Thus spake "E.B. Dreger" <eddy+public+spam@noc.everquick.net>
> When a 30Mpps IPv4 box falls back to <200kpps for IPv6, I
> don't think "not completely functional" is an adequate
> description. To me, that falls into the "not supported"
> category.
Why not use the highest-order 32 bits of an IPv6 address for
interdomain routing... i.e., "overlay" them on IPv4 addresses
and/or a 32-bit ASN? Yes, it smells of classful routing. Call
me shortsighted, but how many billion interdomain routing
policies do we really need?
Most L3 switches shipping today (e.g. the product in question) have
particular ethertypes and destination address offsets hardcoded into their
ASICs. It's not a matter of supporting 128-bit addresses -- they simply
doesn't understand IPv6's header any more than they do DECnet or AppleTalk.
While allocation policies may have an effect on how IPv6 FIBs are most
efficiently stored, address length is a fairly small part of the problem
when you're talking about redesigning every ASIC to handle both IPv4 and
IPv6.
S
Date: Thu, 12 Jun 2003 23:48:18 -0500
From: Stephen Sprunk
Most L3 switches shipping today (e.g. the product in question) have
particular ethertypes and destination address offsets hardcoded into their
ASICs. It's not a matter of supporting 128-bit addresses -- they simply
doesn't understand IPv6's header any more than they do DECnet or AppleTalk.
Yes...
While allocation policies may have an effect on how IPv6 FIBs are most
efficiently stored, address length is a fairly small part of the problem
...but there's no sense making the tries more unwieldy than they
need to be.
when you're talking about redesigning every ASIC to handle both IPv4 and
IPv6.
Less redesign required if reusing the existing IPv4 lookup.
Eddy
The whole 64 bits reserved for a link layer address thing seems silly, why
don't we just put some payload in there and make the packets a fixed
size... 
Petri Helenius wrote:
The inventors of tag-switching^H^H^H^H^H^H^H^H^H MPLS seem to be
firm believers in that if they don�t deliver the goods, the world will
stand still and wait.
And of course you can tag^H^H^Hlabel switch IPv6 through a non-IPv6-aware
core, it just breaks traceroute (as even if you pop a label off the stack
on TTL expiry, your core still can't understand the resulting packet to
return a TTL expired).
But then, much of said vendor's switching equipment with routing
functionality doesn't know about MPLS either - just IPv4.
David.
Date: Fri, 13 Jun 2003 09:37:59 +0300
From: Petri Helenius
One word; multihoming.
How many billion different interdomain routing policies do we
really need?
Eddy
Just 1 is enough to cause trouble. Given strict provider-based addressing,
multihoming leads to rather nasty interactions between host-based selection
of (source address, destination address) & things like the following:
o routing policy
o anti-spoofing... filtering
o quality of service
The IETF drafts I've read have not yet offered what I consider viable
solutions to those issues.