Early Review of draft-filsfils-spring-segment-routing-ldp-interop-03
My previous message has been put on hold by the moderator of the SPRING WG as having too many recipients.
This is partially due to the draft having 12 authors – and have tried to send the review to each of them
I am now re-sending it with a shorter (but, hopefully, sufficient – I believe all the authors are on the SPRING mailing list anyway) list of recipients and
with one more issue I have found in the draft
Email: Alexander.Vainshtein at ecitele.com
Tuesday, July 28, 2015 7:37 PM
'rtg-ads at tools.ietf.org'
bruno.decraene at orange.com; jgs at juniper.net; Jon Hudson; 'Jonathan Hardwick'; 'rtg-dir at ietf.org'; 'spring at ietf.org'; mpls at ietf.org; 'cfilsfil at cisco.com'; 'sprevidi at cisco.com'; 'bashandy at cisco.com'; 'stephane.litkowski at orange.com'; 'Martin.Horneffer at telekom.de';
'igormilojevic at telekom.rs'; 'rob.shakir at bt.com'; 'saku at ytti.fi'; 'wim.henderickx at alcatel-lucent.com'; 'Jeff Tantsura (jeff.tantsura at ericsson.com)'; 'edward.crabbe at gmail.com'
RE: Routing directorate QA review of draft-filsfils-spring-segment-routing-ldp-interop
My name is Sasha Vainshtein, and I have been asked to perform the QA review of
Due to health problems I have been very late (but, hopefully, not too late) with this review, and here it is.
Does the draft solves a real problem
From my POV, definitely yes.
The problem stems from the following combination of
facts fact that
LDP distributes labels for FECs represented by IP prefixes (among other things).
LDP-based MPLS network are very widely deployed
SR operating over the MPLS DP allocates labels for SIDs that can represent IP prefixes (again, among other things)
As a consequence, LDP and SR may effectively map the same IP prefix to different labels.
As SR using the MPLS DP is gaining acceptance in the
industry, the following deployment scenarios look as realistic to me:
Coexistence between LDP-based and SR-based control planes in the same network
Partial deployment of SR in some sub-domains of a network combined with the operators’ need to use the benefits
of SR where it is already deployed
Gradual transition of services (especially L2 and L3VPN) tunnelled over LDP-created LSPs to LSPs set up using SR
(and, possibly, vice versa)
s the draft a good enough start for a WG
? From my POV, yes.
The draft covers multiple (if not all) realistic use cases for coexistence of LDP-based and SR-based control planes
and provides what looks to me as reasonable solutions for these scenarios.
At the same I would think that additional work is required to complete the work.
Specific issues with the current draft:
SR-related abbreviations (e.g., SRGB, SID etc.) are used without expansion at the first use, and there is no “Abbreviation” section in the draft.
The draft currently lists 12 authors on its title page exceeding the recommended RFC Editor maximum of 5 authors.
The draft does not analyse the use case of coexistence between SR and LDP with enabled extension for multi-area LSPs as per
. I think that this use case deserves special consideration in the draft
It mentions the use case of seamless MPLS and refers to the (expired)
Seamless MPLS Architecture
The seamless MPLS Architecture draft, in its turn, explicitly refers to RFC 5283 in Section 5.1.1 to facilitate
binding of labels received via the DoD LDP session while only default route is configured in the RIB of the access node in the seamless MPLS architecture
It is possible that this use case would not add anything to the draft – but I would prefer to see it in the document
with the appropriate explanation
Section 8 “Manageability Considerations” is currently left empty. When this section is written, I would expect at least the following:
Some level of detail regarding local policies defining whether LDP-created or SR-created LSPs should be used etc.
Discussion of using LSP Ping for LSPs that are produced by stitching an LDP segment with an SR one.
Alternatively, it is possible to move this section to a separate dedicated draft
I think more information should be provided in Section 5 to explain operational aspects of SR-assisted LDP FR:
In Section 5.1 I would expect the draft to explain that:
Node D is selected as the RLFA using the same logic that
is defined for selecting RLFA in RFC 7490
The “bypass LSP” is set up by the SR CP automatically
without any need for management intervention
I would like to understand why dynamically set up Targeted
LDP sessions are an operational concern. Such sessions appear also when BGP-based auto-discovery for L2VPN is used, and, according to the analysis in RFC 7490, the scalability problems associated with these sessions look as negligible.
In Section 5.2 I would expect the draft to explain that:
The resulting solution is similar to using a bypass LSP
to a manually selected RLFA that is set up by RSVP-TE (the difference is that a Targeted LDP session to such an RLFA would be still needed)
Set up of the “traffic-engineered bypass LSP” by the
SR CP is triggered by an appropriate management operation
What information should be provided by the Management
Plane for computing the RLFA and the traffic-engineered bypass LSP? Should it be similar to configuration parameters used with RSVP-TE?
Which kind of logic should be responsible for computing
the RLFA and the path to be taken by the engineered bypass LSP: should it be similar to the CSPF used with RSVP-TE?
How the traffic-engineered bypass LSP hat is set up by
the SR CP would react to additional changes in the network topology (the behaviour of the bypass LSP that is set by RSVP-TE in these situations is well known)
In both cases it is not clear how the backup label and NHOP are
FRR with RLFA as per RFC 7490 this is usually done by LDP that receives the backup label as bound to the destination prefix FEC via the Targeted LDP session, retains it and decides to us it when a route to the destination prefix using the LSP leading to the
RLFA as its NHOP is installed in the RIB as the best route to the destination prefix by IP FRR
scheme definitely would not work if the Targeted LDP session to the selected RLFA is eliminated
POV more details are required here, and the critical question here is, of course, when the scheme proposed in the draft requires any changes in LDP
Neither of the issues listed above should be considered as preventing adoption of the draft as a WG document IMO. Actually I think that resolving
these issues in the context of a WG document would be more effective.
Hopefully these notes will be useful.
Alexander.Vainshtein at ecitele.com
Jonathan Hardwick [
mailto:Jonathan.Hardwick at metaswitch.com
Wednesday, May 20, 2015 12:46 PM
bruno.decraene at orange.com
jgs at juniper.net
; Alvaro Retana (aretana); Jon Hudson
Routing directorate QA review of draft-filsfils-spring-segment-routing-ldp-interop
Please would you be the routing directorate QA reviewer for draft-filsfils-spring-segment-routing-ldp-interop?
This initial review has been prompted by the spring WG chairs in parallel with a call for WG adoption. As such, this document qualifies for
“initial QA review”. Please could you provide your initial comments in the next 3 weeks? As per the QA process, we would also like you to stay with the document and review it again when it goes to WG last call.
The following web page contains a briefing on the QA process, and guidance for the QA reviewer.
Please copy your comments to the spring and rtgdir mailing lists.
Please let me know whether you can do it, or not.