Attending: Jens (chair+mins), Dan, Luke, Matt, JohnH, Sam, Brian, Winnie, Ste

The DPM workshop - tomorrow, two half days.  Thought Raul was going but apparently he isn't?
Matt is volunteered to give a presentation on behalf of the UK; inviting feedback from DPM
sites.  For example, Bristol needs to support HFDS which is not available on DOME; Liverpool
plan to migrate to DOME "soon".

Since Alessandra is also speaking at the workshop (on token-based access), the UK is at
least fairly well represented on the agenda, and others will/should be attending remotely.

We have also fed back a great deal of suggestions on improvements to the documentation.
However, theirs is a closed wiki, so we cannot update it ourselves - in which case it would
make sense to update the GridPP wiki instead.  In fact it makes sense to update our wiki
anyway... documentation is more detailed now but assumes you use puppet.

In particular moving LHC tribes off SRM is not so hard; ATLAS command central for example just
need to set the site to non-SRM.  Moving non-LHC off SRM is slightly more tricky.

Is there such a classification as "large T2D", "medium T2C", "medium/small T2C"?  Probably each
site is sufficiently different that we cannot make a broad recommendation for each category.
Also, the future of storage at a given site will depend on the state of the current hardware
and future funding.  A cross-site support model is still hard to implement because universities'
IT teams tend to not like giving the keys to the (virtual) house to someone who is not a member
of staff.


https://indico.cern.ch/event/776832/
Matt
Site-by-site 
Birmingham - decommissioned for EOS (at ALICE request) 
Sheffield, Cambridge - not planning on running an SE for much longer (I could be wrong for Cambridge), may skip DOME. 
Bristol - outlier, ??? 
Brunel, Lancaster, Manchester - DOME'd and no plan to migrate away from dpm. 
Glasgow - Sam can phrase this one 
Oxford - DOME'd, but unsure of long long term plans for SE (they have no storage in warranty iirc) 
Liverpool, ECDF, RHUL - middle sized sites, planning on moving soon when things look safe. 

(looking at it this way makes me feel quite good about the UK DOME situation) 

Other SE flavours - 2 dcache, 2 STORM, 1 EOS, some xcache. 

It might be worth mentioning our users: 
Mainly atlas and cms, but lhcb do have storage at some sites. New user groups coming - a lot of astonomers (ska, lsst, cta). Increased access via DIRAC and the associated toolsets. Potential for user groups to want to move beyond x509, and away from gfal!
MD
Luke
Critical for Bristol: Will HDFS be supported with DMLite in the future?
L
John
You're right for Cambridge - I hope to start ATLAS moving next week.
JH
Daniel
RHUL have issue they would like to use hdsf (it's what they use for the T3) but I understand this is not supported by dome. smae issue for Bristol as they use hdfs now
DP
Matt
Thanks for the feedback John and Dan.
MD
Brian
i have another meeting in 5 mins
sorry I have to leave
BD
Luke
Yes, Kubernetes is nice, but needs some work to set up
L
Daniel
Each site is different. Need to sit down with each site, work out what they want to do /can do and match with what we (gridPP) want them to do with available software/services options.
we need a catalogue of options
DP
Luke
XrootD SE ticks all of our SE boxes (since users want to be able to transfer CMS data to our site): 
https://opensciencegrid.org/docs/data/xrootd/install-storage-element/
L
Samuel
Luke: yes, I think an xrood SE (which is basically just an xrootd server, really) is a good match for bristol
SC
Daniel
with hdfs?, might also work for rhul
DP
Luke
^ yes, time is always an issue
yes, with HDFS
OSG has had a plugin for years
L
Samuel
(the OSG plugin is also sort of what inspired the ceph plugin we're using here / RAL, so it's solid)
SC
Luke
yum install xrootd-hdfs 
add 
xrootd.fslib /usr/lib64/libXrdOfs.so 
ofs.osslib /usr/lib64/libXrdHdfs.so 
to /etc/xrootd/xrootd-clustered.cfg 

done
L
Today at 10:39 AM