Attending: Brian, Jens, John B, Robert, John H, Wahid, Tom, Gang, Rob, Ewan, Raja, Sam, Steve, Matt, David, Chris, Elena Congratulations: Wahid 1. What is this thing N2N anyway and why should it choose to collapse on Betelgeuse 7? It used to be an ATLAS thing to map global names into local names (ie what we used to call LFN to SURL. Today, however, the RUCIO translation is deterministic and an algorithm can do the lookup that previously a lookup did. Most sites have RUCIO renamed - and N2N has been tested by John B on SL6, and at ECDF. Minor niggles that sites occasionally fail FAX. 2. Feedback from TF-Storage (Terena) Jens had attended Terena's storage task force and gave a summary. Discussions about the merits and proper price point of SSDs (and Hyperdrive) - maybe it is now worth re-evaluating the previous tests. Would it be useful, for example, to have SSDs in a WN? Some workflow might be better supported with SSDs. There was a lot of content about SWIFT and Cinder, the object store and block store in OpenStack, respectively. SSD market is changing with pressure for higher densities, so expecting multi-level multi-level cells. If you look at it from the right angle, we have been doing stuff like this for, well, over a decade... also the identity management and authorisation which troubled the discussion of the Terena security-related working group. 3. Round table Finally got it mostly done - update on storage-related things people were doing, or it'd be OK to say "nothing special" if most things are not storage related. - Brian - dark data in ATLAS after RUCIO renaming, still empty subdirs which site admins will need to mop up - Chris - poking people about WebDAV, now usable. Also the "small" VOs. In RUCIO renaming, concerned that some files were absent but they'd just already been renamed - David - Nothing special (see chat) - Elena - EMI2->3 . DPM. LHCb transfer rates (to WNs) - fixed by replicating Liverpool's settings - Ewan - production to get bigger, will be >1PB. Aiming to puppet a new pool. Also the IPv6 DPM is still there for testing! - Gang - working on puppet recipe for gridmap - Kashif also working on this so should discuss - also needs to be in github and doc'd with DPM. - Gareth - test Hadoop - John B - testing hardware, aim to support some locals - John H - DPM upgrade; new SE, needs migration from the old one. - Matt - xrootd niggles. Perfsonar terrible, would follow up by email - Robert - nothing special - Sam - consistency checking. DPM-Argus problem - developers "forgot" - talking to Oliver (as will IanC from T1). - Tom - CERN@School with CVMFS - Wahid - WebDAV renaming, DPM testing fix, aim to hand over tests - Jens - LHCb publishing discrepancy understood and there's an easy fix (lie^W); also trying to use the new CVMFS-backed T1 cloud; trying to get CDMI service into EGI FC. [10:00:55] Tom Whyntie Congratulations! [10:01:15] Wahid Bhimji Kieran Bhimji [10:01:22] Wahid Bhimji 10lb 10ozs ! [10:05:57] Ewan Mac Mahon So atm we've got xrootd-server-atlas-n2n-plugin-1.1-0, so as long as it goes in with the same name and a higher number we should be OK. [10:09:35] Sam Skipsey I see they like their punny names. [10:11:17] Sam Skipsey I'm fairly sure that "cool name collisions" are now basically guaranteed. [10:11:36] Ewan Mac Mahon We need to do the NSA thing of just plunking two random words together. [10:12:03] Ewan Mac Mahon Which is sortof also the Ubuntu thing, now I think about it. [10:18:31] Sam Skipsey Named after the thunder god? [10:18:33] Sam Skipsey Interesting. [10:23:11] Ewan Mac Mahon And I was thinking that NewbieSave sounded like an easy-to-use backup system. [10:23:17] Sam Skipsey Hurrah! [10:23:44] Sam Skipsey (Hopefully I'll get my LFC version of that production ready this year) [10:27:39] Ewan Mac Mahon An SSD is better than a disk. I think it's still less clear that a certain £s worth of SSD is much better than the same £s worth of 15k SAS disks in a RAID. [10:34:29] Ewan Mac Mahon There are several things that might be obvious choices as an alternative to CIFS. iSCSI isn't completely clealy one of them. [10:37:47] Jens Jensen Yes, I found that slightly odd too - I thought it was for making availabel on the WNs in the same cluster. [10:37:52] Ewan Mac Mahon WOS? [10:40:38] Tom Whyntie Is this worth a news item? [10:41:14] Ewan Mac Mahon It's difficult to write anything that's not just going to sound grumpy. [10:41:50] Ewan Mac Mahon What Sam said. [10:41:56] Ewan Mac Mahon Nicely non-grumpy. [10:42:45] Tom Whyntie "An object store by any other name..." [10:43:15] Christopher Walker Technology behind the WLCG - a series... [10:43:18] Ewan Mac Mahon There's possibly an opportunity in the current hardware upgrades - i.e. wait until it's done, then annouce that the GridPP nationally distributed object store has reached a total of ${BIGNUMBER} in size. [10:43:23] Jens Jensen iSGTW [10:43:54] Christopher Walker In addition to the PBs of data stored, analysed and transferred. [10:44:26] Ewan Mac Mahon I think that's partly the sort of thing we should do a) talk about it more and b) when we talk about it, use the buzzword compliant terms, rather than our usual ones. [10:44:52] Sam Skipsey Yeah, some of this is a translation problem [10:45:03] Wahid Bhimji its going to take a long time for this rountable [10:45:06] Sam Skipsey (A lot of "new" stuff looks "new" because it has "new words") [10:45:12] Wahid Bhimji given we are 15 mins past due. [10:45:41] Christopher Walker http://www.chiark.greenend.org.uk/~sgtatham/agedu/ [10:46:53] Wahid Bhimji move on [10:47:14] David Crooks Sorry, audio problems - not much to report for me [10:47:29] David Crooks I'm more working on the general cluster upgrades [10:48:20] Wahid Bhimji WN tcp settings was that ? [10:48:23] Christopher Walker Oh yes, Steve's tests are now working at the tier-1 http://pprc.qmul.ac.uk/~lloyd/gridpp/nettest.html [10:48:29] Wahid Bhimji (sorry I missed what you said) [10:48:58] Wahid Bhimji you saw glasgows blog on dpm puppet [10:49:47] Elena Korolkova Also. I have a problem with overloaded disk servers. I drained disk servers and data are distributed evenly now. Things looks much better now. [10:49:55] Gang Qin https://www.scotgrid.ac.uk/wiki/index.php/Dpm_setup_with_puppet(no_yaim) [10:50:09] Ewan Mac Mahon @Wahid - don't think I did. Where is it? [10:50:18] Sam Skipsey Elena: how full ais your storage in general, out of interest? [10:51:51] Wahid Bhimji https://www.scotgrid.ac.uk/wiki/index.php/Basic_dpm_setup_with_puppet [10:53:16] Elena Korolkova 3% free [10:53:26] Sam Skipsey Thanks, Elena [10:55:54] John Bland rob's not here but he's working on general puppet stuff [10:57:18] Ewan Mac Mahon LFC-fsck ! [10:58:59] Gang Qin To Wahid: in https://www.scotgrid.ac.uk/wiki/index.php/Basic_dpm_setup_with_puppet, we are still using yaim to configure the users and make gridmaps [11:00:17] Gang Qin To Wahid: In https://www.scotgrid.ac.uk/wiki/index.php/Dpm_setup_with_puppet(no_yaim), we configure the users and make gridmaps with a new puppet module dpm [11:02:37] Gang Qin In the previous link, dpm master/disk are installed on one machine, while in the new link , dpm master/disk are spearated on different machines and the module will know the machine's role and take following actions [11:02:49] Wahid Bhimji yeah hanks for clairfiying - and thaanks for the excepllent write ups [11:02:57] Wahid Bhimji bye