VOs/users
to run jobs on their system provided there was tight control over the
executables, but not allowing "just anyone" to run "just anything" on
their systems. *Some* model which allowed sites this kind of control
would surely be valuable, with the flexibility that for "trusted" VOs
and users it would be possible to run arbitrary user-installed software.
Ian
--
Ian Stokes-Rees [log in to unmask]
Particle Physics, Oxford http://grid.physics.ox.ac.uk/~stokes
|