Attending: Jens, Wahid, John B, Chris W, John H, Ewan, Brian, Raja, Duncan, Matt, Steve, Ewan 0. Operational blog posts That there VOMS thang. High t'put of data at L'pool. Overloaded servers: not throttling in xrootd, obviously; this feature is still being discussed and is not available yet. And is the load actually necessary? For example, instances have been seen where the clients had a config problem which made them use GridFTP where they should use direct IO. Almost no data being written, so it is not due to writing. Seen for ATLAS but also biomed. Sam might have options to help - some workaround, limiting threads or something. dCache - proxy handling problem seen with Brazilian CA, due to the issuer name being UTF8 encoded (instead of printableString). Raja sent detailed to Jens separately and Jens confirmed it wasn't a problem for UK CA. dCache have a patch to work around the problem in the jglobus lib. A combination of three different bugs causing the problem. 1. Update on WebFTS testing (if any) Didn't work with Brian's ATLAS magic but it worked for Raul. How do we discover endpoints? In GO they are just advertised using simple metadata strings. 2. Did anybody manage to do some Real Work. in August? August is supposed to be a "quiet" month, so should leave you time to do something Interesting (in theory). Ewan's Argus stuff is available but only used to ban (and unban) users: DPM queries user DNs regularly to ask whether they are banned (resp., unbanned). Not used with CASTOR: CASTOR itself is fairly simple authentication-wise, but the SRM (and GridFTP) could be made to call Argus (in theory, at least). Otherwise the CASTOR user database does not hold DNs. Brian's "longitudinal" study of file sizes on disk. Still trying to get permissions for WebDAV access to delete empty directories. 3. AOB? Was there any DPM suggestions for things to take to the DPM workshop? (Sam || Wahid) will attend and is expected to give a UK summary. It's in Naples which may not be super-easy to get to from every airport in the UK. xroot stuff id developed in the US but DPM could put in a throttle feature, so it may be in scope for the workshop. Reminder that today's GDB has an entry on data management. http://indico.cern.ch/event/272777/ Yesterday's pre-gdb didn't seem to have anything data related. At some point I will try to give a summary of the (past) conference stuff - data related stuff. Also related to OGF, there's some updates to GLUE (to support clouds) and accounting - which I will report later (because I haven't spent enough time looking at it yet.) And we need to revisit the wiki (again): I couldn't find a single data-related change recently: https://www.gridpp.ac.uk/w/index.php?title=Special:RecentChanges&days=14&from=&limit=250 BTW note that CERN will be getting rid of the old Savannah srmsupportuk. ------------------------------------------------------------------------ wahid: (10/09/2014 09:57:47) https://indico.cern.ch/event/336753/ https://indico.cern.ch/event/336753/program https://indico.cern.ch/event/336753/program https://github.com/xrootd/xrootd/pull/22 Duncan Rand: (10:07 AM) I don't see much here http://dashb-atlas-data.cern.ch/ddm2/#date.interval=10080&dst.cloud=%28%22UK%22%29&dst.site=%28%22UKI-NORTHGRID-LIV-HEP%22%29&grouping.dst=%28cloud,site,token%29&p.type=rat&tab=transfer_plots those were inbound. This is outbound: http://dashb-atlas-data.cern.ch/ddm2/#date.interval=10080&grouping.src=%28cloud,site,token%29&p.type=rat&src.cloud=%28%22UK%22%29&src.site=%28%22UKI-NORTHGRID-LIV-HEP%22%29&tab=transfer_plots Raja Nandakumar: (10:10 AM) No.... Thanks! wahid: (10:12 AM) I want Brian to test it ;-) Is there a link by which I can just try it Ewan Mac Mahon: (10:13 AM) I looked at it briefly, it wasn't obvious what to do to actually transfer a file, and I've not got around to going back to it. It's certainly missing the 'big friendly button'ness of GO. Brian@RAL: (10:13 AM) https://webfts-test.gridpp.rl.ac.uk:8446/# wahid: (10:13 AM) Being obvious what to do .is one of the key requirements I would say John Bland: (10:14 AM) duncan: yes, from the logs it looks like biomed is the majority of the gridftp traffic, but I think that's just a background issue, the xrootd overload is what causes the meltdown and that is atlas local transfers wahid: (10:14 AM) well it has a big "Submit a transfer button Raja Nandakumar: (10:15 AM) Just remembered a small issue with dCache - when you have time. Ewan Mac Mahon: (10:16 AM) I'm getting a friendly blue button in the top right that says 'loading proxy...' Shouldn't it have loaded one by now, theyre not very big. And now the webfts pages won't load at all. Brian@RAL: (10:26 AM) try turning on web console to see th error Duncan Rand: (10:28 AM) John: are those xrootd reads grid analysis jobs? Are they staging to the WN or 'direct access'? Raja Nandakumar: (10:30 AM) Apologies - joining another meeting now. wahid: (10:31 AM) https://indico.cern.ch/event/324705/timetable/#20141009.detailed John Bland: (10:32 AM) duncan: as far as I can see it's direct access and it's atlas pilatl jobs. It's also not a hot file issue, when there are hundreds of connections it's to hundreds of unique files might be able to tune some of it away, but ultimately the arrays only have so many IOPS Ewan Mac Mahon: (10:33 AM) And on the continuing webfts testing, the oage loaded, so I clicked 'submit transfer' popped a URL (an https one) frommy SE in the lefthand endpoint path box, clicked load, it says it's loading. It's been saying that for a while now, and there are only four files in that directory. wahid: (10:34 AM) yes Iwill do something Duncan Rand: (10:34 AM) You could cap the number of analysis jobs...not ideal but 157 files for job 2261768033 http://panda.cern.ch/server/pandamon/query?job=2261768033