Re: [v6ops] SLAAC renum: Problem Statement & Operational workarounds

[Mark Smith <[e-mail]>]

On Mon, 28 Oct 2019, 06:06 Brian E Carpenter, <[e-mail]>
wrote:

> On 28-Oct-19 05:29, Fernando Gont wrote:
> ...
> > Robustness has a lot to do with being able to gracefully handle the case
> > when others screw up.
>

> This. We can't make the world perfect, but we can make individual
> components
> that expect imperfection.
>


This is not what I said, as it is implying that I said there is no need for
robustness.

Robustness suffers from the law of diminishing returns. At some point the
cost to design and implement robustness exceeds the either the likelihood
it will be needed or the benefit it returns if needs to be used.

Sugar is sometimes poured into the fuel tanks of cars, which will destroy
the engine. It is not common and is done maliciously. If robustness should
be universal, does that mean car engines should be made impervious to
sugar, at great expense to everybody, despite the need for that robustness
being rare?

Some ISPs are shifting the cost and consequences of their design decision,
which they can change, onto the IETF, CPE vendors and host implementers,
and indirectly their customers'. Are they willing to pay these other
parties to fix a problem that they are both causing and could fix
themselves?

(Can I bill these ISPs for the time I spent implementing the
'DeprecatePrefix' and 'DecrementLifetimes' options in 'radvd' in 2011 that
address this problem? Those options have limitations that I still don't
think are acceptable, which is why ISPs fixing their backend systems is
still the best solution.)



>     Brian
>