Building

Why It Might Make Sense to Use IPv6 in Enterprise Infrastructure Projects

Looking at IPv6 deployment graphs like this one it becomes clear that IPv6 still is not widely deployed in enterprise space (the reason for the apparent oscillation in that curve is the difference between working days – where people use their office computers – and weekend where they preferably use their smartphones or their home equipment connected by means of broadband networks).

There’s a number of good reasons for this (in a nutshell: the overall IPv6 architecture is oriented around, and benefits, the decoupling of mostly autonomous, self-organized endpoints from a well-managed/provider-managed network infrastructure which isn’t exactly the operations model many large enterprise organizations have in mind for their networks. also you might have a look at these slides from RIPE74 to understand some of the reluctance to deploy IPv6 in certain companies).

On the other hand let’s consider how many enterprise-level networks projects with the following situations or attributes we see today:

  • the ambition to perform (network) infrastructure projects in an “agile”/driven-by-MVPs way not surprisingly creates cases where “some ‘provisional’ addresses” are assigned to segments & systems (without proper foresight or coordination within the organization, not least because the latter might require time & resources).
  • many projects are about bringing together/connecting previously separate[d] network segments & systems (think of “lift & shift“ type data center migration projects or all those occasions where you face terms from the “cloud exchange”, “express-sth” or “direct-sth” realm).

I have to say that the more projects I perform in such circumstances the more I value the e2e nature of global IPv6 addressing. Hence when asked about the benefits of IPv6 in enterprise space I nowadays usually have something along the lines of: “given the above type of phenomena I strongly advise to use IPv6 with global addresses from the very beginning in such projects as otherwise you/we’ll have to renumber later” in my response.

From my operational experience “Renumbering Still Needs Work” (as RFC 5887 states it in the title) is a euphemism at best. Avoiding it by clutches like NAT or whatever-type-of-overlay stuff doesn’t help either, complexity- and operations-wise (and, ofc, NAT wouldn’t work with IPv6 anyway).

So when you want (or have to) perform network projects from the above types it certainly makes sense to use IPv6 (with global addresses) from the beginning. Some lucky guy who doesn’t have to renumber (or NAT) in the future will thank you for that one day.

Have a great weekend everybody

Enno