Attending: Janusz (special guest star), Sam, Jens (C+M), Jeremy, Steve, Robert, John H, Matt D, Duncan, David, Ewan, Tom, Raul. Apologies: Wahid (at ATLAS Jamboree) 0. Blog posts - Sam would do one on EC work (hint) We could also do something on the popularity of datasets, or maybe components of the infrastructure (like we would for the big data thing on Monday), saying "look you are talking about clouds and stuff but we have been doing it for years" kind of thing. 1. DIRAC discussion with Janusz There is a server on a VM at IC, and some support for small experiments. Should be sufficient resources at the current level of use. Simon and Daniela also tracking the developments of DIRAC, but as usual the service needs a tradeoff between stability and new features. Is this a *production* instance? It is sort of a "prototype production" instance, but stable enough to do real stuff: CERN@School is the only VO currently to have done much but more will come (eg MICE, NA62). As WMS will not be available, may be a good option for the "smaller" VOs. Currently, has run >6K jobs since 1 April 2013, some of which being Andrew McNab's test jobs. See demonstration/discussion at hepsysman at RAL. There is the equivalent of lcg-cr for moving data, including the catalogue stuff - will be getting support for async transfers. People who want to try to use it should join the GridPP VO, if they aren't already in it. However, Janusz still needs to add the DN to the service. LHCb has a trick up its sleeve to obviate this step. Is there something like a visible workspace? When we get more VOs to try it out (NA62, MICE), maybe some proteomics folks who expressed an interest (Tom), or COMET (Janusz), we could provide "everything a VO needs". Consequently, perhaps a real decision is needed regarding the future of DIRAC (and DIRAC support) in GridPP. In particular, whether this group should be able to support DIRAC data tools (answer: yes, stay tuned.) Sam has a student who'd done work on the client - slightly cranky (the client, not the student) - how is the response time from the developers? - slightly varying reports. 2. CEPH developments Giant should provide a striper for rados (that sentence probably doesn't make sense unless you know something about CEPH). Similar to Brian B: GridFTP for HDFS; Shaun Alistair Dewhurst, Sam: working with the APIs to get GridFTP support for CEPH. Good APIs, but how is the object referenced in an object store? Samuel Cadellin Skipsey: (03/12/2014 10:04:25) Sorry about the, Vidyo is very crashy at the moment. Jeremy Coles: (10:16 AM) What can the storage group do to help get this DIRAC instance to "production" status? Tom Whyntie: (10:17 AM) Hello - sorry, just joined Jeremy Coles: (10:17 AM) The assumption being that we are intending that DIRAC becomes a production service ... which I thought was agreed. Duncan Rand: (10:18 AM) Yes. Tom Whyntie: (10:20 AM) My HEP SYSMAN slides here: https://indico.cern.ch/event/350917/ Full chain example code here: https://github.com/gridpp/dirac-getting-started (Using CERN@school data and examples.) Steve Jones: (10:21 AM) Yes. No timeline. What events need to take palce to bring this into production. I.e. in what way is it not ready for prime time and what activity will occur to address those? Put it in a plan. Ewan Mac Mahon: (10:22 AM) In particular though, as Jens says, what needs to be/can be done by (e.g.) the storage group or anyone else outside the core Imperial team. I think people are happy to help, but need a todo list. Steve Jones: (10:23 AM) Yes. What is it about the current system thast needs work? Samuel Cadellin Skipsey: (10:23 AM) And to set up DIRAC Clients, at least (which I have one half-done here) Steve Jones: (10:23 AM) What doesn't it do that it needs to do? And what does it do that it should not do? Samuel Cadellin Skipsey: (10:24 AM) Also, if we're talking about Dev work, probably looking at the Imperial Dirac fork would help... [I do have some ec stuff that might be committable at some point] [But that's not really "necessary"] Steve Jones: (10:27 AM) A gap analysis between current workflows and new prospective Dirac workflows should show where work is needed. That needs to be turned into a TODO list, and a timeline (.e. Gantt) needs to be made from the TODO list. It.s standard project management planning. Ewan Mac Mahon: (10:28 AM) I think there's an extent to which we're only going to be able to answer those questions by actually pushing small VOs towards it. I think we need to strongly encourage real VOs to actually use the Imperial Dirac, albeit on the understanding that they're using a prototype service with the objective of finding out how t use it, and how well it meets their requirements. But until they try to do (small scale) real workflows through it we're just not going to find the interesting bits. Steve Jones: (10:31 AM) Yes. A working prototype is useful for experimenting asnd finding requirements. That's the main thing they are for, BTW. Let's get a small VO using it and "observe" them using Dirac and interviewing them in a usability study. Write up the results of the study and you have your TDO list. Ewan Mac Mahon: (10:36 AM) Where did we get to with Pheno? Did Mark go and talk to them about Ganga? Ta. Jeremy Coles: (10:39 AM) I have not heard from Mark since he said he was going to chat with them. Duncan Rand: (10:40 AM) Got to go. Steve Jones: (10:41 AM) Sam, you are the one who brought up the KEF/SEF thing in the first place.... I'm happy with SEF, like in Cenotaph (a word with similar provenance). On the other hand, C is always hard in Welsh.... Jens Jensen: (10:45 AM) It's from the greek where it's a K Steve Jones: (10:46 AM) So is Cenotaph, Jens .... there is no answer. Jens Jensen: (10:46 AM) ... Tom Whyntie: (10:46 AM) Ha! Ewan Mac Mahon: (10:48 AM) This is clearly beyond our sen.