On Thu, Feb 11, 2016 at 4:19 PM, Jens Jensen <[log in to unmask]> wrote:
> On 10/02/2016 15:20, Ewan MacMahon wrote:
>>> I'm not sure why accounting would be an issue with a single endpoint (not
>>> an issue at QM with a single storm endpoint) but network clearly can be.
>>>
>> That's "single endpoint in the country" not "single endpoint at the site".
> Or it could be per T2, I guess? so have four in the country plus the T1.
> I guess that should be OK, too.
>
That should be ok too, but in that case user jobs would still need to
know at least about 5 different endpoints to contact for data or when
copy data into the GridPP.
This kind of endpoint should only be the contact point for user jobs
requesting a data file, but then the request should get redirected
from this endpoint to the actual server which has the file, and this
server serves the file to the user job directly in the end without
going through the initial point-of-contact for the user job. Similar
like it is for DPM on a site installation where a user job only needs
to know the DPM headnode which is in that case the endpoint, but gets
the data file from any of the servers on that site - only adding then
some layers on top of it to make the same principle country wide
working.
But for testing maybe we could try to have something like that for
only one of the T2 first and see how it goes... ?
Cheers,
Marcus
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|