Attending: Wahid, Sam, Elena, David, John H, Robert, Steve, Rob, Matt, Duncan, Jens, Brian, Chris, Raja, Ewan, Raul, Pete Apologies: Dave 1. Preparations for the DPM workshop next week http://indico.cern.ch/conferenceDisplay.py?confId=273864 Wahid is presenting the UK summary, and will try to have some slides early, so people can give feedback &c.; conversely, people can send stuff they want to raise. Tutorials are generally considered Useful(tm), so technical presenters should aim to have not just slides but also technical content, particularly if there's stuff you can take away. Fabrizio is generally responsible for the technical side of the agenda. Obviously things which don't exist yet cannot be done in a hands on way. People can still register - video will be available but being there in person is Useful(tm). 2. Any last changes from the VOs for 2013... ding ding! Last orders, please! 2.1. So DPM sites should upgrade - if they haven't - to 1.8.7... supporting ATLAS' use of WebDAV. Last remaining 2-3 sites. What happens if not all disk servers are updated, and the head node advertises itself correctly as 1.8.7 - can the non-upgraded disk servers be a problem? Obviously it depends on the type of upgrade, but maybe baseline requirements should include the minimal acceptable version also for the disk servers. 2.2. A StoRM upgrade is needed but is more an efficiency issue and depends on a new version of StoRM being released, so this one can safely be parked until 2014 rolls along. 2.3. ATLAS space tokens - no more moving stuff around is needed at this point. 2.4. LHCb: need xroot, and on a longer timescale, federated xroot. RALPP already provide xroot, but not Manchester - but there is a general problem with DPM's xroot for LHCb (which is not seen by either CMS or ATLAS) - Andrew McNab knows what it is. Here's an issue for Wahid to mention next week... 2.5. Snoplus - need more storage. What is the procedure for asking for more storage, and, more generally, what is the procedure for growing beyond small generic-space VO to larger tokened-space VO? T2K took some effort also on our part to move [as an aside, T2K don't seem to be using the space token at Oxford, but this seems to be an Oxford issue rather than a generic T2K issue; see chat.] Should we request that Snoplus use space tokens? Depends on their usage patterns; and ultimately the non-LHC use of GridPP resources is limited. Perhaps it is worth documenting this process (adding it to the documentation for non-LHC VOs). In the case of Snoplus, the usage seems to be storing job output prior to squirreling it away at RAL, so the space required would need to be aligned with the number of job slots. In the short/medium term, 2TB should be sufficient. The conclusion of the discussion was that it seems worth moving Snoplus to space tokens, and to give them a TB or two in the token... in which case it'd better wait till 2014. So there is a risk that they run out of space when they work on their Christmas simulations, but setting up space tokens now would be a bigger risk. 5. AOB Matt asks if others are seeing high load on disk servers from xroot. Sam speculated that it was a new kind of job causing a higher load. Brian points out that with sparse reads, the number of open connections will become a much more interesting measure, so maybe it is worth taking this into account in DPM monitoring (next week) as well as for other instances. [09:58:33] Wahid Bhimji hi [09:59:18] John Hill Morning [10:01:46] Jens Jensen http://indico.cern.ch/conferenceDisplay.py?confId=273864 [10:03:59] Jens Jensen Wahid, have we lost you? [10:04:06] Sam Skipsey Wahid, are you there? Step into the light! [10:04:41] Matt Doidge No step *away* from the light! [10:05:22] Matt Doidge (it's probably an oncoming train). [10:06:02] Brian Davies info on how atlas did mass rename using webdav would be intresting [10:06:37] Christopher Walker Wahid, we can hear you - [10:07:01] Matt Doidge Always interesting in tutorials. [10:07:08] Matt Doidge *interested [10:07:14] Sam Skipsey It might be easier if people type things they want Wahid to respond to, as his audio seems to be unreliable [10:07:36] Wahid Bhimji are you talking about tutorials in the workshop ? the agenda is there.. so unless you mean over lunch [10:13:53] Wahid Bhimji forget about publishing [10:14:17] Christopher Walker https://ggus.eu/ws/ticket_info.php?ticket=99401 gfal2 dav [10:15:11] Ewan Mac Mahon My EVO seems to have taken to mysteriously trying to turn the camera on every time I join a meeting until I turn it off. Not sure why. [10:17:28] Wahid Bhimji OK so they will have to run another xrootd server.. are there instructions . [10:17:50] Christopher Walker SNO+ were asking for a bit more storage - they were worried about running out if they get a batch of jobs through. [10:18:28] Wahid Bhimji it works Now ! [10:19:00] Elena Korolkova Can sno set up space tokens? [10:19:11] Wahid Bhimji what is it [10:19:17] Wahid Bhimji if you can tell us we can help [10:19:24] Wahid Bhimji what is it ? [10:19:31] Christopher Walker https://www.gridpp.ac.uk/wiki/WebDAV [10:19:43] Christopher Walker Also does xrootd [10:20:01] Wahid Bhimji forward the ticket to storage list - that would be great - thanks ! [10:20:22] Christopher Walker Can sites fill in status [10:22:16] Wahid Bhimji in practise it means if they lcg-cp in they have to put a --st option [10:24:06] Ewan Mac Mahon I don't believe anyone's got SNO+ space tokens as yet. [10:24:15] Ewan Mac Mahon Some people have t2k ones, some don't. [10:24:23] Sam Skipsey Mainly because people didn't have an idea of how much space SNO+ wanted. [10:24:32] Sam Skipsey Or that they might need reservations/limits [10:25:08] Ewan Mac Mahon There is an issue here that (DPM) space tokens are both reservations and limits. [10:25:21] Sam Skipsey Well, that's a deeper SRM issue, Ewan ... [10:26:23] Elena Korolkova You can reserve 5 TB for SNO [10:26:38] Elena Korolkova and then increase it if needed [10:28:46] Ewan Mac Mahon If we're just talking file staging, then that's likely to be quite steady-state, so suit a space token quite well. [10:29:22] Duncan Rand It depends how much DPMs have unreserved. A few TB doesn't sound much. [10:29:40] Sam Skipsey What we're recommending is that SNO+ think about how much space they need. [10:29:49] Sam Skipsey We can't really do that for them. [10:30:24] Steve Jones Yes; give them 1 month, then scrub the files. [10:30:36] Ewan Mac Mahon Indeed. I think we're recommending that they use a space token, [10:30:56] Ewan Mac Mahon but that they need to carefully(ish) think about whether that's suitable, [10:31:06] Ewan Mac Mahon and if so, how big they want it at each site. [10:31:06] Steve Jones Yes, use a space token where files only remain for 1 month, then chop chop charlie. [10:31:28] Raja Nandakumar Apologies - need to go to another meeting. [10:31:30] Steve Jones They can use it or do something else - their choice. [10:32:21] Ewan Mac Mahon Policy is: [10:32:27] Ewan Mac Mahon Tiny VO: we don't care [10:32:42] Ewan Mac Mahon Medium VO: you can have the space, use a spacetoken [10:32:47] Ewan Mac Mahon Big VO: Give us money. [10:32:53] Ewan Mac Mahon (Roughly, and IMO) [10:33:12] Ewan Mac Mahon You sound like Kenny, Sam. [10:33:20] Ewan Mac Mahon SLightly [10:33:54] Steve Jones Or: Big VO - pay for it, Medium VO: use the "1 month scratch disk or pay for it" [10:33:55] Raul Lopes good rough policy [10:34:20] Ewan Mac Mahon I'd go with yes. [10:35:04] Matt Doidge Agreed, especially as it also gurantees SNO+ the space. [10:35:11] Steve Jones Someone should write up some proposals, and put it to a vote next week! [10:35:28] Ewan Mac Mahon Particularly the ATLAS site have a lot of space in reservations, so the unreserved space is usually small and potentially unreliable (because it's easy to fill) [10:36:14] Sam Skipsey Well, we stick the non-ATLAS stuff in their own pool to protect them from that, Ewan, but yes, in general [10:38:22] Ewan Mac Mahon I'm also assuming that we're not looking to migrate existing data into the token, we'll just do it by natural turn over. Does that sound OK? [10:38:28] Elena Korolkova It's not a big deal for sites. it's more load for vo [10:39:23] Matt Doidge We have 16TB used out of 16TB space for the T2K token at Lancaster. [10:39:32] Elena Korolkova T2K are using spacetoken in Sheffiled [10:39:36] Ewan Mac Mahon I think they are at some places, but there was a problem at Oxford that they worked around by not using the sapce token, rather than telling us so we could fix it. [10:39:57] Brian Davies lcg-infosites --vo t2k.org space | grep T2K 0 17592 17592 0 0 0 T2KORGDISK fal-pygrid-30.lancs.ac.uk 74056 38959 113015 0 0 0 T2KORGDISK gfe02.grid.hep.ph.ic.ac.uk 2884 7010 9895 0 0 0 T2KLIVERPOOLDISK hepgrid11.ph.liv.ac.uk 100 0 100 0 0 0 T2K_DEFAULT heplnx204.pp.rl.ac.uk 999 0 1000 0 0 0 T2KORG_TOKEN se01.esc.qmul.ac.uk 8346 91653 100000 0 0 0 T2KORGDISK se03.esc.qmul.ac.uk 4616 10767 15383 83415 352889 518937 T2KORGDISK srm-t2k.gridpp.rl.ac.uk 938 0 938 0 0 0 T2KIFAEDISK srm.pic.es 1 5799 5800 0 0 0 T2KORGDISK srm.pic.es 1 1 1 1 1 1 T2KIFAEDISK srmatlas.pic.es 1 1 1 1 1 1 T2KORGDISK srmatlas.pic.es 1 1 1 1 1 1 T2KIFAEDISK srmcms.pic.es 1 1 1 1 1 1 T2KORGDISK srmcms.pic.es 938 0 938 0 0 0 T2KIFAEDISK srmifae.pic.es 1 5799 5800 0 0 0 T2KORGDISK srmifae.pic.es 1 1 1 1 1 1 T2KIFAEDISK srmlhcb.pic.es 1 1 1 1 1 1 T2KORGDISK srmlhcb.pic.es 0 0 0 0 0 0 T2KORGDISK srmv2.ific.uv.es [10:40:13] Elena Korolkova We have 25 TB in Sheffield (because they gave us mone for disks) [10:41:37] Ewan Mac Mahon Our 't2k' space token definition includes this: [10:41:40] Ewan Mac Mahon Authorized FQANs: t2k [10:41:47] Ewan Mac Mahon Which I believe eas the problem. [10:42:03] Ewan Mac Mahon IIRC we'd decide to just delete it and start again with t2k [10:43:04] Brian Davies ah oxford is appearing when specifying the VO as " t2k" and not "t2k.org" [10:43:48] Elena Korolkova We have in Sheffield [10:43:56] Robert Frank manchester as well [10:44:12] Christopher Walker QMUL has high load on some of our disk servers [10:44:38] Christopher Walker I blamed a local user, but we are running lots of analysis jobs