IPv6 PI allocation

[David Sinn]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


On May 17, 2007, at 7:32 AM, Jun-ichiro itojun Hagino 2.0 wrote:

>>> 	so cisco is assuming that /64 boundary on all the routes, right?
>>> 	i guess cisco needs some clue injection - i'm happy to do that if
>>> 	they want to.  see /64-or-/128 discussion on my previous email.
>>
>> that does not matter. 144 bits does cover 128 bit anyway.
>
> 	don't we need to keep:
> 	- destination/prefixlen, which would be 128 bits + prefixlen(8bit)
> 	  at least
> 	- nexthop, which would also be 128 bits (could be a index of
> 		routing table, but that would be 20 bit for 1M entries)
> 	for each entries?

He's only talking about the lookup space needed in the TCAM.  The  
size of the results of that lookup is identical for each family since  
it just points on to a further lookup in a adjacency table.  Since  
the platform in question can only have 64k next-hops you only need to  
return 16-bit from that lookup for the follow-on lookup that results  
in the actual next-hop entry.

David
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (Darwin)

iD8DBQFGTK3hLa9jIE3ZamMRAmI3AJ9gcV3tJ2pQvkSk2gCLM34HHsEbEACg84eZ
91+KjcjASzBmVSSUAnw+mQ0=
=5yyg
-----END PGP SIGNATURE-----