Attending: IanJ, Duncan, Jens (chair+mins), Winnie, Teng, Wenlong, PeteC, Sam, Matt


0. Op B P.

   Edinburgh problems with DOME upgrade; upgrading OS and DPM in one go is too great a leap forward.
   Can not generate new quota tokens, but can generate space tokens and convert them to quota tokens.

1. Updates on any, all, or none of the following topics:

CEPH for T2s; federated CEPH(fs)? for IRIS (cf Euclid)

   There are several questions here; trying to unpick them:
   - whether CEPH is a suitable for a T2
   - if more than one T2 provides CEPH, whether they can (or should) be federated
     - using CEPHfs (as kind of suggested in the Euclid talk)
     - Replicating objects (works better on WORM data, immutable)
     - Replicated at a higher level using dynafed, grid, etc.

   The latency may mean that a cross-site distributed filesystem (like DEISA used to run) may not
   be a good idea, or may not work as well as the customer would expect.

   As a meta-comment, we should of course be helpful to the user communities in trying to meet their
   needs, but it is entirely possible for them to ask for something that would take a long time to build
   and might then not work well - we should be realistic about what we can achieve and deliver for the
   users.

DOME upgrade saga

   (Ed problem above) Should it use the system database or whatever it was built for?  Sam thinks the
   only difference between the CentOS6 and 7 versions is the startup, ie whether it uses init.d or sysctl.
   Seems a bit old fashioned from a modern CI/CD perspective, but the differences can be managed.

Rucio for non-LHC users (specifically, docs)

   Revisiting the question of whether and how we should be recommending Rucio as a, or the, data
   "solution" for IRIS, Jens had started setting up a container based deployment, as waiting for the
   shared instance to be developed is delaying the work.  However, there were no other volunteers to
   work on this at this point.  James at Edinburg might be proddable; Pete would try.

Token-based access?

2. Hepsysman is next week and there's a chance to catch up (for those attending): https://indico.cern.ch/event/859095/

   We agreed that a survey/discussion of cluster/distributed filesystems would be a useful entry
   in the agenda.  Winnie actioned to prod Luke to produce slides or at least a few sentences, on HDFS, to Matt;
   Sam actioned to send CEPH slides to Matt.  Matt very kindly volunteered to bring the information together.

   Video (or Vidyo) might be available for the people who cannot attend in person.

3. Milestones for this quarter (I'm working on my report for Matt/Pete).

   The presentations/discussion for hepsysman would be a good example of a milestone, as it shows off GridPP work
   to related communities.  There were no other suggestions, so Jens would have to make some up.  Finishing the DOME
   upgrades at DPM sites this year would not be feasible, as Glasgow would not be able to, for various reasons; and
   people are reluctant for other equally good reasons to roll out major updates in December.

4. AOB

   Pete mentioned as a heads up that our views might be sought on "opportunities", like software dev & support
   where we know we have serious gaps.  Input would be sought on topics like DOMA and AAI.