Attending: Brian, Gareth, Jens, John B, Steve, Matt, Sam, Wahid, Ewan, Elena, Matt RB 0. Operational issues and blog posts (currently we have four things) 0.1. Brian believes that the QMUL WebDAV issue could be the same as one seen by INFN recently, which may also focus the developers's minds on the priority? 0.2. Wahid asks about missing disk servers at RAL, and the time to recover: Brian reports they are "temporarily unavailable", but at what point are they unavailable and at what point should they be declared lost? If the files are not coming back, replicas need to be found (or they should be recovered from tape, if tape backed) - is there any way to decide whether the files are "urgently" needed - here it's 290,000 files, some files also from groupdisk which may not have replicas at other sites. Previously we had the concept of "suspicious files" This is also what federated storage was supposed to be for, but RAL is not in FAX with fallback to the T2s because a T2 might not enjoy being asked for (say) 50,000 extra files, or having its (shared) bandwidth saturated. 1. Summary of the storage-related stuff from hepsysman? Probably not that much, after all... Brian suggests the most useful part of the security training is the ability to compare "normal" storage to what you see, if something dodgy is going on - eg dCache may start lots of Java processes which are neither malicious nor (hopefully) harmful. 2. GDB is today: http://indico.cern.ch/event/272622/ One notable thing (storage related, sort of) is DPHEP. EUDAT is also using Invenio (called B2SHARE in EUDATspeak) Also noted is the MW "readiness procedure" - DPM 1.8.9 is one of the releases; even if individual components will be, er, individually versioned, 1.8.9 is thought to comprise a certain set of such components with minimal versions - so whichever release you run is either the latest of everything, or at least it needs testing, but then we test stuff anyway before running in prod'n. The tests will be usual ATLAS FTS and Hammercloud - Edinburgh participating. The HEPIX summary highlights the use of CEPH, end of life of SL7. For the latter CentOS is thought to take over, indeed some sites are reported to already be using CentOS. While in the past middleware has been amazingly unportable - having trouble porting to the next version of its own OS, for example - EPEL *should* help, and things should be better now, particularly with non-YAIM configurations. CERN will have a SIG on CentOS (CERNTOS??) The safe path is probably to use "whatever CERN is using" but more options should be possible which is a good thing. Pre-GDB on SL6: http://indico.cern.ch/event/272788/ 3. MSST has just finished, here's the link to the presentations: http://storageconference.us/2014/index.html#Program Lots of interesting stuff; Sam points out the Shingled Storage 4. AOB Jones, Steve: (11/06/2014 10:02:49) good thanks Jens Jensen: (10:11 AM) http://indico.cern.ch/event/272622/ Which is unlike maths where people use the same word to mean different things - like "normal" for example. Ewan Mac Mahon: (10:23 AM) Isn't the WLCG baseline for DPM basically 'keep everything up to date all the time'? In practice. Jens Jensen: (10:28 AM) Or get the file back off tape? Ewan Mac Mahon: (10:45 AM) I'm not sure we need to worry if someone's running SLC, SL, Red Hat or CentOS. Most things work fine on all of them. The very few things that break are things like YAIM scripts looking for specific version strings, and that's usually fairly easily fixed. That's mainly because there's very little other difference. wahid: (10:48 AM) sorry didn't realise I was unmuted Matt Doidge: (10:48 AM) np wahid: (10:48 AM) someone asking for Andy.. he's visitng you glasgow guys Jones, Steve: (10:48 AM) We ran grid job of RHEL for a bit. It worked OK. CENTOS would be fine. Gareth Douglas Roy: (10:49 AM) @Wahid Yup Andy's sitting behind me :) Ewan Mac Mahon: (10:50 AM) And we've got a full blown RHEL support contract/site licence. I'm not in any hurry to use it for everything, but I wouldn't rule it out.