Attending: Brian, Daniel, Duncan, Gareth, Marcus, Winnie, Sam, Steve, Tom, Jens (chair+mins), Elena, Ewan, Matt D, Robin, Luke, Govind Apologies: John H, Raul 0. OPerational blog posts Some discussion of GLUE2 - ticket for RAL to provide GLUE2 for CASTOR which may be an opportunity to abandon InstalledCapacity which meant RAL wasn't publishing the dynamic accounting data the experiments expected. Related to this, it would appear that lcg-infosites uses information different from the BDII's, ie it fetches it from the SRM? Which is slightly odd because you need to authenticate to the SRM but the BDII is free... 1. Update on T2C and T2D - in view of mail from Dave to tb-support LHCb should be OK; ATLAS and CMS testing at Oxford. Hammercloud might break. Wrote data to Glasgow as well. Direct IO is new but other stuff is well known. Infrastructure testing, done ad hoc, not fully. For comparison, not a lot of ATLAS stuff is currently done using FAX so relying more on remote IO is relatively untested. How are sites picked? The CMS AAA scale testing had problems at RAL and UK overall not performing so well. With information on secret CMS web sites, who are the right persons to talk to? Nevertheless, maybe scaling for CMS is less worrying than ATLAS trying the same thing. 2. Update on the automated consistency checking (ATLAS, perhaps, in particular?) ATLAS have started ticketing sites. Need relative path length, but relative to what - to what ATLAS think is the (grid equivalent of) mount point. Need to find out how to deal with GROUPDISK. Matt at Lancaster also testing uploading the file list. Quite a lot of work for a T3... could a T3 say no? When the space is just scratch space, the official list of files would be an empty list anyway. 3. Issues so far for the DPM workshop? http://indico.cern.ch/event/319828/ Still open for feedback and issues but Sam will start writing slides. Raul's Ansible config may be of interest. The GDB is the following day and will contain a summary of the workshop. 4. DiRAC backups revisited (in view of Brian's blog post and CASTOR complaints) Too many small files, stressing the nameserver database and the transfers are inefficient due to overheads. Some discussion on alternatives, e.g. using tarballs. Are the requirements fully specified? [Discussion continues on dirac-users list] 5. AOB Hello (again?) to Marcus who usefully also has an LSST hat. LSST have some of the same problems that DiRAC have seen, with expiring proxies for example. Alessandra is working on a proposal/test/solution (delete as applicable) for LSST based on DIRAC (the LHCb one). Worth presenting something at Networkshop next year? Slightly expensive, though. Ewan Mac Mahon: (25/11/2015 10:06:53) It's underway; I think that's about all there is to say about it atm. Latest status is that Alastair is setting bits up in ATLAS, once he's set them up we can start thinking about running tests. I think we're probably working on two things here really - one part is some technical testing, the second part is essentially conciousness raising among the VOs that the UK is rather proposing to actually do this, and they'd better either get used to it or start screaming about it. Jens Jensen: (10:12 AM) ... and whether it scales to when everyone starts doing it Elena Korolkova: (10:13 AM) RHUL? Lukasz Kreczko: (10:13 AM) CMS as fallback-stage-out and xrootdfallback Elena Korolkova: (10:13 AM) They have ~1400 TB Daniel Peter Traynor: (10:13 AM) UCL now points to qm se03. Do we have a feel for how this is working out Ewan Mac Mahon: (10:16 AM) I think there is some merit to the idea that this isn't a completely radical creation, it is indeed based on a lot of parts that exist, we're just joining them together in slightly different ways. It's not an 'is this possible' stage test, it's an 'is this practical' stage test. A beta, not an alpha. Duncan Rand: (10:20 AM) https://indico.fnal.gov/getFile.py/access?contribId=33&sessionId=17&resId=0&materialId=slides&confId=7207 Ewan Mac Mahon: (10:24 AM) Jira, rather than AGIS? The sites might just tell them where they can stick it. Matt Doidge: (10:30 AM) https://ggus.eu/index.php?mode=ticket_info&ticket_id=117883 Jens Jensen: (10:33 AM) give them a zero length file and tell them "this is your file dump" http://indico.cern.ch/event/319828/ Samuel Cadellin Skipsey: (10:37 AM) https://indico.cern.ch/event/432642/other-view?view=standard Ewan Mac Mahon: (10:39 AM) We should just get you a t-shirt printed with a Puppet logo inside a crossed red circle, that would probably save you some time. Duncan Rand: (10:39 AM) I see Performance tests of DPM sites for CMS AAA 25' Ewan Mac Mahon: (10:41 AM) I think someone might have mentioned tarballs. Several times now.