Attending: Steve, Jens (chair+mins), John H, Tom, Lukasz, Raja, John B, Sam, Brian
Apologies: Duncan
New post from Brian on file deletion – good stuff.
No operational problems.
·
Support mailing list works well
- "storage team brilliant"
·
Haven't made use of the list
archives to locate knowledge yet
·
Space tokens and managing space
for new/small VOs
o
Very small VOs are members of GridPP – not worth overhead in setting up anything else
o
Steve suggested sub-VOs.
§ Biomed uses sub-VOs [see details in chat log]
§ And NorduGrid [see chat]
§ And ATLAS (Brian)
§ Also DiRAC was designed with each DiRAC site as a sub-VO, although not configured as such.
Can be reconfigured.
§ IIRC, Pheno was basically individual users
doing different things (so the sub-VO is the individual)
§ Sub-VOs need support for VOMS groups
§ Can space tokens be allocated to individual sub-VOs? Sure.
§ Also need to account by sub-VO, storage usage, jobs
§ Needs support in VOMS impl for some
systems, e.g. when the group is present, it is honoured
in the storage system. VOMS-aware, local user mappings.
§ DPM maps DN->uid; gid
is based on the first VOMS entry (e.g. subgroup).
§ Clearly different procedures for different storage systems to
support this.
§ There is an OGF WG called VOMSPROC which wrote something on
interpretations when multiple VOMS attributes are present
§ There are permissions in VOMS to enable separate adminis
of individual groups
§ Need requriments input: no need to
implement heavyweight security if it is not needed by the VO (cf. DiRAC – separation turned off in CASTOR but could be
reintroduced with gridmap and nschown/nschmod)
o
There is some lifecycley stuff: space tokens tend to be needed only when
the VO reaches "maturity," yet can be difficult to retrofit.
o
Need support for
"default" space token if not using SRM with space tokens; for DPM, it
is assigned to the pool rather than directory (perhaps similar to how CASTOR
selects default service classes?) Not
ideal – and will space tokens continue to exist if experiments want to move
away from SRM?
·
Support for non-WLCG VOs –
already received feedback from Luke, Alessandra, Marcus, Tom
·
Need to document the use of the
infrastructure
o
E.g. squillions of small files are bad, but we mainly know this from
testing it rather than speculating – while it is bad to have to redo data
transfers, sometimes one just need to try stuff to see if it works rather than
spending ages figuring out the best approach (DiRAC
being the example)
o
Can't hack bespoke stuff for
every single VO – well maybe a little bit but best to reuse when pos.
o
Some VOs come to the party bringing
their own bottle and sometimes furniture (see below)
|
|
VO1 |
VO2 |
VO3 |
VO4 |
|
Apps |
VO |
VO |
VO |
VO |
|
Data mover |
VO |
VO |
VO |
GridPP |
|
Catalogue |
VO |
VO |
GridPP |
GridPP |
|
Data tools |
VO |
GridPP |
GridPP |
GridPP |
|
Middleware |
GridPP |
GridPP |
GridPP |
GridPP |
|
Fabric |
GridPP |
GridPP |
GridPP |
GridPP |
o
That sort of stuff. Need to
meet the VO in the middle (or at least somewhere in-between)
o
CERN@School (Tom) - "DFC great!"
o
Sam: Some VOs use us just as a
data infrastructure (DiRAC, SNO+), some use both data
and jobs (LSST), some use only jobs (LIGO, LHCb)
·
Sites of unusual size to
present themselves
o
Is Bristol volunteering as a
"small site"?
o
Oxford as medium site
·
Where are we in 5 years' time
o
T2C
o
T2D for WLCG
o
T2D frr
non-WLCG
o
Maybe sites can be hybrids like
today when a site might be a T2 for ATLAS and T3 for CMS
Tom Whyntie: (30/03/2016 10:00:41)
Mornin'!
Samuel Cadellin Skipsey: (10:12 AM)
Apologies for lateness
Lukasz Kreczko: (10:12 AM)
there seem to be many VOs with sub-vos: http://operations-portal.egi.eu/vo/search
(like nordugrid)
Jens Jensen: (10:14 AM)
thanks
Lukasz Kreczko: (10:19 AM)
biomed has
4 subgroups:
/biomed/bioinformatics
/biomed/medical_imaging
/biomed/drug_discovery
/biomed/lcg1
Jens Jensen: (10:19 AM)
ta
Lukasz Kreczko: (10:21 AM)
quite
different projects:
http://operations-portal.egi.eu/vo/view/voname/nordugrid.org
No SRM at Bristol
Tom Whyntie: (10:58 AM)
Cool, thanks, bye