[ietf-dkim] Issue 1386 and downgrade attacks

[Dave Crocker]

Eric Allman wrote:
> [By the way, there was also some confusion about whether transitions are 
> O(years) or O(days).  Changing selector records is O(days), whether or 
> not those selectors change algorithms, but changing algorithms requires 
> software updates and hence is O(years).]

Important distinction.  Thanks.

It's probably worth noting that a catastrophe with a deployed algorithm, so 
that a rapid transition is required, has no precedent in the large-scale, open 
Internet, and probably would take considerably more effort and mechanism than 
anything we are discussing here.

As such, building in anything designed a) to deal with highly problematic, 
systemic failures, and b) incurring overhead for most/much regular traffic in 
anticipation of that catastrophe is probably not such a good idea.

As we have seen in other such algorithm transitions for mechanisms in 
end-points -- rather than infrastructure -- they tend to have a distinctive 
characteristic:

      While it is O(years) to achieve very broad adoption, it can be O(months 
or even weeks) to gain a useful degree of adoption, within smaller communities 
of interchange.

In general, this means that slower algorithm transitions are acceptable and 
can be handled in the same way as we handle other transitions on the Internet. 
  None of them include a publication mechanism.


d/
-- 

   Dave Crocker
   Brandenburg InternetWorking
   bbiw.net