Starting at 16:00 CET
(notes by Dave Kelsey)
Present: Martin Bly, Nick Buraglio, Tim Chown, Jiri Chudoba, Dimitrios Christidis, Catalin Condurache, Costin Grigoras, Bruno Hoeft, Hiro Ito, Dave Kelsey, Edoardo Martelli, Raja Nandakumar, Francesco Prelz, Duncan Rand, Andrea Sciaba, Tim Skirvin, Edward Simmonds
Apologies: none?
1. DaveK welcomed all. Agenda was agreed.
2. Roundtable Updates
FNAL (TimS). ~700 worker nodes are now dual stack. Currently testing containers running with IPv6.
CMS (AndreaS). No news.
EGI (CatalinC). No news.
ALICE (CostinG). Shows traffic plots. Email sent before the event said:
Here are some historical plots on the topic of IPv6 in ALICE:
https://docs.google.com/presentation/d/1yAXZ6HfuI861KAgfA0-jYcEKVGp8V-21oxrVvDvt5MQ/edit?usp=sharing
We can discuss them a bit tomorrow, in short:
- Slide 1 is for how many of the job slots connect on IPv6 (Java clients to Java services);
- Slide 2 is an aggregation of worker node tests, running `curl -6` and just `curl` to see whether the happy eyeballs is by default too quick in giving up on IPv6.
The numbers are very similar for all the above, about 55-56% of the nodes (or job slots - the distinction can be important in case of many core nodes) can use IPv6, and use it by default.
For each of the above there is a link to the history. I'll keep accumulating these statistics so they can be revisited at any later time.
JiriC asks if the plots can be split by site.
LHCb (RajaN). In Oct/Nov they were merging multiple root files. GGUS Ticket https://ggus.eu/index.php?mode=ticket_info&ticket_id=153653 relates to an IPv6 problem at Nikhef/SARA and dCache. They see a black-hole on some servers.
DUNE (RajaN). No news. Waiting for FNAL.
Prague (JiriC). They added 2 disk servers for ALICE - not IPv6 yet. Traffic off site (IN+OUT) is about 2/3rds IPv6.
ATLAS (Dimitrios). No news. DaveK asks if ATLAS can persuade those sites who have not yet deployed dual-stack. It seems that the US ATLAS status spreadsheet is not fully up to date. HiroI comments that BNL does provide IPv6 storage.
London, UK (DuncanR). No news.
Italy (FrancescoP). Shows camera to prove he is in a meeting room at CERN! Francesco has analysed the KIT network dump. (BrunoH) shows plot of KIT production switches and a list of remote domain names - many at CERN. Shows list of IPv4 top-talkers. These relate to running workload on FNAL HTCondor server (all CMS). Is CMS at FNAL using an IPv4-only HT-Condor network? It was noted that Shawn McK had also reported (at a US meeting) that the move from GridFTP (where IPv6 was used) to WEBDAV (IPv4 only). Java Storage defaults to IPv4 unless the config for Java is changed. TimS and EdwardS say they will help (FNAL).
HiroI (BNL) says they have fixed 3 dCache sites in USA. Also DPM/Java. He will send details by email.
Info from Shawn (sent 14 Feb). dCache Java problem - see https://docs.oracle.com/javase/7/docs/api/java/net/doc-files/net-properties.html java.net.preferIPv6Addresses is "false" by default.
MartinB (RAL). The RAL tape service is moving from CASTOR to Antares (based on CTA). This is IPv4-only to start with. Network changes are required to support IPv6 - these will take several months. There is a major reorg of the RAL network planned for June 2022.
3. CERN Technical Student
Edoardo presents the proposal for a student at CERN to work on IPv6 - draft text shown to us. There is strong support from this WG - the work will be very useful for tracking and undertanding cass where IPv4 is still the preferred protocol.
4. Date of next meeting.
The earlier dates now clash with UK Networkshop50 (8-10 June). We decide to move the two-day IPv6 meeting to Tue/Wed 28/29 June 2022. Perhaps it will be possible to meet in person at CERN?