Re: [dbound] Updated draft-levine-orgboundary-03

[Casey Deccio <[e-mail]>]

On Thu, Nov 12, 2015 at 12:27 PM, John R Levine <[e-mail]> wrote:

> Yes, I know.
>
> My scheme will typically make one or two DNS queries regardless of the
> length of the DNS name.


What I'm saying is that "typically" does not dictate complexity; worst case
dictates complexity.


> Yours appears to do a tree walk for each name component.
>

As I have mentioned both in this thread and in the draft, iteration stops
at an NXDOMAIN--which if I understand correctly, is the equivalent of
hitting a wildcard response in your draft.  I welcome your detailed read
and review of the ODUP draft.

My assumption is that each request is likely to be for a particular
> application.  If you need to enumerate the boundaries for M services, that
> will take M times as long but that doesn't seem like a very common
> situation.
>

Assumptions are fine.  Commonalities are fine.  But again, if you get into
the realm of bounding algorithmic complexity, it's all about worst case,
upper bounding.  Worst case vs. anticipated performance, given the baseline
behavior, are also discussed in the ODUP draft.

I am belaboring this somewhat because we have drafts whose merits are being
discussed on this mailing list, and one of the concerns that has been
raised is "complexity".  Let's be clear about what that means, so the
comparisons are apples to apples.

Casey