Attending: Daniel, Gareth, Henry, Jens, Jeremy M, John B, John H, Matt D, Winnie, Robert, Sam, Steve, Wahid, Elena, Dave Lewney, Govind, Brian, Raul, Chris W, Jeremy C, George V 0. Sam - 2. Everyone else - nil. A reminder to everyone else to score a point or two. There were _no_ operational issues raised. 1. Quick status of Argus - could we meet our long outstanding milestone of Argus integration now? This is a long outstanding milestone in our deployment, and the PMB asked whether we were any closer to completing it. Until fairly recently, the problem was that DPM did not support Argus, but this has changed recently, and tests at Oxford and Glasgow were promising. More surprisingly, it turns out that other sites had quietly rolled it out and it seems that (nearly?) all sites have it? Or at least no one in the meeting would fess up to not having it. Argus support in StoRM is based on access to the SRM, not GridFTP, but that's good enough, at least for the milestone. 2. QMUL discussion - slides from Daniel to be presented - "the QMUL Storage Dilemma" - don't miss it! (The theme is the future T2, not just QMUL) Daniel's slides will be uploaded to the Indico agenda page: http://indico.cern.ch/event/313427/ Daniel from QMUL presented slides on the dilemma of whether to stick with Lustre - which currently has a bug preventing StoRM from fully working - or to start putting eggs into the CEPH basket. More generally, the question is also about the future T2. Brian says migration, by having two SEs, one Lustre and one CEPH, should be the way to migrate - experiments are asked to move their own data over, which updates the catalogues, and also discovers dark data. There would be a need to handle two namespaces [but the catalogues should do that?] At Sussex all data is production, so there won't be much permanent stuff to migrate. Working on parallel Lustre 2.X for some value of X, to have parallel copy along with existing 1.9; then 1.9 can be turned off and serve the grid use only. The "dirty hack" to support Lustre was not incorporated in 2.7, but it will be in 2.8. StoRM developers were not allowed time to integrate at their end. 2.7 also has small file support which is good for informatics users. For CEPH, the recommendation is to use error correction, and then turn off RAID6 on the storage. Would CEPH be better than Lustre for all them new-fangled interfaces we might be expected to run in the future, like clouds and whatnot? But Lustre is tried and tested, and both QMUL and Sussex have expertise with Lustre. Sussex had a 2x disk failure and media failure but managed to recover data. Some alerts had gone unnoticed due to delayed firmware upgrades, so SMART didn't work. Nevertheless, the array was rebuilt, so an interesting "near miss" (and useful to report!) Not much pressure on Intel for unsupported users, so Jeremy is considering buying support for a small test cluster and replicate the problem on the test cluster. Clearly option 2 is the preferred one (stay with Lustre, patch 2.7, move to 2.8 when approp.) CEPH still has unknowns in terms of grid use, and maybe the future is direct xroot to CEPH. Another fix for the StoRM support problem would be to turn off checksums. FTS can be configured to not require checksums. Not sure about Hammercloud; this will need to be tested. Sussex planning to have a new system ~Easter time, will be looking at MDS. 3. Storage outside of firewalls - time for discussion? Postponed - again - in the interest of time. 4. AOB Jeremy raised the question of a "HTTP deployment task force", which would oversee the adoption of HTTP. We have a use case for this, as MICE have been using HTTP access. However, we are slightly concerned that a task force could be required to roll out HTTP to all sites - only sites that need it should, er, need it. Some of the other VOs may also find use cases for HTTP - ATLAS for example. At the moment they are not looking for "volunteers" to join; they are assessing the need. Brian reports a GFAL writing issue at RAL solved (blog post?) Matt Doidge: (18/02/2015 10:03:45) Lancaster's behind on this, but we're behind on argus Paige Winslowe Lacesso: (10:03 AM) My impression from CW@QMUL re: ARGUS & StoRM is that there were bugs in StoRM preventing it working fully correctly Elena Korolkova: (10:04 AM) Are there instructions on the web? wahid: (10:05 AM) http://gridpp-storage.blogspot.co.uk/2014/08/argus-user-suspension-with-dpm.html Elena Korolkova: (10:06 AM) thanks, Wahid Samuel Cadellin Skipsey: (10:16 AM) (I'd argue that you don't really *need* a POSIX interface for Grid storage use...) Jens Jensen: (10:27 AM) Diamond light source next door use Lustre Christopher John Walker: (10:31 AM) The wife of one of the Lustre guys is doing a course at QMUL. - should invite him along for a coffee. He's Italian. Jens Jensen: (10:35 AM) Italian coffee... BTW Sanger are also running Lustre Christopher John Walker: (10:35 AM) https://jira.hpdd.intel.com/browse/LU-1482 Jens Jensen: (10:48 AM) MICE found the problem with the robots.txt file: bots were crawling all over the HTTP access points. wahid: (10:49 AM) Actually now reading more carefully the mandate - I don't object as long as its made clear its about certifying a site as supporting http for wlcg instead of mandating all sites wahid: (10:49 AM) Actually now reading more carefully the mandate - I don't object as long as its made clear its about certifying a site as supporting http for wlcg instead of mandating all sites Henry Nebrensky: (10:52 AM) Sorry - got to run. What technology is used by the CERN public portal to LHC data? HTTP could cover the open public access case