Attending: Chris B, David, Elena, Gareth, Jeremy, Matt, Robert, Brian, Jens Apologies: Everyone at CHEP... Brian: dark data Alastair identified in scratchdisk. Cleaning up 3.5 M empty directories at T1. Rucio should no longer be creating these? Used to give problems if you had 32K directories in a directory (at least on ext2) and of course if you run out of inodes. Having a dir/XX/YY/stuff postpones the problem of having too many subdirectories, but it doesn't solve it. dCache at T1 - didn't test CDMI. Chris: dCache ready to go in the new machine room: on 2.2. Will deploy xroot, WebDAV, and maybe CDMI and other goodies once it's up. David and Gareth: upgraded to 1.8.7; aiming to complete SL6 upgrade on WNs by end of October. Elena: problem with people actually using available space(!), described on the list. Could we bring snoplus in and have a word with them? Also doesn't exactly help that they have large files, about 1TB. How would we do quotas without SRM (and SRM space tokens)? dCache has a means of doing this, assigning space tokens to directories, although this can confuse users who look at the information systems, because they may be using space in a token without realising it. Nevertheless, seems like a useful feature if used in moderation - maybe a feature request for DPM. Ewan: on 1.8.7, mainly the problem with /dpm which causes WebDAV to not work. On EMI3 with SL5, not planning to move - new kit would get SL6, but probably not the old kit. Govind: EMI3. Planning to upgrade to SL6. Jeremy: GFAL->GFAL2 expected "early next year" GLUE->GLUE2 delayed by OSG not supporting GLUE2, maybe Q2 14. (also changes priorities for RAL CASTOR) CHEP had many storage-and-data-management-related talks yesterday afternoon: http://indico.cern.ch/conferenceTimeTable.py?confId=214784#20131015 Matt: Like Ewan, not in a hurry to upgrade to SL6. 1.8.7 still planned, maybe this week. Robert: DPM upgrade went well. SL6 and EMI3. Were already on SL6 beforehand. Will provide more resources for LHCb, another 90TB. Jens: talking storage with EGI FedCloud, would need CDMI interfaces. dCache getting CDMI "improved" so must be important for dCache, too. Also iRODS is getting a CDMI interface, in a separate development. However, the standard EGIFC implementation of CDMI is python based. Also playing with WebDAV. Talked to industry partners about data security. Mostly integrity for us, unlike others for whom confidentiality is top. Of course we have embargos and stuff to let people publish their results, but broadly publicly funded data should be open (cf RCUK data policies.) [10:06:06] Jens Jensen Sorry I am late - my laptop crashed [10:22:50] Ewan Mac Mahon Might be worth checking who's DN has actually been writing the files too. [10:22:58] Ewan Mac Mahon Just in case it's not him.