Re: [Lsr] Working Group Last Call for "Flexible Algorithms: Bandwidth, Delay, Metrics and Constraints" - draft-ietf-lsr-flex-algo-bw-con-07

[Ketan Talaulikar <[e-mail]>]

Hi Shraddha,

Thanks for your responses. Please check inline below for clarifications
with KT.


On Thu, Apr 4, 2024 at 9:49 AM Shraddha Hegde <[e-mail]> wrote:

> Hi Ketan,
>
>
>
> Thanks for the review and comments.
>
> Pls see inline for replies.
>
>
>
> Juniper Business Use Only
>
> *From:* Ketan Talaulikar <[e-mail]>
> *Sent:* Thursday, March 21, 2024 10:07 PM
> *To:* Acee Lindem <[e-mail]>
> *Cc:* lsr <[e-mail]>; [e-mail]
> *Subject:* Re: [Lsr] Working Group Last Call for "Flexible Algorithms:
> Bandwidth, Delay, Metrics and Constraints" -
> draft-ietf-lsr-flex-algo-bw-con-07
>
>
>
> *[External Email. Be cautious of content]*
>
>
>
> Hi All,
>
>
>
> I have reviewed this document and believe it needs some further work
> before publication.
>
>
>
> I am sharing my comments below:
>
>
>
> 1) There is the following text in sec 2.1 and 2.2 for Generic Metric.
>
>
>
> A metric value of 0xFFFFFF is considered as maximum link metric and a link
> having this metric value MUST NOT be used during Flex-algorithm
> calculations [RFC9350
> <https://urldefense.com/v3/__https:/www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html*RFC9350__;Iw!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmuk-veyTXw$>].
> The link with maximum generic metric value is not available for the use of
> Flexible Algorithms but is availble for other use cases.
>
>
>
> I believe the FlexAlgo reference here is to
> https://www.rfc-editor.org/rfc/rfc9350.html#name-max-metric-consideration
> <https://urldefense.com/v3/__https:/www.rfc-editor.org/rfc/rfc9350.html*name-max-metric-consideration__;Iw!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmumv0_0Zeg$>
>
>
>
>
> But the above text does not align with the RFC9350. If a link is to be
> made unavailable for FlexAlgos operating with a specific Generic Metric,
> then the way to do that is to omit that specific Generic Metric TLV from
> the ASLA for flex-algo application. The same would apply for other
> applications - just omit the metric. Why do we need a special
> MAX-LINK-METRIC value for generic metric given that it is a new thing we
> are introducing?
>
>
>
> <SH> I see what you are saying. Text is updated as below for ISIS and
> similar for OSPF.
>
> “A metric value of
>
>    0xFFFFFF is considered as maximum link metric and a link having
>
>    this metric value MUST be used during Flex-algorithm calculations
>
>    as a last resort link as described in sec 15.3 of RFC[9350]
>

KT> Thanks - that works. Perhaps also clarify that to make the link
unusable by FlexAlgo using that generic metric, the advertisement of the
particular generic metric can be skipped.


>
>
> 2) This comment is specific to OSPF given the encoding differences it has
> with ISIS. Section 2.2 allows for Generic Metric TLV to be used in too many
> places without clear specification of what it is used for - this is a
> potential pitfall for interop issues. RFC9492 provides helpful directions
> for us here.
>
> a) This draft specifies FlexAlgo extensions, it is natural that Generic
> Metric be advertised under ASLA TLV. No issues here.
>
> b) This draft does not specify anything about use of generic metric in
> base OSPF and as a reminder there is nothing like L-bit in OSPF encoding.
> Therefore, it does not make sense to advertise Generic Metric outside of
> ASLA and under the OSPFv2 Extended Link TLV or OSPFv3 Router Link TLV.
>
> c) This draft does not specify anything about use of generic metric with
> RSVP-TE/GMPLS. Therefore, it does not make sense to advertise Generic
> Metric in the TE Opaque LSAs.
>
> We can have specific documents that introduce (b) or (c) when there is a
> proper specification.
>
> <SH> Generic metric is a link attribute and can be used by other
> applications apart from Flex-algo.
>
> I don’t see a reason to not take code-points under other applicable LSAs.
>

KT> I disagree on this one. There is a reason why what is proposed in the
draft for ISIS is OK - due to the L-bit in ASLA, we need codepoints both
under ASLA and at the top level for FlexAlgo. There is no L-bit in OSPF and
therefore this does not apply. The code-points can be allocated when the
behavior is specified for those other LSAs and applications (beyond
FlexAlgo) in OSPF. We should not set the precedent for allocating
code-points for TLVs without any defined use or behavior.


>
>
> 3) Please introduce a reserved field to pad the OSPF FAEMD sub-TLV to a 4
> octet boundary as is the convention in OSPF -
> https://www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html#section-3.2.2
> <https://urldefense.com/v3/__https:/www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html*section-3.2.2__;Iw!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmunYcymQgw$>
>
> <SH> OK
>

KT> Thanks.


>
>
> 4) In section 3.2.2 there is the following text for OSPF. Could you please
> use the TLV/sub-TLV name? OSPF folks are not good at remembering numbers ;-)
>
>
>
> The Min Unidirectional Link Delay as advertised by *sub-sub-TLV 12* of
> ASLA sub-TLV [RFC8920
> <https://urldefense.com/v3/__https:/www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html*RFC8920__;Iw!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmun5uDKc7A$>],
> MUST be compared against the Maximum delay advertised in the FAEMD sub-TLV.
>
> <SH> changed to “The Unidirectional Link Delay as advertised by *sub-sub-TLV
> 12”*
>

KT> Perhaps my comment was not clear. The following text would be more
accurate:

The Min Delay value advertised via the Min/Max Unidirectional Link Delay
sub-TLV [RFC7471] of the ASLA sub-TLV [RFC8920
<https://www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html#RFC8920>
], MUST be compared against the Maximum delay advertised in the FAEMD
sub-TLV.


>
>
> 5) Do we want to call out that the reference bandwidth approach requires a
> router to compute and maintain a per link per algo bandwidth metric for
> every link in that algo topology. It may not be very obvious to some.
>
> <SH> updated as below
>
> “Advertising
>
>    the reference bandwidth in the FAD constraints allows the metric
>
>    computation to be done on every node for each link.
>
>    The metric is computed using reference bandwidth and the advertised
> link bandwidth.
>
>    Centralized control of this
>
>    reference bandwidth simplifies management in the case that the
>
>    reference bandwidth changes”
>

KT> The above text is not really needed IMHO. My point was more about the
implementation detail - for the flexalgo computation, we need to maintain
this info on a per link per algo topology basis in the link-state data
store used for path computation. I will leave it to the authors if this is
needed or is obviously clear to implementers.


>
>
> 6) There are a lot of procedures which are common to both OSPF and ISIS
> and are repeated in each section instead of a common section. It would be
> easier (and avoid errors) if there was some consolidation.
> https://www.rfc-editor.org/rfc/rfc9350.html#section-5
> <https://urldefense.com/v3/__https:/www.rfc-editor.org/rfc/rfc9350.html*section-5__;Iw!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmumpoQRYAA$>
> provides a good reference for such an organization of text.
>
> <SH> There is repetition in some cases but its not much so it seems to me
> leaving it as is for clarity may be better.
>

KT> This is an editorial comment so I leave it to the authors. My concern
is that great care/diligence is required through the rest of the
publication process to ensure that when anything is changed or updated for
one IGP, it is done appropriately for the other as well - it may be
copy/paste case when the change is IGP agnostic but may need careful
consideration when related to the specific IGP mechanics.


>
>
> 7) Regarding
> https://www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html#section-6
> <https://urldefense.com/v3/__https:/www.ietf.org/archive/id/draft-ietf-lsr-flex-algo-bw-con-08.html*section-6__;Iw!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmumnzxX7UA$>,
> it seems like we want to retain a numbering ordering of rules/sequence for
> flex-algo as extension documents are introduced. Am I correct? If so, then
> this document should formally "update" RFC9350 since it is changing the
> "set of rules/sequence" for FlexAlgo computations. Further such extensions
> will also need to keep updating the base spec similarly. I would suggest
> that a full set of rules that is a union of what is in RFC9350 and rules
> added by this draft are maintained in an Appendix section. Other documents
> in the future can similarly maintain the latest set of rules.
>
> <SH> This draft is adding 2 new rules at the end of existing rules. Its
> not modifying or changing the order.
>
> I don’t see what value it is going to add by repeating the set of rules in
> Appendix.
>

KT> What happens when another FlexAlgo document adds more rules? What
happens when some FlexAlgo document wants to insert rules in the middle of
existing ones instead of appending at the end? My point is that if there is
a desire to establish a "live" set of rules in specific orders, then we
need to leave a trail of document "updates" on the base FlexAlgo that one
can refer to know how these "live set of rules" are getting "updated" by
this and future documents. I am thinking of this on lines similar to an
update for an FSM.

Thanks,
Ketan



>
>
> 8) Please fix idnits warnings - some are related to obsolete references
> while others are related to formatting. There are also some
> spelling/grammar errors.
>
> <SH> ok
>
>
>
> Thanks,
>
> Ketan
>
>
>
>
>
> On Tue, Feb 20, 2024 at 3:56 AM Acee Lindem <[e-mail]> wrote:
>
>
> This starts the Working Group Last call for
> draft-ietf-lsr-flex-algo-bw-con-07. At least some of the flex algorithm
> enhancements described in the document have been implemented.
>
>  Please send your support or objection to this before March 5th, 2024.
>
> Thanks,
> Acee
> _______________________________________________
> Lsr mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/lsr
> <https://urldefense.com/v3/__https:/www.ietf.org/mailman/listinfo/lsr__;!!NEt6yMaO-gk!DwOojW2YZ48IROz9qMyyh7uKj3rYC-09avEhFQtcPkvxJ5mKF5Cyy6qSVvDJ89s9DmAUVwsT_pgfmukG-EHJRw$>
>
>