=================================== GridPP Technical meeting 2013-12-06 =================================== Present: Andy McNab, Chris Walker, Lukasz Kreczko, Peter Love, Dan Traynor, Ewan MacMahon, Robert Frank, Alessandra Forti, David Colling, Daniela Bauer, Simon Fayer, Adam Huffman, Andrew Lahiff 1. CMS ------ - Andrew Lahiff: - Continued testing of ramp-up speeds on HLT cloud, checking configuration on glideinWMS side, to ensure things working as fast as possible - 28 minutes to startup 2600 jobs; should be faster on next attempt - Database cleaning made an improvement and David said that Tim/Jan from CERN-IT were interested in the results - Andy McNab: could there be a problem in providing enough jobs for this resource from CMS - Andrew Lahiff - Condor would be the limitation and we don't seem to be near that yet 2. ATLAS -------- - Peter Love: - No longer have access to HLT until 2014 - Other work running there - Problems with upgrade to Condor 8, so rolling back to 7 - General lack of jobs in ATLAS at the moment 3. 'Stealth cloud' - Daniela ---------------------------- - see slides - David: how does it handle credentials for data access? - Simon: job arrives with proxy, like any normal job - Chris: 1 VM per job? - Simon: 2 concurrent jobs per VM, will run next job in the queue once initial ones have finished - Each image has 2 vCPUs, glexec deals with changing users - Chris: overhead of starting a VM? - Only a couple of minutes - VMs hang around for 20 minutes - glexec used mainly to protect the pilot proxy used by the cloud - David: could this be used for commercial clouds? - Simon: in principle, yes - Access to storage might be a problem (currently relies on Imperial proxy servers) - Adam: Upgraded GridPP cloud to Havana - David: CERN will upgrade in March 4. LHCb - Andy McNab -------------------- - Vac now working with CVMFS-based contextualization, so ready for those sites that want to test - Ewan interested but this will be delayed by next week's DPM workshop 5. AOB ------ - Alessandra: - WLCG Multicore taskforce approved - She would like to have different sites running with different batch systems to contribute to testing