HEPiX IPv6 Working Group - F2F meeting #27 (Vidyo) 23 May 2013 Present: Thomas Finnern, Costin Grigoras, Bruno Hoeft, Dave Kelsey (Chair), Tomas Kouba, Fernando Lopez Munoz, Edoardo Martelli, Mark Mitchell, Andreas Pfeifer, Francesco Prelz, Duncan Rand, Ramiro Voicu, Chris Walker, Tony Wildish. Apologies: Kars Ohrenberg 1. Dave welcomed all to the meeting. The agenda was agreed. Edoardo requested that we add a discussion on the HEPiX wiki. 2. Roundtable updates. Mark reported that Glasgow is now connected to the testbed but there are still some ongoing problems with DNS. Kars noted in his email giving his apologies that there is nothing to report on dCache and IPv6. CERN. Edoardo reported good progress rolling out IPv6 across the CERN site, but that they had hit a problem with large OSPF packets. Their HP routers and Brocade routers work on their own well in this respect but exchange of routing info between HP and Brocade has a problem. It is felt that the RFC is unclear and that different manufacturers are interpreting the OSPF fragmentation differently. This is being chased with both manufacturers. ACTION 27.1 Edoardo to add details to our wiki/site config/knowledgebase. It is already on the CERN IPv6 knowledge base. See: http://ipv6.web.cern.ch/content/ospfv3-neighbor-stuck-loading Edoardo reported that DHCPv6 is now available at CERN to users on request. They are seeing problems with Linux (SLC6, 64bit) clients where they do not get the correct IPv6 address. The client was sending a DHCPv6 request with a corrupted value in the Client Identifier field. The DUID type was "link-layer address plus time", but the link-layer address was a four bytes string with no clear meaning. See: http://ipv6.web.cern.ch/content/dhcpv6-client-doesnt-get-ipv6-address Francesco suggests trying the extension to authorise by MAC address (a work around not described elsewhere). Edoardo says that Windows clients work well. Francesco warns that this might change when the privacy extension is turned on. Bruno reports that work is ongoing at KIT on their mini-cluster. Tomas. Nothing much new at FZU. Mark reports that the reverse DNS at Glasgow is partially implemented and asks Tony to re-instate the data transfer tests. Francesco says he cannot see the reverse DNS yet. This seems to be a Janet problem. Mark also reports problems with IPv6-only DPM where there are problems with MySQL. On the other-hand dual-stack DPM seems to work fine. Francesco as to when he can do a CREAM-CE test with Glasgow. Mark says he will get back to him. ACTION 27.2 Mark to contact Francesco re CREAM-CE tests. QMUL. Chris reports that he has been testing Jumbo frames at QMUL. Tests of PerfSonar between QMUL and FZU with Jumbo frames worked but he did note routing asymmetries. The return route came back via Tiscali. This is now fixed. Francesco comments that he would have to move peering to move to jumbo frames. He notes that the Tier1/CNAF and Milan already use jumbo frames on IPv4. WLCG today uses various MTU sizes on IPv4. Chris comments that it is the lack of Path MTU Discovery that causes problems. He also comments that QMUL sees more packet loss on IPv6 (during the first couple of hops in Janet) compared with IPv4. The topic of MTU size and IPv6 would be a good topic for a future meeting (July F2F?). 3. Testbed and Data Transfer tests Tony reports that his tests are still running. Imperial is now back in. Glasgow still have DNS problems. Duncan reports that Oxford have expressed an interest in joining the testbed but are concerned by the traffic rates. Tony says there are several ways that the data flows can be throttled back so that is not a problem. Tony notes that to date we have transferred about 800 TB of data on the testbed. For future testing over DPM - Tony confirms that he needs 2 DPM endpoints accepting IPv6 connections (FZU and Glasgow are the 2 foreseen). He would then need SSH access into them to install PhEDEx client software. ACTION 27.3 Mark, TomasK and Tony to take CMS data transfers over DPM testing forward. Dave reminds all that the next point in the testing plan was for Glasgow and CERN to test ATLAS production monte-carlo work (on IPv6). Before the July F2F meeting would be VERY good. This would need a job submission mechanism at CERN and would use the CREAM CE at Glasgow. 4. Draft web site for IPv6 Compliance Testing. Edoardo gives a demo of the new CERN IPv6 drupal web/database that he has set up. http://cern.ch/hepix-ipv6 When you login to this, you get to see more details and can edit the data. The revision history is also available. Edoardo asks if there are any fields missing or badly named? No great response but we suggest that "Software component" is a good name to describe the package, middleware, tools etc. Edoardo will edit. DaveK asks if there is an easy way to achieve our desire for colour-coding? Green means IPv6 works without config work, Amber means that some configuration work, recompiling, relinking is needed, red means it does not work, grey means unknown. Edoardo thinks this may be possible and will investigate. ACTION 27.4 Edoardo to investigate colour-coding on the IPv6 compliance testing web. 5. AOB Edoardo comments that there are problems with the HEPiX IPv6 wiki hosted by Caspur in Rome, particularly as the host certificate is not issued by a trusted CA - this stops Google indexing. He asks if we should just move all the material to the new CERN web. The conclusion of the discussion is that we could start by implementing our own knowledge base on the CERN web site and defer consideration of any full move until later. Next meetings: 13 June 2013 16:00 CEST (Vidyo). 4-5 July 2103 next f2f meeting at CERN. Notes by Dave Kelsey. 13 June 2013