Perhaps I'm showing my conservative nature, but here goes....
On 2016-10-21 10:35, Tom Whyntie wrote:
> a new cluster at the ISS that would be:
> 1) capable of running LHCb/MoEDAL simulations
> (i.e. with CVMFS, AFS, interfacing with DIRAC, etc.) and
> 2) potentially compatible with joining the WLCG at a later date.
The "traditional" path is to chose a batch system that is
supported at many grid sites, and set it up in a manner that
is consistent with the set-up at many grid sites.
Access would be via batch command line interfaces, e.g.
if the platform were (say) htcondor, then users would run
condor_submit from their workstations, which would
be equipped with htcondor clients.
Later, when x509 access is wanted, and access is via the grid,
via dirac, or whatever, the traditional path would be to
install some CE that supports the batch system; for htcondor,
that could be ARC, or maybe CREAM, or HTCondor CE etc.
I.e. business as usual. And since we use the ARC/Condor system
here and it works well, and it's well supported overall, and since
I documented the process we used at Liverpool for making such a cluster
(which I copied from RAL) on the wiki then I have an obvious preference
for that system - we know that it works 100%. The first part of the
set-up (getting htcondor going) is not that onerous. Obviously,
converting it to a grid site later is a big step up.
> Given recent discussions on "grid lite" sites, are there any good
> sources of information or guidelines about what hardware/software
> would be good to go for at this point?
It's hard to say about that. We've done plenty of testing of VAC
using VM payloads, which works very well and I made a poster for it at
CHEP,
but I get the impression that payload construction and distribution is
quite different from "traditional" grid computing. It works well, but
it may involve quite a lot of commitment by the experiment to get it
started up in the first place, esp. when they don't need to be connected
to the grid in the first place. Andrew McNab would say more about that.
So, in conclusion, if ISS gets hardware for the "traditional" path, that
has a level of historical certainly and it can be Gridified later. Or it
can
be converted wholesale to "modern technology flavour of the month" at
some later date, once that Silver Bullet is finally found.
Cheers,
Ste
|