Sandy:

My responses in line below.
I was traveling last week and had limited access to my NIST email.
Hence the delay in responding. 

Sriram

> -----Original Message-----
> From: Sandra Murphy [mailto:[email protected]]
> Sent: Wednesday, April 07, 2010 4:13 PM
> 
> On Tue, 6 Apr 2010, Sriram, Kotikalapudi wrote:
> 
> >
> >> ________________________________________
> >> From: Sandra Murphy [[email protected]]
> >> Sent: Tuesday, April 06, 2010 3:34 PM
> >> .... snip ....
> >> As I understand proxy aggregation, an ISP receives updates from two peers
> >> and discovers that the NLRI in the two updates coincidentally aggregate
> >> and the ISP wants to announce the aggregate.  (If I'm wrong, I should be
> >> corrected - I'm thinking of description in rfc2519 and Geoff's example
> >> in http://www.potaroo.net/ispcol/2006-05/bgp.html.)
> >>
> >> So neither origin AS has authority over the aggregate prefix.  So neither
> >> origin AS would be able to issue a ROA for the proxy aggregated prefix.
> >> --Sandy
> >
> > You are right. What you describe may indeed be more often the case with 
> > proxy
> aggregation.
> > I was considering the other (may be less common) case when the deaggregate 
> > blocks are
> all owned
> > by one common entity, and originated by different ASs which are also owned 
> > by that
> entity.
> > In the case which you described (possibly the more common case), I agree it 
> > is not
> possible
> > to have multiple ROAs for the aggregate.
> 
> If I'm understanding, you are suggesting that the common entity (say
> AS900) holds an aggregate block as well as holding two other ASNs (AS901
> and AS902), that AS901 and AS902 are each originating a distinct more
> specific of the aggregate block and AS900 wants to propagate the aggregate
> route as an aggregation of the two more specific routes, rather than just
> originate a route to the aggregate block on its own.
> 
> That seems odd to me, but I've learned not to bet against the existence of
> odd routing policies.  I'd be curious to hear if this is a recognized
> corner case.
> 
> Your suggestion is that AS900 could sign a ROA allowing AS901 and AS902 to
> originate the aggregate, if I understand.  That is not what is really
> happening here; but the artificial construct would allow the aggregated
> route to validate.
> 
> > I would lean towards having such proxy aggregated updates receive a "Not
> > known" validation assertion during partial deployment period (if no
> > conflicting ROA is found), and disallow them (i.e., Invalid) when
> > deployment is considered to be complete.
> 
> By "such proxy aggregated updates" do you mean subcases like what you have
> just described?  Or do you mean cases of proxy aggregation in general?

I meant the general case as you stated previously,
"As I understand proxy aggregation, an ISP receives updates from two peers
and discovers that the NLRI in the two updates coincidentally aggregate
and the ISP wants to announce the aggregate."  

> 
> Are you suggesting a change to the roa-validation draft's handling of
> AS_SETS originations?

Yes, what I had in mind was that if the 9 use cases (w.r.t. AS_SETs)
that I outlined in my email (link below) were duly considered,
then the treatment of AS_SETs would be possibly much more adequate.
Not calling for any fundamental change in the _principles_ behind the
validation methodology; I was merely suggesting enumeration and
consideration of more (hopefully an exhaustive set of) use case scenarios
for updates involving AS_SETs. With this approach, the adversary cannot
exploit or misuse AS_SET to gain advantage. And also, any updates with
with a single ASN (put in once or repeated multiple times) in an AS_SET 
are duly treated, and can be declared "Valid" if there exists a 
ROA for the prefix with that ASN.
The email where I described the 9 use cases with AS_SETs is at:
http://www.ietf.org/mail-archive/web/sidr/current/msg01524.html

Probably no one is proposing changing the ROA structure -- 
nevertheless, I also think that changing the ROA structure to allow
inclusion of an AS_SET as origin is not worth the effort given that
there is an extremely small number of proxy originations observed
per Danny's email:
http://www.ietf.org/mail-archive/web/sidr/current/msg01519.html

> 
> Are you suggesting that the roa-validation draft should discuss different
> validity outcomes in different deployment periods?

I think this is not an issue really. I was merely thinking that what is 
considered
"Unknown" (i.e., when ROA is not found) during "partial" deployment would 
change to
"Invalid" if and when deployment is considered or declared COMPLETE(!) someday. 
But now I think that it does not matter if the algorithm never makes that 
transition. 
I think there is no harm in continuing on with the decision outcome 
"Unknown" when covering ROA is not found even far out into the future. 
(Here I am speaking in the context of what is already described in the two
ROA validation algorithm documents.)

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

Reply via email to