Attending: Jens (c+m), Matt, Daniel, John H, David, Sam, Brian, Luke, Steve, John B, Winnie 0. Operational blog posts. No operational issues, as we'd expect/hope 1. Report from last week's CIUK - Matt, Daniel, Jens,... High density with 60 or 90 drives in 4U, which of course is why people buy these things, to get high density; however, as we have seen in the past you might have concerns about the bandwidth of getting data into and out of the thing; and for very large systems the time it takes to rebuild an array after failures (however, see also the Cunning Scheme(tm) below.) However, even if you sort this one out, you may still worry about the access patterns and how the bandwidth may limit access to your data. Brian suggests 50 gig/s needed for 1K job slots. And would it need loads of RAM? ZFS with JBOD - Marcus leads the way... 10 gig NICs don't bond well? Reported tricky to get good performance, albeit a while ago. QMUL - 60 drive box holding medical data backups; users also looking at backing with LTO4 tape. also tempted by Mellanox 100 gig switches. And such a box might saturate the site bandwidth - if CASTOR at RAL is doing, say, 10 gigabit per second (not infeasible), most of this traffic is to internal compute clusters. However, in the Glorious Future(tm), if a site is storage only (or near enough) then the cluster-to-storage traffic would necessarily go across the WAN, thus leading to contention etc. On a more "managerial" side, there is interest in GridPP's (and relations') data transfer capabilities, e.g. Globus and GridFTP - talk to Hartree. Also discussions with DiRAC on data (but also the Pathfinder project on authentication and accounting and stuff.) Question is should we be better represented at CIUK (nee MEW) next year? GridPP, like DiRAC, is STFC funded, so we could argue we should (GridPP itself didn't really present anything...) 2. What were your highlights for 2016 (storage and data management-wise, none of that other stuff) - let's accentuate the positive * "boring" storage, even during upgrades. Brian will seek out performance numbers from FTS * DiRAC (+ connecting other infrastructures...) * T2 evolution work * Good collaboration - GridPP as a data infrastructure is a success story (again) 2017 evolution * T2s * Accounting needs revisiting 3. Merry! Matt has a Cunning Plan(tm); Sam and Luke have thought or built something along the same lines, so there would be scope to collaborate. Daniel Peter Traynor: (21/12/2016 09:59:55) hi https://www.hgst.com/products/platforms/4U60-storage-enclosure Matt Doidge: (10:11 AM) 36-bay deja vu I wondered where Dell where. *were Lukasz Kreczko: (10:35 AM) sorry, I had someone in my office, are you talkingbit/s on one machine? we use IEEE 802.1Q, so requires switch support my bad, I mean mode=4 (802.3ad)