Attending: RobC, Jens (chair+mins), Patrick, Winnie, Teng, Sam, Duncan, Dan, Wenlong, Matt


0. OBP

   Sam's written a blog post and a very nice one, too!

1. I'll give a quick summary of last week's whatsit expo (only the interesting bits - or at least excluding the less interesting bits)

   Hmm maybe I should see if I could turn this into a blog post.  Most of the expo was "shaping your DX [digital transformation] journey"
   but there were some highlights, in cloud security, IoT data, cloud data, and the exhibition by WD as the sole storage hardware vendor
   in the hall.

2. LSST Great Leap Forward (RobC or Teng - came up in  TWG)???

   Rob and Teng don't know much more than we do, but it is reported that Singularity images are being rebuilt because
   results differed between workloads run in Europe and the US.

3. DOME status  (DPM sites obvs)

   Glasgow (Sam) and Edinburgh (Rob) presented plans for DOME upgrade of their DPMs.  DPM sites that are not Glasgow and not Edinburgh
   and not Bristol and not on DOME should also share their plans and timescales - whether it involves, like Edinburgh, upgrading other
   systems first (...CentOS7), or whether they have other dependencies.

4. Also from last week, I'd like to spend a bit of time discussing Oliver's talk from the GDB

https://indico.cern.ch/event/739883/contributions/3577297/attachments/1922942/3181621/QoS_Site_Survey.pdf

   This is a very interesting presentation and writeup.  We note the occasional urge to say "I told you so", and how things
   are sometimes being reinvented.  Specific notes from the discussion were:

   * Grid storage is sometimes hard to share; a thin layer interface (xroot, http, StoRM) would be happy to share an underlying
     distributed filesystem.  The economy of a shared storage (grid and non-grid) could be better for a T2.
   * Similarly there is not much point running dCache or DPM on a distributed POSIXish filesystem like CephFS
   * No good way to advertise your QoS.  If you had a disk pool of SHM or SSD, they would require different usage patterns
     (e.g. WORM); even the SRM durability (aka retention policy), latency, and locality would not be sufficient, nor GLUE.
   * Like the cloud storage expo (above), difference between resilience of storage and site and across sites; RAID6 is
     hardware; other resilience is obtained through LOCKSS and more sophisticated software models (and client models, like
     dynafed)
   * Caching, as we've seen, is not The Solution; it is highly dependent, as noted in the report, on experiment access
     patterns.  CMS has become slightly more cache friendly; ATLAS might do the same.  However, even if a T2 were to transition
     to a cache-only store, it might need to support other use of its storage system (like most sites do).
   * For non-LHC use, as we've seen in IRIS, there's the question of bringing new users to the grid or bringing the infrastructure
     to them - 

5. AOB




From Daniel Traynor to Everyone:  10:11 AM
Anything about CaaS? Containers as a service, seems to be the next bus word