Date: August 17, 2007 12:32:50 PM EDT
Subject: Gen-art review of draft-ietf-ccamp-inter-domain-pd-path-comp-05.txt
reviewer for this draft (for background on Gen-ART, please see
http://www.alvestrand.no/ietf/gen/art/gen-art-FAQ.html).
Please resolve these comments along with any other Last Call comments
you may receive.
Document: draft-ietf-ccamp-inter-domain-pd-path-comp-05.txt
Reviewer: Elwyn Davies
Review Date: 16 Aug 2007 IETF LC End Date: 16 Aug 2007
IESG Telechat date: (if known) 23 Aug 2007
Summary:
I think this document needs significant work on the core description of the algorithm. I found s4 to be difficult to read and it appears to contain a number of ambiguities and inconsistencies that should be fixed before it goes to the IESG. The various sub-cases and routes through the algorithm are not very clearly expressed IMO. That being said, I think this is essentially a descriptive problem rather than any issue of principle. There are also a number of editorial issues (mostly need for acronym expansion) that need fixing in due course.
Comments:
s3.1: To avoid the sense of surprise when a Path message appears in the last bullet, I would suggest s/The path (ERO)/The path (specified by an ERO in an RSVP-TE Path message)/
Fixed.
s4, para 2/3: Presumably para 3 is talking about the 'next hop' being loose or abstract as is implied by items 1) and 2). It would be good to make this clear. It would also be worth inserting a sentence making it clear that if the next hop is strict there isn't anything to do, since otherwise one has to scan the entire section to verify that this is the case. It is just possible that I have misunderstood what is going on and some of the stuff *does* apply to strict hops - in which case the section needs a whole lot more clarity. There also needs to be some words about the 'simple abstract node' case.
You're absolutely right, this has been clarified.
s4: Sending of PathErr. This section states that PathErr messages 'SHOULD be sent' in several places. Whilst this is consistent with RFC 3209, both of these documents appear to be (or may be) inconsistent with RFC 2205 which does not appear to provide any alternative to sending a PathErr when there is a problem processing a Path message. If it is deemed correct to allow not sending the PathErr, reasoning should be given as to why this might be desirable, and what is the alternative (presumably silently ignoring the message?) and its consequences.
As pointed out in the text, the sending of a Path Error message is compliant with section 4.3.4 of RFC3209.
s4, item 1): I think the first sentence ('If the next hop is not present in the TED.') is probably redundant and is certainly confusing.
Indeed, we removed the redundant text, thanks.
Part of the confusion is that the rest of the item appears to only concern loose next hops but there is no pointer to what happens with the other case of abstract nodes. I think the description would be more logical with the paragraph on abstract nodes, from later in the section, moved to before item 1). In that way the case splitting (strict/loose/abstract) would be much clearer. BTW doesn't the simple abstract case have to do some of the non-simple work?
I see where the confusion came from. I added a paragraph clarifying that s4 only applies to the case where
the ERO contains a loose hop or a non simple abstract node (an abstract node than identifies more than one
LSR).
s4, item 1, bullet 2: Which domain has to be PSC? The current one or the one containing the next hop?
The Next hop domain (text added).
s4, item 1: Worth making even clearer that *both* conditions have to be satisfied.
Fixed.
s4, item 1: 'If the next-hop is reachable, then it SHOULD find a domain boundary LSR...': What does 'it' represent here?
The "boundary LSR" (text fixed).
The path necessarily crosses the boundary (unless we have some very weird topology here ;-) ) so there is *something* on the domain boundary. What else could it find on the boundary? I think this is probably a badly expressed story about some other point. Reflecting on this, it strikes me that this is the key point where the routing information is pulled into the TE and this is very poorly expressed IMO.
The point is in that case to find the boundary LSR: "If the next-hop is reachable, then it SHOULD find a domain boundary LSR (domain boundary point) to get to the next-hop. The determination of domain boundary point based on routing information is what we term as “auto-discovery” in this document."
s4, item 1: 'the ABR in the case of inter-area TE or the ASBR in the next-hop AS in the case of inter-AS TE should be the signaled loose next-hop in the ERO': Does this mean in the expanded ERO or the original one?
The original one.
If it is the original one, how is the implementor supposed to check it is an ABR/ASBR?
In the absence of auto-discovery, the user must indeed know and configure all domain boundaries. If there is a succession
of IGP areas and ASes for example, as pointed out in the text, the ERO must contain all boundary nodes (ABRs/ASBRs).
s4, item 2): The term 'ERO expansion' is not defined in any of the standards - it is referred to as an alternative shorthand in RFC 4736 (requirements doc). It needs to be defined.
It is actually defined in RFC 4736, but you're right that it should also be mentioned here. We added some text.
"an ERO expansion (process consisting of computing the constrained path up to the next loose hop and add the list of hops as strcit nodes in the ERO)"
s4, item 2), bullet 1: This section contains 3 instances of 'SHOULD'. I am happy with the first one (policy applies). The third one is covered by the discussion on PathErr above. The second 'SHOULD' strikes me as a 'should' or even 'is'. What else would be done if it isn't treated this way? Bullet 2, sub-bullet 1 has a similar construction.
Very good point indeed. "If no path satisfying
the set of constraints can be found, then this SHOULD be treated
as a path computation and signaling failure "
is actually non appropriate and should read "If no path satisfying
the set of constraints can be found, then this is treated
as a path computation and signaling failure "
s4, item 2), bullet 2, sub 1: The condition at the beginning of this bullet could probably be written down more clearly.
If OK with you we kept the text unchanged since the terminology matches with the other companion draft (ccamp-inter-domain-rsvp-te)
s4, item 2), bullet 2/sub 1: 'If the boundary LSR is a candidate LSR for intra-area H-LSP/S-LSP setup (the LSR has local policy for nesting or stitching),...': Which LSR has the policy? The boundary LSR or the one asking? There is an equivalent question for sub bullet 2.
Right - The "boundary LSR" (text added in both places).
s4, para after item 2): 'Note that in both cases, path computation and signaling process may be stopped due to policy violation.': Does this mean some other policy violation than is already documented? If not I think this para should be removed as it it just raises doubts as to whether there is some other cause.
You're right, text removed.
s4, 'optimization technique': I am confused about what protocol instance would accept the proposed flooded information if the IGP is not running on the inter-AS link concerned. Does this require some special functionality in the IGP?
This is because it is carried as a non routing information. The TE link is advertised by the IGP using
an opaque LSA in OSPF and TE TLV in IS-IS. Thanks to local configuration, the IGP only advertises
the Inter-ASBR link as a TE link even if there is no routing adjacency over that link. Note that there are
implementations supporting that mode.
ss4.1/4.2: I believe that each of the examples references one or other of the Figure 1/2 configurations - this should be made clear explicitly.
Right, text added.
s4.1.1, para 4: Although it is not strictly normative, harder advice should be given on back-off times to avoid DoS/congestion.
As you pointed out, this section is clearly not normative and the security aspects are being addressed separately.
s5: The 'tapping' problem may be well-known but not to me. It needs either a brief explanation or a reference.
OK I added an example.
Editorial:
Abstract: Expand IGP acronym (and add to terminology?).
Done.
s2, para 4: The term Head-end is not defined.
Text added: "LSR that inititiated the TE LSP"
s3.1, bullet 3: Expand ERO acronym.
Fixed + added to the terminology section.
s3.1: To avoid the sense of surprise when a Path message appears in the last bullet, I would suggest s/The path (ERO)/The path (specified by an ERO in an RSVP-TE Path message)/
Done.
s3.2: Expand ABR
s3.2: Expand ASBR
Done.
s3.2, Figure 2: s/(CE to ASBR =>/(CE to ASBR) =>/
Fixed
s3.2, last bullet: This contains no verb! The bullet could (usefully) say: 'The scenario supports an inter-AS TE LSP...'. This bullet might logically appear as the 2nd on the list.
I see a verb here ...
- The example depicted in Figure 1 shows the case where the Head-end
and Tail-end areas are connected by means of area 0. The case of
an inter-area TE LSP between two IGP areas that does not transit
through area 0 is not precluded.
s5, para 3: A reference to s2 would help.
Not sure why/where ?
s6.1: A reference to how/where the Make-before-Break procedure is defined would be good (RFC3209?).
Added, thanks.
s8: Expand FRR, MP and PLR acronyms.
Done + reference added.
s9, para 2: s/verison/version/
Many Thanks again.
Diff: draft-ietf-ccamp-inter-domain-pd-path-comp-05.txt - draft-ietf-ccamp-inter-domain-pd-path-comp-06.webarchive
draft-ietf-ccamp-inter-domain-pd-path-comp-06.txt