On Dec 19, 2012, at 1:13 PM, Randy Bush wrote:

>  o there is a useful poster child for the need for O(minutes) rpki (or
>    perhaps just roa) propagation.  as folk have repeatedly said, we
>    need concrete examples and goals.

I really think these should feed requirements analysis too [1].  I think there 
are a few other easy examples of this type of use case, but for requirements, 
maybe we need just one good one?

>  o unfortunately, there is no internet technology today to completely,
>    accurately, and securely maintain a globally distributed database
>    (e.g. at every non-trivial pop) with time constraints on the order
>    of a minute.  it's a research problem.  and a *really* interesting
>    one.

I feel _this_ is where a lot of our group's false-logic starts.  This (and 
similar) statement(s) assume that we must deploy a global replicated state 
machine, which has a complete and coherent image of a globally distributed set 
of resource holders' states, at all RPs at all times in order for validation to 
succeed.  Moreover, the state of this global replicated state machine has no 
semantic for data authorities to inform or influence the coherence of 
replicated data held by RPs (in caches, etc).  It is true that this model is 
mandated by the current _design_, but I don't see that this is motivated by any 
real engineering _requirement_.  Why do we need to crawl all repos, for all 
resource holders in order to route?  What motivates _this_ as the only way to 
accomplish routing security?

In the spirit of sound engineering: since this aspect of BGPSEC is not derived 
from a requirement, and it is limiting us from reaching other required goals 
(see above), it ought to be reconsidered/thrown out.  afaik, RPKI+BGPSEC (note 
the BGPSEC part, not just RPKI) is something of an outlier in the field of 
Internet-scale systems, in that it requires everyone to have this full globally 
consistent cryptographic state maintained reliably at all RPs.  I point this 
out just to underscore the notion that we should all be circumspect in our 
optimistic appraisals of this design's operational viability.

I feel this design element needs to be challenged, and if it cannot be 
supported by requirement it should be removed from the design.

Eric

[1] http://www.ietf.org/mail-archive/web/sidr/current/msg03671.html
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to