[discuss] Real world Impact of multiple roots

Suzanne Woolf suzworldwide at gmail.com
Wed Jan 29 14:07:47 UTC 2014

On Jan 28, 2014, at 4:35 PM, Brian E Carpenter <brian.e.carpenter at gmail.com> wrote:
> Facebook (etc.) are voluntary. If you choose to use a non-global service,
> a.k.a. a walled garden, that's your choice and I respect it. That is quite
> different from splitting the global namespace - I imagine that FB users
> would be quite annoyed if connecting to facebook.com took them to
> different services on different occasions. But of course somebody would
> soon implement a fix, so that users could choose between (I am making
> this up) facebook.com.oldroot and facebook.com.newroot. That's why
> computer scientists are so fond of recursion.

I'd agree with this but state it slightly differently: Facebook et. al. are applications, in the network sense. They provide certain services and a certain interface to users, and in turn they use certain infrastructure and rely on certain properties of that infrastructure. 

DNS is part of the infrastructure, along with IP routing and a very few other things. Changing the properties of DNS infrastructure, including introduction of new local dependencies for resolution, changes the properties available to applications and changes what applications that use them can deliver. Having global and unambiguous properties to the public DNS namespace is useful in ways that aren't all completely obvious, and changing it needs to be done carefully if we value that.

To a great degree, the assumption underlying both the technology and the policy discussions, and therefore much of the contention, is that a unified, self-consistent namespace is a valuable thing. Accordingly, it's tolerable and even occasionally useful to have inconsistent local namespaces at the edges of the network, but at the core there should be an infrastructure that's available to anyone without requiring prior arrangement or software compliance outside of published, open standards. In general, experience to date suggests that one of the properties at that core is preferred to be consistency at the root of the mathematical tree used to implement the namespace.

The question isn't whether it's possible to have multiple namespaces available to a user or a service, or how to do it (there's lots of ways, although you do have to pick one at a time if you're using DNS-- there really can be only one DNS root in that sense). The question is whether there should also be a namespace that's public, more or less global (i.e. available easily anywhere that the TCP/IP internet reaches), and consistent in its naming and behavior regardless of who uses it, for what, or from where. The answer to *that* question has seemed to be "yes," and the technical mechanism we have for it is DNS, so we're left with a question of coordinating its contents and properties to maintain and extend its usefulness as infrastructure.


More information about the discuss mailing list