[ih] IP addresses are not phone numbers, was Some Questions over IPv4 Ownership

[Scott Brim]

On 10/17/2010 17:58 PDT, Miles Fidelman wrote:
> Scott Brim wrote:
>> On 10/17/2010 12:43 PDT, Miles Fidelman wrote:
>>   
>>> I guess I should also add the case of roaming, where one wants to
>>> maintain TCP sessions while handing off link-level connectivity across
>>> assets owned by multiple carriers.
>>>      
>> When multiple carriers are involved, why does the hand-off have to be at
>> (so-called) link-level?
>>    
> Well... for current software to work right, maintaining a TCP connection
> requires that the underlying IP addresses remain the same (at least I
> think it does).  So somehow the IP numbers have to stay the same, while
> the underlying network connection changes.
> 
> It's not necessarily a link-layer function to do the handoff, but what
> is actually changing is happening at the link layer - one physical link
> is dropping, and a new one is coming up.
> 
> The hand-off is really a routing function - changing the mapping between
> a specific IP address and a specific physical link.  That shouldn't be
> too hard to do if one carrier is handing things from one asset to
> another, seems a bit more difficult when two carriers are involved - and
> brings us back to the original question of who "owns" the IP address
> when more than one carrier is involved. :-)

Got it.  You're trying to work with current software.  That can be done
below the L3/L4 monolith, or above it (in the app layer) and of course
we're trying to break the monolith apart.  Every mechanism has areas
where it's useful.