Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of Gordon, JC (John)
said:
> It would also be more effective if we said what we wanted to
> happen. In
> responses to tickets I have seen biomed blame the middleware but they
> are using lcg-cp. If they used lcg-cr in every job, would it make
> multiple copies to the local SE or is it smart enough to notice that
> there is already a copy locally and only make the first copy?
As I understand it the problem (well, one of the problems) is more on
the reading side, they seem to be trying to read the files remotely
rather than sending the jobs to the local site. More generally they seem
to be hitting scale limits in a variety of places. In a sense they
aren't doing anything wrong, just trying to use the grid. What we should
be doing is learning where the bottlenecks are and either trying to fix
them, or to put limits that stop people exceeding what the system can
do, or at least providing clear guidance so users can know what the
system can and can't cope with.
Stephen
--
Scanned by iCritical.
|