Attending: Brian, Gareth, Jens, John B, John H, Sam, Steve, Raja, Adam, Govind, Elena, Jeremy, David

Apologies: Wahid

0. Operational blog wiki issues

High memory use of DPM xrootd (swapping) : DPM 1.8.8.  How would you inspect a
runaway process to see whether it really is a runaway process or it is just
very busy, or it is buggy, or something else?  Probably too complex to
document, but it may be a suitable thing for a tutorial.  Sam will raise the
question of the xrootd process with the developers.

LHCb reported timeout of xroot at T1s, all using dCache, but not all dCache
T1s: SARA is fine; IN2P3, PIC, and GridKa get the timeouts.  All their dCaches
are 2.6.X, and so is RALPP, so maybe worth checking RALPP.

The problem reported by Govind was discussed.  Maybe worth inspecting the logs
further and doing more tests (without letting ATLAS loose on it just yet.)

1. More on feedback from poll

Feedbacks are remarkably similar - we have seven responses now - generally
they like the format, and there were suggestions to have more technical talks
and perhaps even a hack day where people could get together and hack stuff.

2. Feedback from VOs? Either now, or questions we should ask them to
answer for GridPP33.

T2K still aiming to do work with (Web)FTS before reporting back to us.

Raja reported the operational problem for LHCb, but otherwise LHCb have no
problems related to GridPP storage.

As the proposal is to ask experiments/VOs to feed back on storage for
GridPP33, were there any particular questions we could usefully ask them to
feed back on?  Ewan suggests future planning, like whether they move to/from
xroot, SRM, smoke signals, telepathy, etc., and then disseminating which
choices they make.  Also whether they will be needing cloudy type storage.

Of course we should also do feedback in these meetings, but it can be hard to
persuade perhaps particularly the smaller VOs to come round because they are,
by definition, small.  So GridPP33 may be a good opportunity.  Feedback to
Dave Colling with suggestions for agenda.

3. Round table of Interesting Things., like GFAL plugins, catalogue
synchronisation, WebDAV, WebFTS, etc.

Sam
  - summer student working on erasure codes integrating with DIRAC's file
    system.  Could usefully be turned into a blog post when done.
  - DMLite with GFAL2, and GFAL2 with RADOS (CEPH's low level object storage
    library)
  - consistency checker for DIRAC7 (aka HEAD, not yet released), which seems
    to be well written (and it's in python).
  - S3 plugin for DPM (ie to make DPM talk to S3) a bit crufty
Ewan
  - DPM/ARGUS integration usable, or near enough.  Needed a few docs and
    tweaks.  No man page in rpm; but will be in next release.  Could also
    usefully be turned into a blog post.
Brian
  - GFALFS rpms, using WebDAV to mount filesystem, could be possible to mount
    T2 storage?
Jens
  - Spent some time getting CDMI storage working for EGI, not very
    successfully.  Stoxy didn't compile (documented on EGI wiki) due to bugs
    in old python build systems; recent dCaches not configurable using
    documentation so may try to roll back to 2.6.  Status of CDMI in dCache
    somewhat unclear, but that's what we're aiming to find out.  There is now
    also a CDMI implementation from GRNET.  Also tried the SNIA one but tried
    to compile in Eclipse which is not a good idea...

4. This might be full of Interesting Things. as well:

http://www.snia.org/events/dsicon2014/dsi_presentations2014

5. AOB


Raja Nandakumar: (06/08/2014 10:07:45)
http://lblogbook.cern.ch/Operations/18956
Govind : (10:09 AM)
/raid xfs logbufs=8,noatime,nodiratime
Raja Nandakumar: (10:16 AM)
Tutorial will be nice :)
Ewan Mac Mahon: (10:17 AM)
Difficult to do though, I think.
Raja Nandakumar: (10:17 AM)
But I am not a sysadmin ...
Ewan Mac Mahon: (10:17 AM)
It's not a simple idea to document
I think Adam at Imperial might have tried it - he was looking at doing the necessary for EGI fedcloud integrtion in general.
Speaking of interesting things, didn't someone mention something about dmlite ceph plugins recently?
jens: (10:35 AM)
:-)
https://www.gridpp.ac.uk/wiki/Quick_Guide_to_Dirac
Ewan Mac Mahon: (10:38 AM)
We should reclassify /all/ our source as documentation. That ought to have an impact on the documentation metrics.