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

Eliot Lear lear at cisco.com
Mon Oct 27 08:02:04 PDT 2008


Robin,
> Also, I think you meant 10^6 rather than "2^6", meaning a million
> enterprise networks:
>    

Uch.  Indeed.  Thanks for the catch.

>    
>> 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.
>>      
>
> I guess from this that you can imagine the DFZ being OK at a million
> plus prefixes.
>    

I don't know about the "plus", but we really aren't all that far from a 
million now (I'm thinking orders of magnitudes).

> Maybe it can (I tend to think not), but doesn't the average
> enterprise network have lots of physical sites, each of which it
> would want to multihome via two or more local ISPs?

Of course we always get into trouble when we talk about "average".  I 
think what often happens is that an enterprise hands a contract off to a 
service provider for both internal and external connectivity.  Discount 
structures being what they are, I suspect this will tend to be the case 
into the future.  There are other models too that would scale more 
poorly than the existing the routing system, but the degenerative case 
is certainly the walking human network of devices.  And that may be a 
case we choose not to handle within the routing system.


>  From your
> message, I got the impression you thought a million enterprise
> networks could be accommodated without any new system, using current
> BGP techniques.
>    

Yes.

> What assumptions are you making about the total number of prefixes a
> million "enterprise" networks would advertise?

O(10^6) routing entries (yes, I have that right, thanks).

>    Or did you mean a
> million prefixes for all enterprise networks, which I think is a
> very much smaller number of enterprise networks.
>    

And of course I would have to say that any system that escapes these 
engineering boundaries is preferable to one that does not, other things 
remaining equal.

Eliot


More information about the rrg mailing list