On Dec 6, 2012, at 5:11 PM, Eric Osterweil <[email protected]> wrote:

>> As Chris mentioned down-thread: Could you explain how these considerations 
>> differ from considerations around hosted services for critical services, 
>> e.g., DNSSEC?
> 
> This system (and therefore, its considerations) is/are VASTLY different than 
> those of DNSSEC... In a lot of ways:
> 1 - This repository system requires RPs to have a FULLY synchronized copy of 
> all objects at all times.  DNSSEC does not.
>       a - This means all systems have to be high performing/reachable/etc. or 
> there begin to be problems
>       b - This means that global crawling must be done before computation can 
> be reliable
>       c - …

DNSSEC requires you to have up-to-date information on all the things you care 
about.  It's just that in BGPSEC, everyone cares about everything :)

W.r.t. (a) -- Doesn't this argue in favor of hosting?  Isn't the "all" less 
burdensome if there are fewer things to keep alive?  I know I would rather have 
my web site on Dreamhost than on my on host in a colo.

W.r.t. (b) -- Again, this seems to argue in favor of the hosted model, since 
you have to crawl fewer repos.  

And it's not really true that you have to crawl the whole tree before you can 
do anything.  If you have a partial tree, then you can validate part of the 
ROAs.  Especially if you crawl intelligently, e.g., trying to avoid missing 
links in a cert chain. I believe that RPSTIR does something like this.


> 2 - This system's hosting model is proposing to put all of the object 
> generation information in a very few places (the hosted model), DNSSEC does 
> not (the hosting colocation coefficient in DNSSEC is vastly                   
>       different that 5, for rpki)
>       a - I am not even encouraged to have all my DNS zones hosted by RIRs... 
> I can get a registrar or hoster to do it, but I can choose from many many 
> options...
>       b - …

Not sure where you're going with this.  I thought you were arguing that the 
hosting model in general was a bad technology, which would not depend on the 
coefficient or which entities were doing the hosting.


> 3 - The downside of borking routing is all your [connectivity] bases are 
> belong to us... Lots of problems exist when destinations are unreachable that 
> do not exist (or are semantically different) when names are unresolvable.  In 
> short, these are very different systems, with _very_ different architectures, 
> with very _very_ different goals, etc.  I could go on, but is this sufficient?

Different services have different values for different networks.  There are 
certainly lots of stakeholders for whom loss of DNS reachability would be as 
bad as loss of IP reachability, as would seem to be attested by the predominant 
focus in the cyber crime community on DNS hijacking rather than BGP hijacking.


To be clear, I have no particular dog in this fight.  I'm not even really clear 
on why this discussion is happening on the SIDR list, given that it's a 
deployment decision between an CA and its customers.  

--Richard
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to