Very good points and clarifications. :-)
Regarding "are border nodes allowed", I think if inter-domain communication is 
required, the border node should allow, shouldn't it?

Also I am wondering whether this requirement is only applied to H-PCE case? 
H-PCE could be a typical use case. However I believe SVEC tuple is also applied 
to other use cases
Otherwise why RFC5440 specify it In more generic way.

Regards!
-Qin
From: Ramon Casellas [mailto:[email protected]]
Sent: Friday, September 13, 2013 3:49 PM
To: Qin Wu
Cc: [email protected]
Subject: Re: [Pce] A comment regarding domain diversity in 
draft-ietf-pce-hierarchy-extensions

Qin,

Please see in-line tagged with Ramon>

El 13/09/2013 9:22, Qin Wu escribió:
Sounds reasonable, I am wondering whether it is sufficient for just defining 
one new 'D' flag in the SVEC object? Why 'D' flag was not defined when RFC5440 
was documented.
How do I know computed paths don't have any transit domains in common.
Ramon> there are several things to consider:

*) the D flag is generic, attached to the SVEC tuple, only to convey a group 
request constraint. This applies regardless of the actual method used in 
computation. We should discuss what are the implications, such as "are border 
nodes allowed (corner case)? Of course, a PCC can also play with inclusion and 
exclusion provided it has the information on the interconnection of domains, 
but I don't see this as a common case.

*) I understood the requirement was scoped to the H-PCE case, where it is the 
responsibility of the p-pce to ensure that they are domain-disjoint when 
selecting the domain sequence for both paths, and before segment expansion. How 
the parent does this I would say it is is implementation specific (k-shortest 
disjoint paths, iterative, etc.)

*) I would say that, without more information, a PCE or PCC cannot simply 
deduce whether two paths don't have any transit domains in common, unless this 
information is conveyed in the path ERO or attributes. I was thinking that the 
use of domain objects could be used
http://tools.ietf.org/html/draft-ietf-pce-pcep-domain-sequence-03

if the EROs are "tagged" with domain information, it can be deduced whether 
both EROs are disjoint or not


Domain diversity is more complicated when computed paths don't even share 
ingress domain and egress domain.

Ramon> I guess we agree :o)

Thanks
R.


_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce

Reply via email to