On Thu, 2006-04-13 at 12:25 +0100, Coles, J (Jeremy) wrote:
> Dear All
>
> Minutes from yesterday's meeting have been uploaded:
> http://agenda.cern.ch/fullAgenda.php?ida=a06682.
Hi Jeremy,
Many thanks for that. (A review of the presentations given at HEPiX has
been most interesting.)
I do have a couple of questions (not necessarily aimed at Jeremy):
* I get the impression that the experiments are expecting that the
different VOMS roles for their VO are mapped to different sets of pool
accounts at each site. (This is that different fair-share policies can
be implemented per-role.)
However, this will possibly lead to an explosion in the number of pool
accounts needed at each site, which may not scale. (Both in terms of
absolute numbers of accounts needed -- think 16bit UIDs -- and in terms
of response-times for implementing support for new roles.)
Has any thought been put into using some mechanism other than a job
owner's UID/GID to use as the basis for local site scheduling? (I'm
thinking that the RSL 'project' field, mapped to some appropriate
batch-system variable, could work nicely here..)
* What's going to happen to the DTEAM VO once the OPS VO has been
established? Presumably there will be some sort of transition period
when support for both VOs must exist in parallel, but what are the plans
for the DTEAM VO after that -- will it be disbanded?
Cheers,
David
--
David McBride <[log in to unmask]>
Department of Computing, Imperial College, London
|