Re: GSE IDs [Re: IETF multihoming powder: just add IPv6 and stir]

[Brian E Carpenter <brian@hursley.ibm.com>]

David Conrad wrote:
> 
> Brian,
> 
> On Wednesday, May 7, 2003, at 02:08  AM, Brian E Carpenter wrote:
> >> Or you simply rewrite the first 48 bits when the packet traverses the
> >> border into the destination site with the same value the packet had
> >> prior to traversing the border of the source site.
> > This was not part of any GSE proposal I ever saw,
> 
> Different terminology would be helpful.  Of course, this implies my
> messages on this topic have been whistling past people for the last two
> months.
> 
> > and it involves stateful
> > distribution of mapping information.
> 
> Yes.  In one approach, a mapping of a globally unique 48 bit value to a
> set of other globally unique 48 bit values where that distributed
> lookup of that mapping occurs at the source border only.
> 
> > A very different beast from GSE,
> > and it sets off my stateful=bad alarm.
> 
> Your alarm must go off continually today,  as the mapping is more or
> less equivalent to either the DNS or routing tables, depending on your
> point of view.

Right, and we know that faulty DNS setups cause connectivity glitches
on a daily basis - because DNS state is created and managed by every site
worthy of the name, and mistakes are inevitable. We also had years of 
connectivity glitches caused by BGP4 configuration errors - most operators 
seem to have that figured out now, but it was bad for a while. Introducing 
a third distributed state mechanism that will be critical for connectivity 
is indeed a big step. 

   Brian