[rrg] Agenda request: Presentation on new host stack architecture

[Christian Vogt]

> Therefore, your proposal addresses a very important architectural
> problem of the Internet. If deployed, it would allow a much easier
> deployment of new techniques, whether HIP, LISP, IPv6 or anything  
> else.

Hi Stephane -

Thanks a lot for your review and feedback.  It is highly appreciated.
And I apologize for getting back to you with delay.

> * a weaker form of your proposal is implemented in many programming
> languages (even in C if you use libraries like neon). The program can
> connect to a program on another host using just host names (for
> instance, I believe Christian Huitema mentioned several times here  
> that
> there is such an API in Microsoft products). It is weaker than your
> proposal since everything is implemented in userland and therefore  
> such
> connections typically do not survive a renumbering or rewriting.

That's right.  And I think the popularity of these evolved APIs is a
good indication that application developers will adopt also the new API
provided by a hostname-oriented stack architecture.

Also, you are right that the existing evolved APIs are weaker than a
hostname-oriented stack:  First, because they do not provide an Accept
 From Hostname method.  Second, because they cannot handle address  
changes
without application-layer reconnects.  A hostname-oriented stack would
provide both.

> * at least for debugging purposes, it would be great to be able to
> retrieve technical connection details such as the IP addresses  
> actually
> used. Should you plan to develop a concrete API, this would have to be
> handled.

Yes, I agree that this would be useful and necessary.

> * Security is of course the big problem and the current proposal is a
> good start, but insufficient.

Are you referring to hostname registries potentially not being
trustworthy?

> * Your plan would make us more dependent on the DNS. Today, an
> application may run entirely without the DNS, which would no longer be
> possible with your plan. Disclaimer: I work for a domain name registry
> so I find it a very good idea :-)

Right, a hostname-oriented stack would make DNS a first-class entity.
I believe this is feasible because it is true for many applications
already today.  Having said this, I also acknowledge that there are
mission-critical applications that must continue functioning in the
event of a DNS failure.  It may be necessary for those applications to
operate on IP addresses directly.  I envision a non-default mode that
enables this.  Note that a similar mode will be necessary to support
legacy applications.

- Christian