Choose timezone
Your profile timezone:
HEPiX IPv6 working group F2F meeting
→
Europe/Zurich
31/3-004 - IT Amphitheatre (CERN)
Description
WIFI access: pre-register your MAC address here (contact person: David Kelsey)
DRAFT agenda. Topics may change. Timings are approximate.
The PIN for Vidyo has been distributed by email.
Registration
Participants
Hepix IPv6 WG F2F meeting minutes (CERN)
Day 1 - 2016-05-18
Present: Edoardo, Duncan, Andrea, Ulf, Marian, Kashif, DaveK, Bruno, Alastair, Francesco
Remote participants: Tiju, Fernando, Raja.
Agenda was agreed
http://indico.cern.ch/event/501358/
Review of Actions
Alastair: done
Edoardo: done. Link is in CERN IPv6 knowledge base.
Round table updates.
CERN (Edoardo): no news apart from they are recovering IPv4 addresses though optmisation of the way in which addresses are used. Will deploy a new Wifi network across campus. Currently using 8000 but will need 16,000 IP addresses in the future. Plan to deploy less fixed sockets. DK: Network monitoring, does CERN keep historical data? Yes: to some extent, see CERN IT network monitoring pages. Edoardo will take a snapshot each F2F meeting.
Imperial (Duncan): new monitoring enables display of Ipv4 and IPv6 traffic. Saw CMS Phedex production transfer from PIC recently that went over IPv6. Dave K queried timing of enabling dual-stack on routes taken by Ipv4 and IPv6 e.g. over OPN etc. Alastair comments this is what perfSonar is meant to be spotting.
FTS monitoring: DK: is it possible to get a high level view of the proportion of traffic is going over IPv6. Not at the moment, bit there is an FTS monitoring plugin that can be enabled. Perhaps something could be done with this.
RAL (Alastair): No news.
Ulf: not much (IPv6) news. CA monitoring continues. The importance of making the CRLs available over IPv6 was reiterated. At the moment these are obtained from each CA. We think Nikef runs IPv6 mirror. Generally site is now dual-stack. Finally got logstash feed of dcache logs working again so can start analysing again.
Desy: no significant news, some more non-WLCG services have been made dual-stack. (LRZ site dcache was made dual stack on 9th March).
Marian: Firstly ETF service is in production for all LHC experiments but is IPv4 only. Is there something we would like to do to add in some sort of IPv6 testing? Could start with pre-production instance. Good to know if ETF is having problems interacting with dual-stack site nodes. WLCG IPv6 task force acts as interface between HEPix WG and WLCG Ops meetings. Andrea might be able to monitor the pre-prod instance if IPv6 is enabled. Can of course revert back if there are problems. Are IPv6 only machines supported at CERN? Depends on the services required. Might be possible. Perfsonar meshes will be rearranged. Perfsonar 3.6 will come out at the end of the year and the upgrade would be an opportunity to make other changes if required.
Action: Edoardo to think whether it is possible to run IPv6 only or to appear to be IPv6 only from outside. Could block IPv4 traffic from outside.
Action: Marian, Andrea and Edoardo to look into viability of IPv6 only ETF.
Alastair: the document is suggesting sites can be all IPv6 rather than just the WN.
Francesco: noticed that the IPv4 and Ipv6 maddash mesh plots show quite a lot of red and yellow rather than green. It was pointed out that >0.9 Gbps is required to be green. Asked whether we should start considering NAT64, the protocol translator. There is a standalone user space implementation for linux, Tayga. Are other sites able to test it? On perfsonar, Bruno at an LHCONE meeting tried to improve the greenness of the LHCOPN perfsonar plots. Would like to understand the Pic xrootd problem discussed on the list.
RAL (Kashif). GOCdb developers have been given a dual-stack machine it to work with. UK light router (connection to OPN) being worked on (removed) which should open the way to adding IPv6 capability. Adding more (~10) hosts to the IPv6 testbed. Discussion as to why RAL's perfsonar host is not in the dual-stack mesh - likely because not in the GOCdb.
Bruno (KIT): no plans to increase number of dual-stack services at the moment.
Fernando: T1 LHCb dcache pools and IFA T2 dcache IPv6 were enabled end of April. Received tape system (Enstore) that is IPv6 capable. Enabled dual-stack on one of two CMS VO boxes. Will also add it to CMS xrootd redirector. Dual-stack LHcb jobs running OK. CMS OK apart from xrootd-access. No feed back from ATLAS. Seeing IPv6 traffic with Imperial and ARNES (Slovenia).
Tiju: SSM secure stomp messager?
Experiment updates and WLCG IPv6 proposal
Raja (see slides): Dirac status. Networking bug fixed.
LHCb status: dual-stack services in production now. Running limited number of services. Waiting to check all is OK before adding more services. LHCb has new servers which are part of re-organisation and are planned to be all dual-stack. So started on way to make IPv6 WN useable by LHCb. Alastair requested a note of this good progress be added into the document.
Alastair/Atlas (slides) : CPU only resources. WN will need to talk to panda, rucio services and pilot factories. All use http(s). They're all based at CERN. APF dual-stack production. Testing of other services being done before production instances made dual-stack.
IPv6 only document: Available online in github. Trying to encourage experiments to enable IPv6 on central services. Some Canadian sites are interested in running IPv6 only resources, could run a few dual-stack services. Don't just want a CVMfs stratum one at CERN, would need at least four. Could have a totally new site wanting to be IPv6 only or an existing site that might want to bring on some IPv6-only WNs.
CMS would not like the removal of old IPv4 addresses on existing hardware before the end of Run 2 (Nov 2018)- this is in order to minimise disruption.
Andrea/CMS (sides): Preliminary at the moment, should be finalised by the June GDB. Interested in IPv6-only opportunistic resources. Today: ~11 sites with dual-stack - used in production. Some xrootd redirectors are dual-stack, cmsweb should be soon, glideinWMS almost fully validated, pilot factories are dual-stack or in process. Frontier not yet fully compatible. Wherever possible sites should keep IPv4 connectivity until the end of Run2. Priority should be given to dual-stacking xrootd (storage). For services HTCondor is the most important to be made dual-stack. CMS is close to being IPv6 ready.
Day 1 minute taker - Duncan Rand (19 May 2016)
Modified by Dave Kelsey (1 Jun 2016)
Day 2 - 2016-05-19
Present: Edoardo, Kashif, Duncan, Ulf, Kars, Marian, Francesco, Bruno, DaveK, Andrea
Remote participants: Raja, Fernando
preGDB IPv6 Workshop (7 june 2016) schedule:
- We need Alistair to present, even if it's remote.
- xrootd storage demonstration from a dualstack machine?
- Offtopic: should we schedule downtimes to test dualstacking? So far it has not impacted production.
- The projects should make up their mind of what they want and the most important thing it to get their central services dualstacked.
- guide to set up perfsonar with IPv6
- should the tests only be about storage?
- "IPv6 sysadmining for dummies" manuscript should be found. Noone is really sure of what it should contain. Some firewalling guides at least.
- PIC will do a dcache + ipv6 talk
- NDGF-T1 will do another dcache + ipv6 talk (transfer statistics)
Dates of next meetings
Summary of Actions
Day 2 minute taker - Ulf Tigerstedt (19 May 2016)
Modified by Dave Kelsey (1 Jun 2016)
There are minutes attached to this event.
Show them.
