Present:
Edinburgh: Phil, Andrew
RAL T1: Steve
RAL ADS: David
RAL Storage: Owen, Jiri, Jens
Apologies:
Alessandra, Jeremy
Long summary today, despite the fact that meeting finished on time.
New: I think it is now worth keeping a list of actions:
grep '^ACTION' `mhpath +drafts last`|awk '{print $2}'|sort|uniq -c
1 ALL:
1 Andrew:
6 Jens:
1 Jiri:
1 Phil:
1 Steve:
Don't worry, I'll delegate something :-)
0. Site update:
Went through this extra quick this time.. if you have anything here
you think we should all know about, please mail it to the list.
The Tier 1 production dcache is running with SURL
srm://dcache.gridpp.rl.ac.uk:8443/pnfs/gridpp.rl.ac.uk/data/dteam/
There is a subdirectory which will allow you to write files to tape
but I'm not telling you about it yet because we've found a(nother)
bug which we need to fix first. Also, at the moment it only works
if you're in dteam, but that's easy to fix. I expect the serious
bug to be fixed by today, will let you know when you can start.
1. Status of storage deliverables:
Two level 1 deliverables for GridPP2, Grid Storage, and Local
Storage Management.
They're there (except that there is still a bug in that interface)
but at the moment not very visible until the web server comes up
and is accessible. As of 1300 today, the name is in DNS but we
still need the firewall opened and ssh doesn't work... (it never
works the first time regardless of what they tell you).
ACTION Jens: Check if people outside can get write access (ssh).
Check whether we are allowed to use GridSite.
ACTION Phil: Check if Edinburgh can provide bugzilla.
ACTION Jens: Check if we can go via the helpdesk.
ACTION ALL: Send external URLs to Jens.
ACTION Jens: Update project page on CCLRC e-Science server and
on storage.gridpp.rl.ac.uk with external URLs.
2. dCache performance and testing
Status of client support:
srmcp: works
GFAL 1.4.0: reports protocol error - it should use DCAP though.
GFAL 1.5.x: apparently reported a TURL error, being investigated.
lcg-cr: Maarten reported errors against dCache 1.2.2.7
- latest.
We aren't testing gLite IO until it stabilises further.
Regarding documentation, Phil reported there's a postdoc at Glasgow
who's working on similar documentation stuff.
My tests show 8 secs to write a small file; 12-13 to write a 50
MB file. That's a single client write; I really would like to
start stress testing the Tier1 dCache...
Steve had 450 Mb/s peak from CERN to Tier 1 dCache at RAL.
Occasional problems would result in 0 length files being created,
and then the transfer could not be restarted because the file
exists (dCache doesn't allow that).
Since the meeting, I (Jens) was asked by Owen to resurrect the
SRM C API. I will do this (eventually) and put it in CVS.
ACTION Jiri: Figure out why GFAL doesn't use DCAP.
ACTION Jens: Update client documentation to include lcg-cr.
ACTION Steve: Send report to list on data transfers etc.
ACTION Jens: Resurrect C Client API
3. Deployment
Postponed for now, Jeremy has gathered all T2 deployment coordinators
here at RAL and storage deployment will be discussed on Friday.
ACTION Jens: With Jeremy, plan for deployment to more T2 sites.
4. DPM
Andrew has received source from Jean-Philippe Baud and functionality
test suite from Gilbert Grosdidier. It's using a socket interface,
the SRM interface is not yet running.
ACTION Andrew: Send URL to list (or me) when he wants us to start testing.
5. AOB
The question of Open-Sourcing dCache was discussed. Tony Doyle has
offered to get in touch with Mike Ernst and at least make the source code
available to us. Long-term view is still uncertain because we don't
know how dCache will perform in the tests, or whether it is feasible
for us to contribute to the development. Adding to the uncertainty is
the DPM and RAL's SE which are both alternative solutions.
Thanks,
--jens
|