Hi Mark,

> Jeremy, should we have a section on the GridPP Wiki for this programme?

Yes, but as far as possible we should contribution to the HEPiX working group shared documents. As Ewan suggested during the discussion it would be useful to set a series of GridPP targets, beginning at a low level, such as can I ping a CE across IPv6. To begin with the wiki can record these enabling activities, those involved, progress and helpful UK links.

For those who missed the discussion, a big driver now is that CERN may run out of IPv4 addresses next year ("Current VMs adoption plan may cause IPv4 depletion during 2014). See Martelli's talk at http://indico.cern.ch/conferenceDisplay.py?confId=228172.

Jeremy




On 29 Jan 2013, at 16:09, Mark Mitchell wrote:

> Hiya,
> after discussion with Ewan and Chris after the Ops team meeting anyone who is interested in joining the HEPIX IPv6 working group can find instructions here on what is required for joining the test bed can be found here:
> 
> https://w3.hepix.org/ipv6-bis/doku.php?id=ipv6:testbed
> 
> There are several mechanisms for the delivery of IPv6 to a server or Cluster.
> 
> A basic overview can be found here:
> http://en.wikipedia.org/wiki/IPv6#Transition_mechanisms
> 
> This entry includes many of the relevant RFCs and delivery mechanisms for IPv6 into a cluster.
> The Glasgow implementation utilises SLAAC as the primary method for addressing the test cluster.
> http://tools.ietf.org/html/rfc4862
> 
> Also AAAA (Quad A) DNS records are required:
> More information can be found here:
> http://tools.ietf.org/html/rfc4472
> 
> 
> A basic recommendation is that the IPv6 network and associated servers are separate from the production cluster environment, to start with, at least.
> There are multiple security and operational loop holes currently with IPv6 and these have not yet been fully explored by the HEPIX working group in relation to Grid Middleware. 
> To limit a potential number of problems and issues further testing is required prior to testing this on full production environments. 
> Anyone running an ARC production instance can run IPv6 and should be completely ok as this has been tested in NorduGrid. Cream CEs still require testing.
> 
> The plan at Glasgow is to re-image the test cluster shortly to allow for the testing of DPM. Glasgow has installed a limited number of nodes in a dual stack configuration, running both IPv6 and IPv4 as part of the distributed test but we wish to further test Puppet, DPM and CVMFS hence the re-imaging work.  It is envisaged that Glasgow should have a fully functioning dual stack test cluster which will hopefully show up issues with the current software implementations as we will try to emulate the production cluster as closely as possible. Certain pieces of software such as AFS do not support IPv6 presently and may not in the future.
> 
> David Kelsey is the primary point of contact for the HEPIX programme. The primary focus this year for the programme will be component testing of software for various Experiments throughout the collaboration. How far we get on this is partially down to how much feedback we get from the Experiments on what software components are critical to current and near future operations. 
> 
> Jeremy, should we have a section on the GridPP Wiki for this programme?
> 
> regards,
> Mark
> 
> 
> 
> --------------------------------------------
> Mark Mitchell,					
> ScotGrid Technical Co-ordinator,		                 
> Rm 481, 
> Kelvin Building,                      
> School of Physics and Astronomy,             
> University of Glasgow,                  		
> G12 8QQ, UK      
> Telephone: +44-141-330 6439
> E Mail: [log in to unmask]
>