202203071610.AYC Re: Making Use of 240/4 NetBlock

Sun Mar 13 08:29:23 UTC 2022

On Sat, 12 Mar 2022 at 18:19, Abraham Y. Chen <aychen at avinta.com> wrote:

> 3)    " ... Changes to hardware and software to make use of 240/4 as ordinary unicast IP addresses can and should proceed in parallel to such debate.  ":     Agreed. Since through the EzIP Project, we have identified that the hardware does not need change, and the software modification is minimal, we should proceed to discuss what is the best application for the 240/4 netblock, after is re-classified as an ordinary unicast address pool.

It would surprise me greatly if there isn't hardware in the field
which physically cannot be retrofitted for 240/4, as we have a very
diverse set of hardware which very liberally makes assumptions of the
environment it will be in, to both reduce cost and to improve PPS.
These assumptions cause issues already in the environment they are in,
because engineers aren't always correct. It is crucial to understand
not all lookups are equal, and determinism isn't perfect, the devices
are not complex, but they are more complex than that.
And of course a lot more which by software do not, and will not, as
they've stopped receiving software updates. The marginal increase in
cost and effort in any of these cases between CPR and IPv6 is trivial.

Any narrative to prolong dual-stack with the argument 'it is just SW
update' is broken. Only reason we are not 100% IPv6 today, is because
we failed to foresee IPv6 won't take off, we all kind of assumed it
will of course happen organically in due time. I don't think anyone
really believed in 2002 that in 2022 we are still IPv4 and IPv6 is
after thought. And now we should understand, the organic market-driven
move to IPv6 may not ever happen, and are we going to accept all this
cost involved maintaining dual-stack or do we want to reduce CAPEX and
OPEX and have specific plans to return to single-stack world?

What if many/most large CDN, cloud, tier1 would commonly announce a
plan to drop all IPv4 at their edge 20 years from now? How would that
change our work? What would we stop doing and what would we start
doing?

-- 
  ++ytti