IPv6 – to dual stack or not? and when to start the transition?

I was particularly interested by this morning’s session on IPv6. All three talks gave me some interesting updates into the status of IPv6. 

I spent a lot of time planning and performing a similar network transition some years ago – 20 years ago I now realise. We were then moving the large worldwide High Energy Physics and Space (NASA/ESA) network to OSI routing, i.e. from DECnet Phase IV to Decnet Phase V.  What I learned then was that planning was the most essential aspect of the project plan.

We have been told that the IPv4 address space will run out in about two years for almost as long as I can remember. Many NRENs have been IPv6 capable for a (large?) number of years butstill end users are just not moving.

Today I learned from Dave Wilson’s talk that the current estimate for the exhaustion of IPv4 addresses in just one year away. So, it is already far too late to install a dual stack on all worldwide clients and servers. I heard today that we are going to have to deal with IPv6-only systems.

The good news is that there are solutions to connect IPv6-only systems to an IPv4 world via the use of NAT64 and DNS64 – as has been demonstrated on the network at this conference – some of you will have tried it here. The talk from Jean-Philippe Dionne explained how all this works.

We also heard from my colleague in the EGEE project, Xavier Jeannin, as to the successful work they have done to make sure that all of the EGEE gLite (Grid) middleware is IPv6 compliant. This was very good news, but there are apparently many other software packages we use in the HEP community which are not yet there.

So… what should we make of all this? For my user community (high energy physics), I have already been thinking for some time as to when would be a good time to start our move to IPv6. I am now fully convinced that now is definitely the right time. There was no consensus amongst the speakers as to whether to start with dual-stack or not, but to me that seems a good place to start.

What I do know, based on my earlier experience with the migration to DECnet Phase V, is that we (my scientific community) need to start the planning activity soon. We in the Grid will continue to experience a rapid growth in end system network addresses, particularly with virtualisation growing in popularity. There is lots of work to be done in documenting and analysing all of our software packages to see which are still not IPv6 compliant (reminds me of the Year 2000 problem!).  I think we should get started.

Advertisement