Attending: Dan, JohnH, Winnie, Raja, Sam, Simon, Steve, Teng, Jens (chair+mins), Brian, Matt, Luke Sam, Raja, Steve, Rob, and Teng were at the DUNE singalong (the DUNE TUNE?) There is a DUNE test instance with large datasets - using FNAL's SAM catalogue. SAM+Rucio => both are tracking data locations, Rucio is getting support for metadata as KV pairs, not super-sophisticated but it's there. Plan to migrate to Rucio, in two steps. Rucio+DIRAC => integration? Need for workflow workload, e.g. GlideInWMS Based on OSG federated authentication, Data lakes, and DOMA technologies. Currently run stuff at CERN with participation of GridPP sites, notably Manchester with 1PB; also Lancaster and Imperial are involved albeit currently with only a few TB. There is a 1PB quota allocated at RAL, but RAL offers only an S3 interface, not xroot, and it seems to put a file size limit at 5GB, where DUNE have 8GB files. There is a ticket against davix. Also, DUNE could get involved with DOMA. The ATLAS view - discussing Simon's options - is that some data caches are better than others, in the sense that some files are reused and others not. Simon's only buying 0.3PB which is not enough to upgrade the whole SE. Ultimately, it's either running the new storage as a second SE, or put it in the old one through the usual processes. Using HDFS as a backend would not work with DPM DOME. However, SRM-less access _should_ be OK for ATLAS. Discussion of IRIS WGs postponed till next week. hi Steve. you could just stop your video broadcast. DP Ste you tell it to us, sam... Sorry about the broad cast .... there was a dog assaulting me with licking... Called Tinker... SJ Daniel is there a link for the talks that non dune people can access DP Teng https://indico.fnal.gov/event/18094/other-view?view=standard Can you access this? T Samuel https://indico.fnal.gov/event/18094/ SC Daniel yes DP Yes, those links work, thanks. Brian So does that mean though DUNE still can't use RAL due to 5GB limit? email alastair regarding endpoint for DOMA-TPC endpoint BD Matt Ask me again in 60 minutes. MD Brian DUNE will likely follow CMS as both are heavily FNAL dominated. BD Raja Yes - DUNE still cannot use RAL, until FTS is instrumented to do the multi-part transfers R Samuel Certainly, there were a lot of CMS people in the room at the DUNE meeting, but it really wasn't that clear cut, Brian. There was definitely interest in non-CMS solutions SC Brian It does not make sense for RHUL which has been stated not ot be one of the large T2s we are going forward to with storage to just to buy extra capacity for current soluiotn BD You mean not a T2D..? Daniel I think the argument is that RHUL are too big to be just a cache. DP 'cos storage can also cache etc Brian NO site is too big just to cahce, just need to change job profile sent to site. think of site as a HPC.. BD Raja Apologies - got to leave now R Bye Raja; thanks Daniel dpmlite on top of hdfs, DP Simon Brian, we currently do the most ATLAS analysis in the UK, is this compatible with an HPC-style cache-only site? SG Luke POSIX is not a strict requirement, but users do prefer it L Brian aim is to get to a system that is fewer components to make sit elighter wieght surely? not just running DPM in DOME mode just with http endpoint???? my 2p worth , decom DPM and just put an xCache on HDFS? good fo reveryone? BD Luke It would have my vote ;) L Simon Interesting thanks Brian SG Daniel also I would check with Brum about EOS. Oxford and Durham with glusterfs. DP Brian it runs as "Volatile storage" for ATLAS BD Daniel but glusterfs would requires dpmlite ontop DP Brian UCL, BHAM BD Ste Re: "to get a system with ..." fewer moving parts, one often has to try out a lot of systems that don't work out. We need to do that to find the few systems that do work better and deliver what they promise, in a field where 9 things out of 10 soon disappear without a trace. SJ Samuel hi Ste: yes, this is quite true, and a thing which is a big problem for "moving forward" SC Simon Thanks everyone. SG (document URL) Ste Thanks Jens. I've done some work for the T2 forward look (skills section) so I would like to listen in. SJ Brian I though t this doc was WLCG limited, ie not worrying about storage outside WLCG? VOs want to do data managament themselves so providing rucio is enough? non WLCG VOs really aren't in scope for thsi document# BD Paige Sorry, gotta go PW Today at 11:01 AM