Minutes of the storage EVO meeting, 21 July 2010 Present: Glasgow: Sam Lancaster: Matt Liverpool: Stephen Bristol: Winnie Sheffield: Elena QMUL: Chris W RAL: Brian, James, Jens (chair+mins) Apologies: Edinburgh: Wahid Birmingham: Chris C 0. Review of actions - long overdue, quick review 1. Post WLCG workshop discussion See the agenda here: http://indico.cern.ch/conferenceOtherViews.py?view=standard&confId=82919 Like Amsterdam, lots of discussion re caching at T2s. Data moved when it is wanted. Plots of access of files show lots of files being accessed infrequently, and few files being accessed frequently. Replicate files which "look hot". ARC discussed, the cacheing mechanism. Some sites are not providing the resources required. When sites get resources, they should publish them immediately. Overall a feeling of "happiness but more could be done to optimise." Schism between "file" (ie NFS4) and "xroot"ers. Move away from using tape as working repository; use instead as backup and archive. There was some discussion about taking over whole WNs, ATLAS and CMS seem to have suggested this. This makes sharing resources more difficult, eg if you have local users. There was discussion about virtualisation and a report from the HEPiX WG, courtesy of Tony Cass. Also talk about CVMFS - CERN VM file system, by Rod Walker. What are the implications for the T2s? More bandwidth? Need to publish resources immediately. We should still follow up on our demonstrators, despite them not being demonstrated at the workshop. Sam says the replica resilience is being built into lcg-cp, according to Zsolt Molnar. 2. Current versions of middleware revisited (again) Sam has tested the DPM 1.7.4-7 release. Sites should not upgrade to 1.7.4 unless they downgrade VOMS - there is a problem with the VOMS API RPM: apparently the API is not thread safe. dCache - Brian and Rob evaluating fresh install. 3. Writeups revisited What happened to the file integrity? [Jens] Hadoop? [Brian] Plans - orphaned files [Brian again] 4. AOB Sam discovered reason behind anomalous load on pool nodes: datasets unbalanced, causing reads to focus on that pool. Sam is currently writing a tool to analyse this. It may be feasible to use ext4: there is some evidence that ext4 behaves better than xfs under high load, but there is not much difference under normal load. Current list of actions 322 15/04/2009 Replicate Glasgow DPM database indexing at Oxford Ewan Low Open No news, none at all... 367 24/02/2010 Add alternative inputs to script (eg list of files) Sam Low Open Not needed by ATLAS any more - Cedric ran Cedric's tool to do the ATLAS checking. Make this a task in Savannah instead. 368 24/02/2010 Look into a local StoRM dumper (analogous to dpmDump) Wahid Low Open Wahid is not present... 382 07/04/2010 Document samplemathics Jens Med Open Done. Probably a note is needed in the File Integrity Testing wiki page - also needs Brian's numbers. 388 28/04/2010 Clean up wiki front page (the storage one presum.) Sam Med Open Is this part of 401? Make it so. 391 05/05/2010 Investigate StoRM metrics for QMUL Chris+et al Med Open No news? (Chris cunningly joined late, after we'd done the actions) 397 19/05/2010 Upload workshop writeup and circulate to relevant PMB members Jens Med Open Will be done this week as part of the QR. 398 26/05/2010 Check whether list archives are 600, 640, or 644 Jens Med Open Did it just now - archive seems to be world readable, so 644. 399 26/05/2010 Test DPM 1.7.4-6 on SL5 and report Sam Med Open Done, but writeup not written up - see item 2 on agenda. 400 26/05/2010 Investigate upgrading B'ham to SL5 Brian+Chris C Med Open Chris reports hardware obtained, but problems with draining filesystem on head node. 401 02/06/2010 Clean up the wiki ALL Low Open Action! == CHAT == [10:00:47] Winnie Lacesso joined [10:01:39] Elena Korolkova joined [10:05:44] Ewan Mac Mahon joined [10:08:45] Jens Jensen http://indico.cern.ch/conferenceOtherViews.py?view=standard&confId=82919 [10:09:01] Queen Mary, U London London, U.K. joined [10:12:18] Queen Mary, U London London, U.K. Bloom filters mentioned as a possible way of publishing files available on an SE [10:28:55] Queen Mary, U London London, U.K. Private conversation indicated that gridFTP piplining was in the works for FTS [10:33:02] Elena Korolkova SAM, I've missed. You do not reccoment latest DPM upgrade, do you? [10:34:31] Jens Jensen Don't upgrade to 1.7.4 unless you downgrade VOMS [10:35:08] Elena Korolkova we are running 1.7.4 since april [10:35:16] Jens Jensen If you have no problems then fine [10:35:45] Elena Korolkova I meand the latest (3 weeks) ago update which was announced [10:38:38] Elena Korolkova left [10:38:39] Stephen Jones left [10:38:41] Winnie Lacesso left [10:38:41] Sam Skipsey left [10:38:42] James Thorne left [10:38:45] Ewan Mac Mahon left [10:38:45] Brian Davies left