Re: [bess] New short WGLC for draft-ietf-bess-evpn-unequal-lb

[Neeraj Malhotra <[e-mail]>]

Hi Sergey, Bruno,

Many thanks for the detailed review and comments. Discussed with co-authors
and have enhanced the 'value' encoding to explicitly encode weight units as
default or generalized as suggested below. I have published a revision that
incorporates all comments discussed on this thread, as well as some
additional editorial comments from Jorge. Please let me know if there are
any additional comments (in particular on section 4).

Thanks,
Neeraj

On Fri, May 7, 2021 at 12:33 PM Sergey Fomin <[e-mail]> wrote:

> Hi Neeraj,
> thank you for the feedback.
>
> I agree that allocating a new code point for every possible calculation
> method would be cumbersome.
> How about a middle ground solution, with allocation of just a few values
> (even 2 would suffice: 1 for BW in whatever units we decide; the other
> "catch-all" generalized weight)?
> This way operators will get at least basic visibility & cfg. error
> prevention mechanism in place (I have a preference towards explicit
> signaling from ops point of view; but would be glad to hear operators
> opinion on this one).
>
> My concern with the current text is, while striving for maximum
> flexibility, we may end up in a situation where 2 implementations won't be
> able to interoperate with each other (e.g. vendor X decides to implement
> link BW as a metric, vendor Y decides to implement link count as a metric -
> since support for any method of weight calculation is completely
> optional). I believe we, as a technical community, should give guidance on
> implementing sensible defaults (i.e. "an implementation SHOULD support BW
> weight calculation method") where it makes sense.
>
> --
> Sergey
>
>
> 07.05.2021, 09:18, "Neeraj Malhotra" <[e-mail]>:
>
>
> Hi Sergey,
>
> Thanks for the comments. We would like to avoid adding another code point
> within the value field, as every new encoding of the "weight" will require
> a new code point. Agree that we should specify the default units (for which
> the consensus on the thread appears to be Mbps), but would rather leave it
> to implementations to support any additional units such as number of links,
> configured weight, or anything else, than have a standardized code point
> for each unit. Please let me know if this sounds fine. I will fix the units
> in section 5.2, and the text in section 4.1 as suggested by you and Bruno.
>
> Thanks,
> Neeraj
>
> On Thu, May 6, 2021 at 9:19 PM <[e-mail]> wrote:
>
> Hi Bruno, Jorge, et al.,
>
>
>
> Jorge,
>
> *> Among the co-authors we also discussed the possibility of defining two
> ECs: one for BW and one for a generalized-weight, so that the remote PE can
> catch if the multi-homed PEs were indeed using the same meaning of the
> weight. However, we thought it was easier/simpler to use a generalized
> value in a single EC sub-type, and add the sentence below. *
>
> Have you considered reserving a byte in the value field to signal a weight
> type? Similar to how it is done with DF election framework (and allocate a
> few right away, e.g. 0 for physical link BW, 1 for link count, 2 for
> manually configured weight, ...). This would help not only with interop
> cases, but to prevent & diagnose potential configuration mistakes as well.
>
>
>
> For interop reasons, I would support Bruno & argue that it makes sense to
> use "SHOULD" normative language at least for one way of calculation (e.g.
> BW).
>
>
>
> A couple of minor comments on newly-introduced text in -09 and -10:
>
>    1. Revision -10 introduced a curious change in section 5.2 “Remote PE
>    Behavior”, replacing “1000 Mbps” with “1000 megabytes” in an example. It is
>    not correct in the current form, since right above it there is an explicit
>    statement that links are Gigabit Ethernet: *"As an example, for a CE
>    dual-homed to PE-1, PE-2, PE-3 via 2, 1, and 1 GE physical links
>    respectively".*
>
> (keeping the Mbps units there would be appropriate, since this is a weight
> calculation, not a community encoding example)
>
>    1. Section 4.1 “Usage of EVPN Link Bandwidth Extended Community”:
>
> *“An implementation may support one or more of the above ways of encoding
> the value."* -> I would assume this should be "MUST support at least
> one", since every other possible option is included in bullet#2?
>
>
>
> --
>
> Sergey
>
> *From:* BESS <[e-mail]> *On Behalf Of *
> [e-mail]
> *Sent:* Thursday, May 6, 2021 5:46 AM
> *To:* Rabadan, Jorge (Nokia - US/Mountain View) <[e-mail]>;
> Neeraj Malhotra <[e-mail]>
> *Cc:* [e-mail]; [e-mail]
> *Subject:* Re: [bess] New short WGLC for draft-ietf-bess-evpn-unequal-lb
>
>
>
> Hi Jorge,
>
>
>
> Thanks for the feedback.
>
>
>
> Regarding the first point, I can live with the current text. But I think I
> would prefer that the text favour one option, and leave it to the
> responsibility of the SP for others usages. E.g.
>
>
>
> OLD:
>
> EVPN Link Bandwidth Extended Community value field is to be treated
>
>    as a 6 octet unsigned integer that may be set to:
>
>
>
>    o  total bandwidth of PE's all physical links in an ethernet segment,
>
>       expressed in bytes/sec.
>
>
>
>    o  or a generalized weight that may be set to link count, locally
>
>       configured weight, or a value computed based on an attribute other
>
>       than link bandwidth.
>
>
>
>    An implementation may support one or more of the above ways of
>
>    encoding the value.  Operator MUST ensure consistent encoding of this
>
>    value across all PEs in an ethernet segment.  Procedures related to
>
>    signaling and handling of this extended community defined in this
>
>    document use "total bandwidth in bytes/sec" encoding as an example to
>
>    illustrate its usage.
>
>
>
> NEW:
>
>    EVPN Link Bandwidth Extended Community value field is to be treated
>
>    as a 6 octet unsigned integer representing total bandwidth of PE's all
> physical links in an ethernet segment,
>
>       expressed in bytes/sec.
>
>
>
> Note however that the load balancing algorithm defines in this document
> uses ratio of Link Bandwidths hence the operator may choose a different
> unit or use the community as
>
>     a generalized weight that may be set to link count, locally
>
>       configured weight, or a value computed based on an attribute other
>
>       than link bandwidth. In such case, the operator MUST ensure
> consistent usage of the unit
>
> across all PEs in an ethernet segment. This may involve multiple routing
> domains/Autonomous Systems.
>
>
>
>
>
> But I leave this to you.
>
>
>
> Thanks,
>
> --Bruno
>
>
>
> *From:* Rabadan, Jorge (Nokia - US/Mountain View) [
> mailto:[e-mail] <[e-mail]>]
> *Sent:* Thursday, May 6, 2021 10:36 AM
> *To:* DECRAENE Bruno TGI/OLN <[e-mail]>; Neeraj Malhotra
> <[e-mail]>
> *Cc:* [e-mail]; [e-mail]
> *Subject:* Re: [bess] New short WGLC for draft-ietf-bess-evpn-unequal-lb
>
>
>
> Hi Bruno,
>
>
>
> Thanks for your comments.
>
>
>
> About the first point, we do have use cases where the bandwidth is not
> what we want to encode in the EC but rather a generalized weight that is
> derived from the link count, logical weight or simply a configured value.
> Among the co-authors we also discussed the possibility of defining two ECs:
> one for BW and one for a generalized-weight, so that the remote PE can
> catch if the multi-homed PEs were indeed using the same meaning of the
> weight. However, we thought it was easier/simpler to use a generalized
> value in a single EC sub-type, and add the sentence below.
>
>
>
> The sentence can be modified/fixed. But the gist is that the multi-homed
> PEs may support multiple meanings for the weight (BW, link-count, etc), but
> at least one of those MUST be common across all PEs and the multi-homed
> routes must use it consistently. Would it be enough if we fix it?
>
>
>
> About existing implementations, a new EVPN sub-type was defined only a
> couple of revisions ago, where, before, the existing non-transitive link BW
> EC was used, so there’s been some churn in the use of the EC anyway. I
> think it is important to get it as soon as possible, but get it right
> rather than finding gaps later once the document is done. But let us know
> your thoughts too.
>
>
>
> Thank you.
>
> Jorge
>
>
>
>
>
> *From: *BESS <[e-mail]> on behalf of
> [e-mail] <[e-mail]>
> *Date: *Thursday, May 6, 2021 at 10:04 AM
> *To: *Neeraj Malhotra <[e-mail]>
> *Cc: *[e-mail] <[e-mail]>, [e-mail] <
> [e-mail]>
> *Subject: *Re: [bess] New short WGLC for draft-ietf-bess-evpn-unequal-lb
>
> Hi Neeraj,
>
>
>
> Thanks for considering my comments.
>
> Much better from my perspective. Thank you.
>
>
>
> I have two comments on the changes:
>
> - Regarding deployments
>
> §4.1 allows two rather incompatible encodings/usages with no way to detect
> which one is used: some PE could advertise the bandwidth in bytes, while
> some other PE could advertise a general weight. I understand that both
> works, but to me there is a significant risk of issues over time or between
> domain/SP. I’d prefer that you only chose one in order to favour
> consistency in deployments and usage and I would prefer the real bandwidth
> (at least for consistency with the name of the community, but also because
> this is not subjective)  (And if a SP really wants to put an arbitrary
> value, I think he will figure out by himself, that it can do so).
>
> If you disagree with the above, then I would have a comment on the two
> below sentences:
>
> An implementation may support one or more of the above ways of
>
>    encoding the value.  Operator MUST ensure consistent encoding of this
>
>    value across all PEs in an ethernet segment.
>
> Logic dictates that the second sentence (MUST) can only be fulfilled if
> the first sentence mandates that all implementations MUST support both
> options, or one specifically defined.
>
>
>
> - Regarding existing implementations:
>
> previous version of the draft did not really specify the format of the
> EVPN EC. I had personally assumed that the format was similar to the
> draft-ietf-idr-link-bandwidth link bandwidth community hence encoded in
> IEEE floating point format. Latest version of the draft defines it in
> unsigned integer. Integer looks good to me, but for an existing
> implementation this may be seen as an incompatible change very late in the
> process. Obviously, if there are no implementation, there is no issue. In
> which case, you could also express the bandwidth in unit of bit/s _*if
> you*_ wish to. (I have no preference). However given that the draft had
> indicated a codepoint, there seem to be a risk of existing implementations
> hence incompatible change. BTW the codepoint is squatted even though the
> registry is FCFS hence easy to request.
>
>
>
> Thanks,
>
> --Bruno
>
>
>
>
>
> *From:* Neeraj Malhotra [mailto:[e-mail]
> <[e-mail]>]
> *Sent:* Thursday, May 6, 2021 7:41 AM
> *To:* DECRAENE Bruno TGI/OLN <[e-mail]>
> *Cc:* [e-mail]; [e-mail]
> *Subject:* Re: [bess] New short WGLC for draft-ietf-bess-evpn-unequal-lb
>
>
>
>
>
> Hi Bruno,
>
>
>
> Many thanks for the review comments. We have revised the draft addressing
> your comments.
>
>
>
> More inline.
>
>
>
> Thanks,
>
> Neeraj
>
>
>
> On Mon, May 3, 2021 at 2:20 AM <[e-mail]> wrote:
>
> Hi Stéphane, authors,
>
>
>
> I have not followed the discussions on this document, but I’ll nonetheless
> raise one point  regarding the bandwidth community (better safe than sorry).
>
> - why has [BGP-LINK-BW] been moved to informational reference while its
> reading seem mandatory?
>
>
>
> [NM]: There was a leftover reference to this in one of the sections that
> has been fixed now to use new EVPN EC. With this, reference to
> [BGP-LINK-BW] is purely informational (as was intended).
>
>
>
> - A new EVPN Link Bandwidth extended community is defined, but I could not
> find its specification. I guess that this is the same format as
> [BGP-LINK-BW] but transitive. Could this be explicitly stated?
>
>
>
> [NM]: clarified in section 4.
>
>
>
> - [BGP-LINK-BW] advertises the bandwidth in unit of bytes (not bits!) per
> second. Could the unit of the new EVPN Link Bandwidth extended community be
> also clearly spelled out? Especially give the history on this (cf below).
> Also in order to avoid misleading the readers could the examples use the
> correct unit (vs bits per seconds as writen)
>
>
>
> [NM]: done.
>
>
>
> - 10 years ago or so, I had raised a similar point (distinction between
> bits and bytes) on [BGP-LINK-BW] in the IDR WG. And it turned out that 1
> major implementation had implemented and deployed “bytes per second” as per
> the spec, while another implementation had implemented and deployed “bits
> per second” which is the typical unit of link bandwidth. Given the
> deployments, none was willing to change its implementation as it would be a
> non-backward compatible change with themselves. What’s the status on this?
> Could we have an implementation status on this?
>
>
>
> [NM]: I don't have this information. Perhaps someone else could comment.
>
>
>
>
>
> Thanks
>
> Regards,
>
> --Bruno
>
>
>
>
>
> *From**:* BESS [mailto:[e-mail]] *On Behalf Of *
> [e-mail]
> *Sent:* Monday, May 3, 2021 9:21 AM
> *To:* [e-mail]
> *Subject:* [bess] New short WGLC for draft-ietf-bess-evpn-unequal-lb
>
>
>
> Hi WG,
>
>
>
>
>
> We got final updates from authors on draft-ietf-bess-evpn-unequal-lb.
>
>
>
> I'm opening a new short Working Group Last Call (to be closed on 5/10) to
>
> get any last comments before moving to the next step.
>
> However, the document having normative references to EVPN PREF DF, and PER-MCAST-FLOW-DF, the draft will not be sent to IESG until these drafts are ready.
>
>
>
>  Feel free to send comments to the list before next Monday.
>
>
>
>
>
> Thanks,
>
>
>
>
>
> Stephane
>
>
>
> https://datatracker.ietf.org/doc/draft-ietf-bess-evpn-unequal-lb/
>
>
>
>
>
>
>
>
>
> _________________________________________________________________________________________________________________________
>
>
>
> Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
>
> pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
>
> a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
>
> Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
>
>
>
> This message and its attachments may contain confidential or privileged information that may be protected by law;
>
> they should not be distributed, used or copied without authorisation.
>
> If you have received this email in error, please notify the sender and delete this message and its attachments.
>
> As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
>
> Thank you.
>
> _______________________________________________
> BESS mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/bess
>
> _________________________________________________________________________________________________________________________
>
>
>
> Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
>
> pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
>
> a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
>
> Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
>
>
>
> This message and its attachments may contain confidential or privileged information that may be protected by law;
>
> they should not be distributed, used or copied without authorisation.
>
> If you have received this email in error, please notify the sender and delete this message and its attachments.
>
> As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
>
> Thank you.
>
> _________________________________________________________________________________________________________________________
>
>
>
> Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
>
> pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
>
> a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
>
> Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
>
>
>
> This message and its attachments may contain confidential or privileged information that may be protected by law;
>
> they should not be distributed, used or copied without authorisation.
>
> If you have received this email in error, please notify the sender and delete this message and its attachments.
>
> As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
>
> Thank you.
>
> ,
>
> _______________________________________________
> BESS mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/bess
>
>