Agenda: https://indico.cern.ch/event/615970/
Attending:
• Alastair Dewhurst
• Andrea Sciaba
• Bruno Hoeft
• Costin Grigoras
• David Kelsey
• Duncan Rand
• Edoardo Martelli
• Francesco Prelz
• Kars Ohrenberg
• Ulf Bobson Severin Tigerstedt
a) The main CHEP2016 paper needs to be revised and resubmitted. Alastair will propose the changes.
b) DPM issues. See GGUS ticket. (GGUS: 127285) Globus, FTS and dCache all break the RFC. Andrea to talk to Oliver Keeble. Ulf and Raul(?) report problem to IETF.
c) CMS testing of IPv6-only at Brunel - on hold. Any news?
d) Edoardo to add 2 columns to Tier 1 table (done)
e) Need to work on plan for training half-day in Manchester - June 2017.
f) All to help Andrea with documentation on deployment in the
Andrea Sciaba (CMS): Still testing glidein-wms job submission. Still have problems, updated to latest version of HTCondor. Server attempts to make a connection to the WN from the outside. Needs extra component. Only one of the xrootd redirectors in Europe is dual-stack and this failed causing problems - acts as a single point of failure.
Costin (Alice). Recently had a workshop with sites, IPv6 was asked about. Most sites aware but few have done anything concrete - details on agenda. Oxford will fully support IPv6 in the next 6 months. RAL T1 making good progress but awaiting xrootd-ceph plugin so no ALICE support as yet. Ceph can accept names without a ’slash’ in it but xrootd cannot. Nothing to do with IPv6. Unique to ALICE. Working with CMS at the moment, ALICE is next. (ALICE uses third-party copying to transfer data). US (ORNL, LBL) sites awaiting EOS to support IPv6 (should be soon). Romania - work in progress. IN2P3 - Tier-1 has dual-stack storage. INFN will act in an coordinated way, but no exact estimate. Kisti within a month. Japan - little progress. NGDF - IPv6 has been working for some time.
Francesco - At Milan he has not been given permission to dual-stack things yet. Reminded us that Terry had reported that Vidyo doesn’t work with NAT64. Edoardo reported that the Vidyo developers say it won’t work with IPv6 anyway.
Bruno: Server people are setting up new dCache file system with dual-stack.
CERN: storage team is ready with EOS4.1 (supports IPv6). Will start with LHCb instance this week. If OK will plan upgrades for the other experiments.
Francesco: What about the DHCPv6 RFC that has come out? Edoardo: we’ve been testing it locally and are moving forward. There is a new IETF working group on DHCPv6 version 2.
Duncan - v4 of perfSONAR was recently released. Most hosts will have updated the software automatically. Duncan has been helping UK sites to get (dual-stack) mesh working well after the new release. Traceroute results now appears in the mesh http://maddash.aglt2.org/maddash-webui/index.cgi?dashboard=UK%20Meshconfig. The issue of making all the PS meshes show IPv6 data has been recently discussed.
NDGF: T1 working fine. Will implement perfsonars at each of the sites (it is a distributed T1).
Alastair (ATLAS): In March Panda-Server front ends were made dual-stack. Run jobs on IPv6 only WNs. Upgrading Frontier to dual-stack (numerous operation problems (not IPv6 related) has slowed things up). Small but non-trivial number of sites accessing Panda servers via IPv6. BNL have recently (since Sept 2016) actively implemented IPv6. Made their FTS dual-stack but had to roll-back. BNL dCache now dual-stack. Note: BNL is 25% of ATLAS T1s. Reported at BNL HEPiX meeting: https://indico.cern.ch/event/595396/contributions/2544103/attachments/1447978/2231941/BNL_Site_Report_HEPiX_Spring_2017.pdf
Fernando: Not much news from PIC. Put HTCondor into production < 50 nodes. All HTCondor installation is dual-stack.
There were a few corrections which have been implemented.
Two new columns have been added to the table http://hepix-ipv6.web.cern.ch/sites-connectivity, with percentage dual-stack storage offered by the site by three dates. Several T1s have not filled this in yet and so Bruno created GGUS tickets (keyword WLCG-IPv6 Tier-1 readiness) enquiring as to progress. Bruno will summarise the status for Martin to present at the the WLCG meeting tomorrow.
Dave K enquired about Francesco’s web page (http://orsone.mi.infn.it/~prelz/ipv6_bdii/) showing the number of WLCG hosts that are dual-stack, there has been a recent increase - seemingly at SurfSara.
There was a discussion about traffic in and out of CERN: https://netstat.cern.ch/monitoring/network-statistics/ext/?q=IPv6&p=EXT&mn=Internet&t=Monthly and progress on getting monitoring of IPv6 traffic on the LHCOPN.
Andrea has some information which he has put on the web including perfSONAR, dCache etc which support IPv6 ‘out of the box’. DPM was slightly more complex with the redirect issue. http://hepix-ipv6.web.cern.ch/content/how-deploy-ipv6-wlcg-tier-2-site
It was suggested that the WLCG SAM tests could be used to check IPv6 status at sites. The main service to be tested would be the storage. A new profile for each LHC VO could be created for IPv6 and suitable tests created. The existing WLCG operations reporting could be used to summarise state across the whole WLCG. How will sites moving to IPv6 be supported. We could use lcg-rollout - this is a rollout after all. Perhaps we should start with a small set of sites so as to keep things manageable.
Oliver had previously asked a question about DPM redirection to another host and the RFC. The RFC had previously been ignored by others. Should we try to change the RFC which is old and they probably didn’t think about this use case when it was written. There is the possibility to submit an RFC errata report.
(minutes taken by Duncan Rand)
- Review of agenda. Shorten meeting to 11:00.
- Perfsonar?
- IPv6 tickets: a few sites responded to getting pushed.
- Training: 2 1½h sessions
- intro to ipv6 basics
- is my traffic on ipv6 or not?
- data transfer tests
- set up perfsonar
- slide-session on wednesday afternoon
- edoardo will talk about dhcpv6
- Cheat-cheat cards for subnetting
- Ulf will do transfers
- Should we explain why we are doing this?
- Session on "how to prepare a T2 in general"
- Hands on perfsonar
- we need more sysadmins to the session
- preliminary f2f in september 11-12.9
- 13.7 vidyo meeting
- DPM/globus problem. Packages in epel-testing about to be deployed.
(minutes taken by Ulf Tigerstedt)