Attending: Dan, Winnie, Steve, Jens (chair+mins), Alastair, RobA, Teng, JohnH, Sam, Matt, RobC, Brian

0. Difficulty at the Beginning

lcg-utils are not on CentOS7 systems (resp., UMD4).  However, lcg-utils are deprecated or at least
retired, and should not be used.  People should be using GFAL2 although it is understandable if they
already know and love lcg-utils...

1. Preponderance of the Great.

Round table of updates (but accompanied with discussion) of GridPP40.

Brian - WLCG workshop, UKT0, and future storage and future T2, storage and federation.  Although presumably evolutionary,
can include DataLake, dynafed, etc.  Combined DLMA and storage steering working groups.

Dan - alternative workflows - such as Spark and Hadoop.  Discussion about the (ab)use of EOS; Sam had suggested improvements
to DPM which would also bring to DPM some of the advantages of EOS.

Matt - xroot proxy cache.

Steve - politics.  Can we optimise code for storage like we can compute (cf. Dell's talk, and Graeme's).  Some
optimisation of data placement, and perhaps also in some of the code.  Successful scalability of WLCG to 10-100PBs only
through the tightly controlled data models.  Also a question of the mindset of users: managing their expectations.
Brian mentioned the "data train" and "data carousel"

Teng - T2 evolution and IPv6.  Evolving GridPP T2 sites into diskless and cache-only sites.

Jens - UKT0, how to make use of GridPP expertise but without forcing users into a model which is not suitable for them.
Notably other communities like climate or bioinformatics do their own thing (ie different from HEP), and climate at least
is not in the UKT0 community.  Thus, need to ensure all of GridPP is roughly aligned with ideas of how to present our
work and options in UKT0.  Also EGI fedcloud, which we've discussed a few times, is potentially relevant because UKT0
will be (is) using EGI's CheckIn service for federated identity management.  RAL's cloud deployment ought to be fully
forward and backward compatible with EGI fedcloud once CheckIn is enabled.  In general excellent representation of GridPP
storage and data management - we did get an agenda together in the end and managed to mostly almost stay on time.
Other than business as usual talks and evolution talks, the types of talks were also aiming a bit in the direction of
other technologies (eg Hadoop, iRODS) which we are not using much but are good to understand.


As regards the GDB, no one attended directly.


2. Return.

Jens apologised for missing the later suggested additions to the agenda (sent to mailing list just before the
meeting.)

Dan had requested a slot on performance (see chat) but will talk next week when he's done some more work on it.

Alastair mentioned the RUCIO trial for SKA at RAL: Rohini Joshi from Manchester and Mario Lassnig from CERN will
be visiting RAL later this month and will do some testing - could involve other sites (see chat).  Could it be provided
to others, e.g. the GridPP or dteam VO, so we'd know what we have?  This is second on the todo/priority list.  Currently
looking at S3 integration; can use WebDAV.

OSG and CERN have RUCIO servers; OSG looking at using it for LIGO which currently has a relatively simple data model
(everything is copied everywhere) but may have to evolve into a more sophisticated model as they get more data.

Alastair will contact QMUL and Cambridge about involving them in the test.


Daniel Peter Traynor: (18/04/2018 10:00:44)
hi
I had some slides to show on storage perfromance with latest itel microcode
j: (10:08 AM)
https://indico.cern.ch/event/684659/timetable/
Brian: (10:16 AM)
jens is still tlaking steve
Steve Jones: (10:17 AM)
afraid my sound is gone.... 
Brian: (10:17 AM)
in terms odf data management. fact CMS are looking into rucio means it sprobably sopmething we need to look further at.
Daniel Peter Traynor: (10:18 AM)
EOS is an "invented here solution"
I had a mild cold. I blame Steve
Ste Jones: (10:20 AM)
Yes... I gave Matt the cold so fast his feet didn't touch the ground.
Matt Doidge: (10:21 AM)
I think I imported my own cold. It might have received reinforcements from Steve's.
It was a fairly sickly sounding meeting.
Robert Andrew Currie: (10:24 AM)
Not sure which Rob that was (I think I'm having mic problems). For me, EOS looked interesting but thinking back I wonder whether the IPv6 work will push for all of the storage at sites to go dual-stack or just the dpm head nodes?
Matt Doidge: (10:26 AM)
I think just the head nodes will break things, thanks to the "not falling back to IPv4"-ness.
Robert Andrew Currie: (10:29 AM)
That was my assumption but we're tempted to make our production dpm dual-stacked next week and see what (if anything) falls over.
j: (10:34 AM)
https://indico.cern.ch/event/651352/
Daniel Peter Traynor: (10:45 AM)
we already support SKA for storage at QM. webdav works
John Hill: (10:47 AM)
So do we at Cambridge, though only for a nominal amount of storage