Mark Norman Robert Frank John Hill Alessandra Forti Jeremy Coles Sam Skipsey Gareth Roy Stephen Jones Matt Doidge Govind Songara Ewan Mac Mahon Brian Davies Elena Korolkova Jens Jensen Chris Walker 1. Mark's problem is not the same as the problem John had a while ago; there is one easily solved and another more mysterious which is being investigated at the moment. You should be able to have a separate MySQL instance... but it is also shared with the old DPM instance. 1 November for having sites on EMI - same as everything else. It would be better if sites didn't have the threat of being decertified. People should upgrade their head nodes if they can't upgrade anything else - at least it would appear as an EMI service. The services are released by dpm-listspaces (via GLUE). Slightly broken as name of RPM version becomes unset. 2. Roundup of GridPP last week Do we have consensus on not supporting GPFS? Would it lock you in to a certain vendor... Where are we in 2015 if we need to spend money supporting proprietary "solutions"? xroot may be part of this but is also HEP-specific. The group should be recommending which POSIX interface filesystem we should recommend. E.g. Lustre. Federation gives you resilience. But we still need something like FTS. The T1 "EOS evaluation" is also relevant, cf Shaun's presentation. Brunel "volunteered" to move away from DPM. Raul is meant to do it. Sam has talked to dCache from DPM? If we understand the metadata, perhaps this is sufficient. dCache can consume DPM data - potentially - but maybe it would be possible. Strategies * Delete files and reimport from original sources * Set up two endpoints - old and new - and FTS files over. Would need sharing disk servers carefully. And having to maintain two systems at the same time. * Move data to T1 and move it back once the new endpoint is set up. * Move in-place, migrating the metadata to the new system. On VO support - "Small" VOs are not necessarily small any more - at least wrt data volume; maybe wrt number of people. - CMS support? They can attend this meeting more often, of course :-) The impression that we are mainly DPM is not entirely right, although we spend more time on DPM because we have more DPMs (and maybe the dCaches are just ticking along nicely without any problems.) Of course we are happy to help anyone in GridPP. Stuart is leaving and we don't know who will replace him. Cloud storage question? What does "cloud" mean in this context. Probably the same as it means for the rest of the world, for once. 3. This month's pre-GDB on storage (although we have potential apols from Wahid for tomorrow..) http://indico.cern.ch/conferenceDisplay.py?confId=208241 Is there any DPM input? 4. It is the end of the quarter - the usual reporting stuff Publications, interactions with HEP and WLCG and others. Conferences, meetings. 5. AOB There will also be a DPM forum later this year, in Paris. Date tbd but expecting 3-4 Dec. Also planning for the first one next year in the UK - so if anyone is interested in hosting it... Could also be hosted with the next GridPP meeting or the next EGI TF. DIRAC - moving data around. UK3A has gone in to BIS. Interoperation would be good for both data and job submission - and either would be useful. There is another proposal going in for assessment. It is also worth emphasising business impact, but HPC and HTC are lumped together. It may be worth looking for more business cases. E.g. NGS (Next Generation Sequencing.) [10:00:36] Jeremy Coles joined [10:00:40] Sam Skipsey joined [10:01:35] Gareth Roy joined [10:01:50] Stephen Jones joined [10:02:20] Matt Doidge joined [10:03:39] Govind Songara joined [10:04:02] Ewan Mac Mahon joined [10:04:09] Brian Davies joined [10:05:25] Matt Doidge At Lancaster we hope to have it done this month [10:05:31] Elena Korolkova joined [10:05:36] Matt Doidge To avoid any rush [10:06:11] Ewan Mac Mahon You sort-of can see pool nodes remotely too, if anyone bothers to try. [10:06:31] Govind Songara Is any one having SL6 headnode in prodcution ? [10:06:33] Ewan Mac Mahon I'm not sure I'd want to rely on CSIRT not looking behind the curtain. [10:08:33] Ewan Mac Mahon Pfft. Still not going to happen. [10:08:49] Ewan Mac Mahon (and doesn't need to anyway) [10:09:22] Christopher Walker joined [10:09:33] Christopher Walker left [10:09:38] Christopher Walker joined [10:10:16] Christopher Walker left [10:10:39] Christopher Walker joined [10:11:09] Christopher Walker left [10:11:47] Jeremy Coles Ewan it may not be EGI CSIRT but our security team (as a responsible NGI). So now you know where to look! [10:12:04] Christopher Walker joined [10:14:21] Ewan Mac Mahon Well, from having seen the way things have gone even at the EGI level with the recent WMS problems, I just wouldn't want anyone to get the impression that CSIRT feel in any way restricted to looking at the information system - they don't. [10:14:44] Ewan Mac Mahon People are quite happy to open gridftp connections to nodes and poke around their filesystems directly. [10:16:30] Stephen Jones Chris, yousound like a dalek. [10:16:43] Jens Jensen It's the same consideration for the T1 diskonly eval. [10:16:53] Jens Jensen Prefer things run by other sites. [10:17:10] Jens Jensen ie outside hep too [10:19:22] Alessandra Forti indeed [10:19:57] Christopher Walker Apologies Stephen, I'll try and do something about my microphone [10:20:24] Ewan Mac Mahon I think this is something of an 'I wouldn't start from here' problem - I'm happy with the moving away from SRM idea, but I think we have a legitimate story as to how we got ourselves here in the first place. [10:20:44] Alessandra Forti Chris: I did tell you few days ago [10:21:17] Jens Jensen Or maybe you have a bad cold, Chris. [10:22:05] Ewan Mac Mahon Plus, of course, when we were initially running DPM to aggregate disk servers, Lustre &c. weren't really a thing. [10:23:30] Ewan Mac Mahon So, if a funder asks why we did this, I think we just tell them. If they ask why we want to insist on carrying on doing this, I think we tell them that we don't. [10:25:11] Ewan Mac Mahon Anyway; so we're OK with a Storage Group consensus position of 'Hell no' on GPFS then? [10:25:18] Sam Skipsey Yes. [10:25:56] Ewan Mac Mahon There's a blog post in that - title 'GPFS', body, 'Hell no'. Ticks us a metric box too [10:27:21] Christopher Walker Ewan, my personal position on GPFS is that it's a POSIX solution that does much the same as Lustre in much the same way. It locks you into one vendor - which is clearly a concern, but the cost is AFAICT, similar to proper Lustre support. The great advantage of Lustre is that you can get away without paying for it. [10:28:08] Alessandra Forti indeed if we have alternatives where we don't pay is silly to pay isn't it? [10:28:27] Ewan Mac Mahon It does have technical downsides too (like the kernel updates problems), but I think it's quite fair to discount it on the grounds that it's equivlent to Lustre, but with added vendor lock-in. [10:28:31] Christopher Walker Depends what you think you are paying for. [10:28:40] Ewan Mac Mahon That's not a bad ground to make a judgement on. [10:28:51] Ewan Mac Mahon Er - on which to make a judgement. [10:30:16] Christopher Walker 1Gbit=10TB/day, 10Gbit=100TB/day = 15 days for QMUL [10:34:43] Ewan Mac Mahon Sam is (as per usual) correct. [10:35:37] Elena Korolkova apparently more than 50 5 cms sites in UK are dCache [10:35:46] Elena Korolkova sites [10:36:40] Christopher Walker Ah yes, clouds are the other interesting thing [10:38:45] Ewan Mac Mahon That would sort-of be the Tuesday ops meeting. [10:38:52] Ewan Mac Mahon (ish) [10:39:52] Christopher Walker s/welcome to/CMS are encouraged to attend/ [10:41:37] Jens Jensen http://indico.cern.ch/conferenceDisplay.py?confId=208241 [10:45:03] Alessandra Forti can you remind when they are? [10:45:22] Alessandra Forti the xrootd federated things? [10:46:19] Sam Skipsey Every other monday, via Vidyo. I'll forward you the email, Alessandra. [10:47:33] Alessandra Forti no agenda? [10:47:40] Alessandra Forti thanks btw [10:47:50] Sam Skipsey there's usually not a detailed agenda [10:50:30] Ewan Mac Mahon Well, we should clarify whether there's an application in for storage stuff specifically, or just for the UK3A thing. [10:50:58] Ewan Mac Mahon If there's a storage specific proposal, it would be nice to know what we were being proposed to do. [10:51:20] Ewan Mac Mahon Seeing as, you know, we're going to be the people doing it. [10:56:15] Ewan Mac Mahon And the wonderful thing with the Welcome Trust folks is that they think like academics but have that whole commercial overlap thing. [10:57:28] Ewan Mac Mahon That sounds monumentally dodgy Jens. [10:57:46] Jens Jensen [10:57:49] Ewan Mac Mahon 'contacts in the pharmacutical industry' could cover a multitude of things [10:57:54] Jens Jensen Yep [10:58:34] Gareth Roy left [10:58:35] Mark Norman left [10:58:35] Brian Davies left [10:58:35] Elena Korolkova left [10:58:35] Robert Frank left [10:58:35] John Hill left [10:58:37] Sam Skipsey left [10:58:38] Govind Songara left [10:58:40] Jeremy Coles left [10:58:40] Ewan Mac Mahon left