Present: Brian, Sam, Wahid, Jeremy, Ewan, Gareth, Stephen, Chris, Duncan, Alessandra, Jens, Elena 1. DPM collaboration discussion Tools and other non-code input would be reasonable contributions. We already provide ~1/2 FTE. Somebody needs to provide the security fixes etc... and the coordination on development. Suggestion that the effort from the UK should be in core development rather than on additional tools. Either we train people to do DPM development (which may be difficult given its history of rotating developers which wasn't ideal), or provide funding (but a CERN FTE is more expensive than a UK one?) Code refactoring may make it easier. Only two countries currently provide contributions, UK and France. Maybe we will need three parties for an MoU. Alternative would be dCache or StoRM/Lustre. StoRM would also need post-EMI support. CNAF need StoRM. Do we need SRM? E.g. GridFTP to Lustre (or similar distributed filesystem), or some sort of "cloud" storage. Ricardo may provide some effort. Do we need different scenarios for various levels of Ricardo-contributions? Experiments seem to "not care" what the storage is(?), but they would probably care if it stopped working, or if their storage becomes unsupported and we would need to go through large migrations. Will dmlite be an option? Plugins may be easier to develop. Stuart will send the instructions to the list on how to get the source code - or Sam. Old code was less clean, apparently. 2. RFIO - what next...? Do we need to look at xroot, too? Sam and Stuart looked at the DPM RFIO, Jens at CASTOR RFIO. For DPM, is it possible for a single disk server to overload if getting too many RFIO requests. Maybe RFIO can have a transfer cap. Stuart thinks the cap can be set low enough to complete within the TCP timeout. With Lustre, an admin can set the limit for a disk server. xroot is improving for DPM, and is "the way forward" in CASTOR land, so maybe worth documenting as well. Federation is less tested. Does xroot offer the load limiting? We'll need to check. RFIO to HTTP may be better in many ways... load caps are available for web servers. Can also do clever things with squids. How do squids work with certificates - for WAN where you need security it will be a problem but using HTTP internally could be OK. 3. "Federation" workshop at IN2P3 - who is going? Shaun is going. It is 13-14 Sept. 4. Targets for the next few months - quick discussion. 5. Technology review - suggestions? 6. AOB [10:00:43] Jeremy Coles joined [10:00:57] Stephen Jones joined [10:01:11] Gareth Roy joined [10:01:32] Ewan Mac Mahon joined [10:04:08] Wahid Bhimji joined [10:04:16] Wahid Bhimji sorry [10:04:50] Queen Mary, U London London, U.K. joined [10:06:00] Duncan Rand joined [10:07:10] Wahid Bhimji CERN people cost a lot ! [10:07:53] Ewan Mac Mahon [10:08:06] Ewan Mac Mahon OTOH, the current developers know their way around the code [10:08:53] Sam Skipsey In fact, the dmlite thing is happening partly because the current developers wanted code they knew their way around better. [10:10:10] Ewan Mac Mahon Turning this round, maybe we're better off bringing the development to someone UK based because they'll churn less than CERN people will. [10:10:31] Alessandra Forti joined [10:11:05] Duncan Rand maybe the french could pay us [10:11:19] Jeremy Coles In Euros! [10:12:00] Ewan Mac Mahon OK, so in principle we've already got 2FTE on storage that we can play with. [10:12:22] Ewan Mac Mahon Without needing new people. [10:12:34] Ewan Mac Mahon But yes DPM isn't the whole of our storage. [10:12:44] Jens Jensen Do we have the development skills? [10:13:03] Ewan Mac Mahon Wailing, gnashing of teeth, rending of garments, end-of-days. [10:13:07] Ewan Mac Mahon Not good. [10:14:06] Wahid Bhimji (PS we have 3 FTE as Brian is also storage [10:14:24] Alessandra Forti none of you is a developer though [10:15:00] Wahid Bhimji quite - a change of role (and actually the gridpp4 funding wasn't for a developer more for support and for use of storage [10:15:19] Alessandra Forti indeed [10:15:48] Ewan Mac Mahon In which case we might be more OK with the 'minimal maintanence' model. [10:15:52] Gareth Roy Sorry need to do more recovery work [10:15:56] Gareth Roy left [10:16:27] Duncan Rand if that the first option in the doc Ewan? [10:16:45] Ewan Mac Mahon Yes, pretty much/ [10:17:04] Ewan Mac Mahon Option 1 seems to be basically no-one does anything and we all try to get off to something else. [10:17:55] Ewan Mac Mahon If 'something else' were a simple lustre+gridftp 'classic SE' approach then that would be somewhat less painful than moving to some other things. [10:18:18] Ewan Mac Mahon The doc doesn't address the question of what the 'something else' might be at all though. [10:19:00] Elena Korolkova joined [10:25:01] Wahid Bhimji IThe contribution might end up argue the opposite (other countries are doing it so cern doesn't need to ) [10:25:57] Ewan Mac Mahon Indeed. I think the argument that Alessandra is making is that we explain to CERN how horrific this would be, and that that's why they need to continue to maintain their effort. [10:26:08] Sam Skipsey quite. [10:26:34] Ewan Mac Mahon If they just plain refuse to do it though, then there's still the question of what we do next and how. [10:28:09] Duncan Rand but dcache and storm are independently supported [10:28:21] Duncan Rand by US/DE and I [10:28:39] Sam Skipsey I think we generally agree, Chris [10:29:43] Elena Korolkova When we start to talk to people we should find a write person. [10:30:07] Elena Korolkova I saw some people are sceptical about DPM [10:30:24] Wahid Bhimji Difficult to answer (what I want personally) - but I think we need to make the decision what we want first - then if we can do it. (I personally like DPM at least) [10:31:33] Duncan Rand how much are CERN willing to contribute? [10:32:07] Jeremy Coles They will not commit at the moment. Perhaps an FTE. [10:32:12] Queen Mary, U London London, U.K. Don't forget the LFC by the way. [10:32:16] Wahid Bhimji (They are being wooly - more than 0.5 - they balked a bit at 2) [10:32:21] Queen Mary, U London London, U.K. AIUI, that's part of DPM [10:33:06] Duncan Rand so we need to find 1.5 FTE? [10:33:11] Alessandra Forti what's the meeting tomorrow at 2:30 pm? [10:33:38] Jeremy Coles To come to some conclusions that I can present to the PMB next Monday. [10:33:48] Ewan Mac Mahon It sounds to me as though we want to start off trying to make a case for why CERN should keep their level of support up, failing that we want to contribute development effort ourselves, and failing that we want to work on the transition plan. [10:34:16] Duncan Rand if nobody helps will CERN let it die [10:34:27] Jeremy Coles Bug fixes perhaps. [10:34:30] Alessandra Forti Jeremy: is it on EVO? [10:34:35] Ewan Mac Mahon But we need to be clear that the third of those options is the last ditch avoid total disaster option, not simply the least preferable of a range of OK things. [10:34:37] Jeremy Coles Yes on EVO [10:34:42] Jeremy Coles Pwd: dteam [10:34:44] Alessandra Forti ok [10:38:39] Wahid Bhimji or good [10:39:30] Wahid Bhimji its going quite well (there are manual tweaks required) - and you need all disk servers on 1.8.3 [10:42:17] Wahid Bhimji I can't go [10:42:24] Duncan Rand when is it? [10:43:32] Jeremy Coles GridPP29 is 26th/27th September in Oxford. [10:43:54] Duncan Rand URL? [10:43:56] Wahid Bhimji GDB in on the wed then 13-14 is this lyon meeting... I can't find the link [10:44:03] Duncan Rand very nice lunches [10:44:13] Duncan Rand 3 courses + wine [10:44:33] Wahid Bhimji did you go last time duncan (where are they advertised I never see them [10:44:37] Elena Korolkova left [10:44:39] Alessandra Forti left