below...

Bob Hinden wrote:
> 
> Keith,
> 
> >operationally I think it would be a mess to have site-locals routed
> >differently within a site than globals.  it's not that you can't do it,
> >it's that it makes life more difficult, and GUPIs seem to be a better
> >way to solve the same problem.
> 
> I am not sure there is that much difference.  In the current /48 provider
> based global addresses, each subnet is identified by a /64 prefix.  Both
> approaches will generate /64 routes.  I suspect that only in the largest
> sites, will the subnet field also be used for aggregation inside of the site.
> 
> This is all about tradeoffs.  For example in the site-local prefix
> FEC0::/10, there are 54-bits available for global identification and subnet
> numbering.  If 16-bits are for subnets, then there are 38-bits left for
> global site identification.  Is this big enough for a global
> token?  Probably not if it a random number or based on some other existing
> global identifier.  Might be OK if it was centrally allocated, but who is
> going to do that?  Also, what about sites that need more than 16-bits of
> subnets?
> 
> We could use a shorter prefix, but how much of the total IPv6 address space
> do we want to use for this?  For example, a /2 prefix would allow 16-bits
> of subnets and a 46-bit token.  But this would use 1/4 of the total IPv6
> address space.  That doesn't seem wise.
> 
> So I am not sure there is any solution that has the three properties:
> 
>   1) globally unique
>   2) 16-bit subnet field
>   3) Uses a limited amount of the IPv6 address space
> 
> We may have to pick the two we think are the most important.  The current
> site-local definition has 2) and 3).  My global site local proposal has 1)
> and 3).  Andrew White's draft has 1) and 3).

There is the alternative that can definitely be satisfied:

  1a) probably unique
  2) 16-bit subnet field
  3) Uses a limited amount of the IPv6 address space

i.e. hash something into 38 bits that has a very high probability of being
unique among a finite number of sites.

   Brian
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to