Last Call Review of draft-ietf-teas-lsp-diversity-07

Request Review of draft-ietf-teas-lsp-diversity
Requested rev. no specific revision (document currently at 10)
Type Last Call Review
Team Routing Area Directorate (rtgdir)
Deadline 2017-05-24
Requested 2017-05-05
Requested by Deborah Brungard
Authors Zafar Ali, George Swallow, Fatai Zhang, Dieter Beller
Draft last updated 2017-05-23
Completed reviews Rtgdir Last Call review of -07 by Bruno Decraene (diff)
Secdir Telechat review of -08 by Benjamin Kaduk (diff)
Genart Last Call review of -08 by Linda Dunbar (diff)
Opsdir Last Call review of -08 by Ignas Bagdonas (diff)
Prep for Last Call.
Assignment Reviewer Bruno Decraene
State Completed
Review review-ietf-teas-lsp-diversity-07-rtgdir-lc-decraene-2017-05-23
Reviewed rev. 07 (document currently at 10)
Review result Has Issues
Review completed: 2017-05-23



I have been selected as the Routing Directorate reviewer for this draft. The Routing Directorate seeks to review all routing or routing-related drafts as they pass through IETF last call and IESG review, and sometimes on special request. The purpose of the review is to provide assistance to the Routing ADs. For more information about the Routing Directorate, please see ​<>

Although these comments are primarily for the use of the Routing ADs, it would be helpful if you could consider them along with any other IETF Last Call comments that you receive, and strive to resolve them through discussion or by updating the draft.

Document: draft-ietf-teas-lsp-diversity-07
Reviewer: Bruno Decraene
Review Date: 2017/05/23
IETF LC End Date: not initiated AFAIK
Intended Status: Standard Track
I have some minor concerns about this document that I think should be resolved before publication.
I have only very basic knowledge of RSVP-TE and none of GMPLS. I found the document very clear. Thank you.
Major Issues:  None
Minor Issues:
To provide diversity, my understanding is that the signaling of the second LSP is extended to detect the crossing of the first path, and then is re-routed to avoid it. In some cases, I guess that providing path diversity is not possible without re-routing the first path. It's not clear to me how this is possible with this proposition. If it’s not, this limitation should probably be highlighted somewhere in the document. (and if it is, sorry for having missed it). Especially since this case seem to be in scope as per the Introduction: "Similarly, an LSP from EN2 to EN3 traversing CN1 needs to be diverse from an LSP from EN2 to EN3 going via CN4. [...]  This document addresses these diversity requirements "
"There are scenarios in which the ENs have the following requirements"
OK. Are there other scenario? (seem so as per the wording). What are their requirements and why aren’t they considered?
"-  Both client and server understand the identifier."
Not sure what is meant by "understand". At this step, it's not clear why the identifier can't be an any number/string that no one understand but is treated as an opaque value/identifier.
"The identifier is to be stable for a long period of time."
Easy comment but "long" is not very specific and different person may have different interpretation. I guess that the goal is that the identifier be stable for the duration of the diversity requirement.
" These requirements are met by using the LSP identifier. The LSP
      identifier uniquely identifies an LSP in the network and
      comprises of the following fields: IPv4/IPv6 tunnel sender
      address, IPv4/IPv6 tunnel end point address, Tunnel ID, LSP ID,
      and Extended Tunnel ID. These fields are defined in [RFC3209],
      sections and"
It's not clear to me that this choice meet the requirements. As per my quick reading of RFC 3209, Tunnel ID remains constant for the lifetime of a tunnel. If the node (e.g. EN1) reboot, it seems plausible that this Tunnel ID be changed. That seems to be an issue given that this identifier seems to be statically configured on the other node (e.g. EN2)
"In order to
      maintain diversity between these two connections within the core
      network, it is assumed that the core network implements Crankback
      Signaling [RFC4920]."
Don't you mean :s/assume/REQUIRED  ?
§1.2 PCE-allocated Identifier
If PCE is used to compute the paths, paths diversity needs to be handled by the PCE (to compute diverse paths). In which case, the problem seems to be solved with no need for further RSVP-TE extensions.
Figure 2 seem to illustrate that this section seem to consider the case where PCE is only partially used (in one domain). This is not stated at the beginning of 1.2 hence this does not help the reader to understand the case really being considered.
"the concept of a Path Affinity Set (PAS) is defined for abstracting SRLG information."
It's not clear to me how the second node (EN2 using the example from the introduction) is communicated this PAS from EN1, and how it's kept up to date as the path used by EN1 may change.
IOW, this PAS does not seem to fulfill the 3 latest requirements listed in 1.1:
"       -  It is necessary to be able to reference the identifier even if the LSP referenced by it is not yet signaled.
        -  The identifier is to be stable for a long period of time.
        -  The identifier is to be stable even when the referenced LSP is rerouted."

It's not clear to me what benefits this PAS brings compared to the client initiated identifier. While it adds the above disadvantages.
"The means by which the processing node determines the path corresponding to the PAS is beyond the scope of this document."
But this doesn't remove the problem to be solved. You seem to assume that there is one single database, typically centralized. An alternative option may have been to cypher the "detailled SRLG list". Why has this option been discarded?

"The means to distribute the PAS information within the core network is beyond the scope of this document. For example, the
      PAS and the associated SRLG information can be distributed within the core network by an Interior Gateway Protocol (IGP) or by other means such as configuration."
I don't think the use of the IGP would be such a good fit, in term of frequency of update (ok, this is deployment dependent) and scalability (a priori the number of PAS is o(N^2), N being Core Nodes.)
Abstract: " This document specifies three new route exclusion types."
From the IANA section and the document ToC, I'm seen only 2 types: "IPv4 Diversity subobject", "IPv6 Diversity subobject"
"IANA section"
You do not define a registry nor a registration policy for the "DI Type"
The A-Flags field is 4 bits longs and this document already allocates 4 flags, meaning that there is no room for extension (although there is a 4 bits Reserved field available)
If A-Flags "0x01 = Destination node exception" and "0x04 = Penultimate node exception" are both set, it seems that the penultimate link (between these 2 nodes) should also be excluded from the exclusion list. If so this should be specified.
"When the diversity identifier type is set to "IPv4/ IPv6
              Network Assigned Identifier", the value MUST be set to the
              IPv4/ IPv6 address of the node publishing the Path
              Affinity Set (PAS)."
Given that the way the PAS is advertised (I read 'publish') is out of scope of this document, I'd rather not use the term "publishing". I guess "allocating" would be better and probably more accurate.
"the diversity subobject must be kept while other subobjects may be removed."
Do you mean :s/must/MUST  ? (it looks to me that this is required for interop, hence a MUST)
"all Diversity subobjects in an XRO/ EXRS MUST contain the same Diversity Identifier Type."
Could you clarify the reason?
What if someone wants to be diverse with 2 others LSP: one intra-domain using a client-Initiated Identifier, and another one inter-domain using a "PCE-allocated Identifier"?
"it MUST return a PathErr with the error code TBA3 "Routing Problem" and error value of "Unsupported Diversity Identifier Type"
I think I would propose
"code "Routing Problem" (24) and error value of "Unsupported Diversity Identifier Type (TBA3)"
"The transit nodes in a domain and the domain egress node SHOULD NOT process the signaled diversity information."
This does not seem to match " In order to
      maintain diversity between these two connections within the core
      network, it is assumed that the core network implements Crankback
      Signaling [RFC4920]."
As I understand, by the latter, that all RSVP-TE nodes need to process the diversity information. But this may comes from my lack of knowledge of RSVP-TE.
Impressive list of contributors: 1,5 pages, 16 persons (in additions to 4 authors)