[rrg] Renumbering can never be secure, testable, reliable or routine - draft-carpenter-renum-needs-work

[Eliot Lear]

Hi Robin,

> Short version:   Does anyone agree with Eliot's view?
>
>                     "... our scaling problem is NOT with enterprise
>                      networks, but with home and personal networks."
>
>                   I don't.  I think that we should aim for a single
>                   solution to be attractive to end-user networks
>                   from universities and corporations to DSL-connected
>                   SOHO users.
>    

Let's not conflate the problem with the solution.  By problem I mean 
that whatever number of enterprises are multihomed can and should be 
dwarfed by the number of home users and small/medium businesses (SMBs) 
in the timeframes we're discussing.  Whether the solution fits both is a 
separate discussion.  Ideally you want one, but sometimes the best 
approach is a short menu.

> OK.  My "design assumption" is that since we can't move to a
> completely new set of protocols which would free us from the
> requirement that IP addresses be generally stable, and that since
> renumbering is a major upheaval for the network, and cannot be
> reliably, securely automated that we cannot base a scalable routing
> solution on "routine renumbering" of end-user networks.
>    

And I agree.  I do not see this happening in even the medium term.

> My belief, argument, assumption or whatever is based on things which
> I think can never change as long as we use the current protocols -
> which is itself something I think we cannot change in any time frame
> relevant to the routing scalability problem.
>    

There are many reasons why I buy into the above design assumption.  As 
someone else wrote, many of the problems are not with protocols but both 
in implementation and also with the trust model of DNS.

>
>> In those latter two cases I hold out more hope.  If enterprise
>>      
> networks
>    
>> were merely a 2^6 exception, we would be in good shape.
>>      
>
> I don't understand your last sentence.
>    

What I am saying is that if we had 2^6 enterprise networks announcing, 
using BGP, we could survive and survive well.  We can debate about the 
2^6 #, of course.
> While I wouldn't completely rule out different scaling solutions for
> radically different sizes and types of networks, I have argued that
> we should try to find a single solution which works equally well,
> and is highly attractive, for networks of all sizes, from university
> and corporate networks to single homes and offices currently
> dangling from a DSL line.
>
>    Map-encap space only for small end-users? PI space prices
>    http://psg.com/lists/rrg/2008/msg00454.html  2008-02-22
>
> If there were two solutions, or no solution for "larger" networks,
> then those smaller organisations which aim to be large will tend to
> adopt the "large" solution (if there is one) from the outset.
>    

This very much depends on the upgrade paths available from one to 
another.  If it's easy to go from small to large solutions, or if the 
cost of going to a large solution is very high no matter what you're 
using, then I would think small organizations would start small.

Eliot