IPv6 Working group 10-11 April 2014 ----------------------------------- Day 2 Notes by Chris Walker 11 April session. Present: Jerome Bernier; Alastair Dewhurst; Marek Elias; Dave Kelsey (Chair); Fernando Lopez; Edoardo Martelli; Kars Ohrenberg; Francesco Prelz; Duncan Rand; Michail Salichos; Andrea Sciaba; Ulf Tigerstedt; Christopher Walker (minutes); Tony Wildish (remote), Raja (remote) Introduction ============= David welcomed Michail Salichos to the meeting. Michail is one of the FTS3 developers, and is based in CERN. Revisit testing plans ===================== SRM --- Testing of SRM via Phedex (ACTION:Duncan/Tony). dCache ------ KIT, PIC, (Imperial waiting for something ready to be tested) Fernando: as noted yesterday, dCache srmcp doesn't work. ARC lcg-cp doesn't work either. ACTION: Francesco and Ulf and Karl will contact dCache team about gridftp redirect not being supported in IPv6. DPM --- Glasgow: Mark Mitchell (previous IPv6 rep at Glasgow) has moved elsewhere in Glasgow. ACTION: Duncan to contact Sam Skipsey at Glasgow to encourage him to continue the work. Issues: Mysql doesn't work over IPv6. At pre GDB, the following to talk about/give a slide on appropriate storage systems. StoRM: Chris Walker - StoRM has been deployed dual stack with all components on a single machine, but needs more detailed testing. As MySQL is used, if different components are on different systems it is likely to run into the problems that DPM does - though it should be possible to get the MySQL parts to talk IPv4, and the rest to talk IPv6. For the pre GDB on IPv6, the following were volunteered to give a slide or two on their storage system. Storm: Chris Walker dCache: Ulf DPM: Duncan to ask Glasgow (Sam Skipsey). Pre-GDB: Tony Wildish to talk about his data transfer tests. Some standard tests would be useful - both automated tests, and simple tests that a sysadmin can run interactively. Action: Andrea Sciaba - Some standard commands to run against an SE to be documented on the wiki. Action: Duncan to run nagios SAM tests on storage elements using the ipv6.hepix.org FTS3 ==== Michail: ActiveMQ is known to work (but needs to be tested). There is a piece of the FTS3 client that is known not to work. Once it does work, FTS3 should be fully dual stack. There is a testbed infrastructure so he can, and will, test this. Action: Michael: fix FTS3 client, then test it. Michael: ActiveMQ is required for the WLCG transfers dashboard. Atlas Rucio framework will also need information from FTS3 which it will consume via ActiveMQ. This needs to be tested too. On the pilot instance (used for atlas functional tests), 6 months ago, IPv6 support was enabled as an experiment, efficiency dropped from 98% to 70%. Imperial has tested with a dual stack FTS3 and not seen the same problems - it may have tested at a different time (or different sites). Action: Michael/Duncan/Alastair follow up on this. WORKLOAD MANGEMENT ================== Pre GDB: Francesco to report on workload management. - Cream CE works - ARC CE Works - OSG CE - we may have results by this time - perhaps Brian Bockleman can present - ideally in person, but perhaps remotely. Batch systems: Condor works. Slurm - ULF - known to be IPv4 only (slurm) should work over IPv4 only on a dual stack system, ULF to test this. GridEngine - Imperial running dual stack. LSF - used by CERN - no information Torque/Maui - rumour is that it doesn't work - Could we contact Nikhef for Torque/Maui - though reports are that it doesn't work. David: Note even if there are IPv6 worker nodes, they can have a private IPv4 address - so dual stack for internal use (such as the batch system). It was suggested that for WLCG, if the CE can run on IPv6, then the internals of a site (dual stack or not) are a site concern. Experiments testing =================== CMS (Andrea Sciaba). Could try to setup GlideinWMS - but this has not been tested yet. The people involved are very busy and understaffed. Duncan: Very dependent on Condor - either IPv6, or IPv4, but no dual stack. ATLAS: APF development machine dual stack at CERN. Peter Love is making progress. LHCb: Dave Kelsey to talk to Raja. Alice: no rep present, Dave to contact Costin to give an update. Experiments should present the use cases they want to test in the context of IPv6 only worker nodes (with IPv4 routable across CERN). Alastair (atlas): Mentioning requirement for IPv6 to people has resulted in many the initiative and fixing the things they are responsible for. IPv6 Site Readyness survey ========================== A revised version of the site readyness survey was prepared. Action: Andrea to look into turning that into an IPv6 status table on the Twiki. Plans for dual stack on the WLCG ================================ In particular, what can be shown to be dual stack by the pre GDB in June. Imperial and Umeo have production dual stack services. Nebraska believed to be dual stack. Could (some of) lxplus be made dual stack? Alastair: How about presenting timescales promised by the Tier-1s put it in the slides of the pre-gdb - then you can hold them to it. Next Meeting dates: =================== Phone conference on Thursday 8th May 16:00 CST (CERN time). Phone conference on Tuesday 27th May 16:00 CST (CERN time). Face to face 10th June - pre GDB. WLCG workshop Barcelona 7-9 July will have something on IPv6, but difficult to schedule an add on IPv6 meeting, so have a phone meeting the next day. Thursday July 10th - Vidyo meeting 15:00 CST (CERN time) - two hour meeting.