Bruno - I have read your post carefully - but it is still possible that I have misunderstood some of what you have written. If so, please help correct my misunderstandings.
Please see my responses inline. > -----Original Message----- > From: bruno.decra...@orange.com <bruno.decra...@orange.com> > Sent: Tuesday, March 28, 2023 6:45 PM > To: Les Ginsberg (ginsberg) <ginsb...@cisco.com>; Christian Hopps > <cho...@chopps.org> > Cc: Huaimo Chen <huaimo.c...@futurewei.com>; draft-chen-lsr-isis-big- > tlv.auth...@ietf.org; lsr@ietf.org > Subject: RE: [Lsr] Comments on draft-chen-lsr-isis-big-tlv-00 > > Les, all > > > From: Les Ginsberg (ginsberg) <ginsb...@cisco.com> > > > > Chris - > > > > <snip> > > However, that is the missing piece, so it works if we also add a capability > > bit. > If we have the capability bit you now know which routers are processing the > container TLV and which aren't. That should be enough info to route > correctly. > > > > Using a container TLV *and* a capability bit is not a free lunch, but it > > should > work to allow incremental deployment safely. If that's something we want as > a WG. > > <end snip> > > > > No - this does not work. > > Customer deploys some features. They expect all routers in the network to > be able to correctly calculate topology and correctly forward for the features > they support. > > They do not deploy a feature and expect only a subset of the routers in the > network that are configured to support the feature to correctly calculate > paths. > > > > There is no way to successfully support incremental deployment. > > [Bruno] > 1) globally consistent TLV or not > Probably the discussion we could make a distinction between: > - a- TLVs which are required to be consistent (typically the one involved in > distributed SPF e.g. IP and IS reachability) > - b- TLVs which are not (e.g. SRv6 locators for algo 0 & 1, but there are > probably others easier e.g. local adjacency SIDs for SR-MPLS) > > For "a" there is no way to do incremental deployment. So it's either relying > on careful deployment (and implementation) or relying on a capability. Both > solution seems equal on this. > For "b" the big TLV will allow to deploy larger TLV without impacting legacy > implementation. That's a benefit. > [LES:] What you are proposing is that we categorize TLVs into two different categories - and use different encoding strategies when more than 255 bytes of information is required to be sent. I think this is undesirable. It further complicates implementations - both in how to format advertisements when sending information and in how to process received information. The justification you provide for doing this is the suggestion that "Big TLV" supports incremental deployment - but this is not the case. To understand why I say this it would be good to review the example use case I provided below in my response to Chris. [LES:] The terms "incremental deployment" and "backwards compatibility" are sometimes abused. When introducing a new advertisement (e.g., a new sub-TLV for a Neighbor advertisement) it may indeed be possible to make such claims. The new sub-TLV may be used only for some new feature. Nodes which do not support the new feature have no need to parse the new sub-TLV and it is therefore safe to introduce such advertisements even when only some nodes in the network support the associated feature. [LES:] However, that is not what we are dealing with in the case of how to advertise more than 255 bytes. Existing defined content (e.g., existing link attribute sub-TLVs for Neighbor advertisements) may be sufficient to require more than 255 bytes to be advertised for an object. This means that all nodes in the network need to be able to parse all the advertisements. Placing some of the advertisements inside a new container TLV does not mean that those advertisements need not be parsed by all nodes - yet that is exactly what will happen when these advertisements are placed inside a new TLV that some nodes do not understand. [LES:]The claim that BIG TLV supports incremental deployment is simply not correct. > 2) failing versus failing > Even when we need global consistency , there may be multiple failure modes > to consider. > - One is not achieving global consistency, which will trigger adverse > consequences. E.g. persistent forwarding loops for some path for data > involved in SPF computations. Both solutions have the same properties > - One is that the new advertisement is not correctly understood by legacy > implementation and confuse them. This may trigger additional issues. E.g. > larger set of inconsistencies as the first TLV instance/part may not become > inconsistent anymore. Also, unfortunately, it's not unheard that some > implementation are light on the error handling part (e.g. "undefined > behavior", IS-IS process crash). In that regards, big-tlv seems to have an > advantage in that legacy implementation are not impacted and everyone > have a consistent view of the first TLV part. > . [LES:] I agree that the consequences of advertising more than 255 bytes about an object when not all nodes support such advertisement can vary depending on the information being advertised, the placement of information in the TLVs used to advertise the information, and even the syntax of the advertisement. But it is not correct to say that big-tlv has any advantage (please see my comments above). [LES:] Given that, introducing multiple ways to support advertisement of more than 255 bytes only serves to complicate implementations and increase risk of interoperability issues. It goes against the principles that we have used when extending the protocol over the last 20 years. Multiple ways of doing same thing are simply bad for everyone - implementors as well as users. [LES:] I am reminded of the often quoted : “Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away.” > > > > >>>We are not naïve - we understood very well that if not all routers in > the > > > network supported at least reception of MP TLVs that there would be > > > deployment issues. > > [Bruno] > Did network operators had the opportunity to comment on these > deployment issues and discuss the involved trade-off? > I have never assumed that you were naïve and I don't feel that this was > implied in the original email. IMO the question is that different person may > have different perspective (which is just fine). So if the decision is taken > by a > single set of persons having a similar perspective, it's likely that the > outcome > optimization be only local. e.g. pushing back the issue/complexity on the > network operation side versus on the design/implementation side. > [LES:] Implementations that support advertising more than 255 bytes today do so using the syntax explicitly defined in existing RFCs. What is lacking in existing RFCs is explicit indication of the use of the defined syntax for some TLVs to which it can be logically applied. (This is discussed in more detail in the MP-TLV draft.) [LES:] (Speaking only for myself) If I did not consult operators regarding the application of the defined syntax to TLVs where it had not previously been used it is because I saw no reason to consider an alternative syntax. The protocol defined the syntax long ago. [LES:] You are suggesting that there actually is a choice where a different syntax provides significant advantages over MP-TLV. But IMO there is no such choice. Big-TLV does not provide the functionality you claim for it - and I have provided detailed responses as to why I believe that is the case. If I have not convinced you of this, then I suggest we focus on the claims of incremental deployment support and try to demonstrate them with real world examples. I have already provided one such example which demonstrates that incremental deployment is NOT supported. Les > --Bruno > > > > > I already gave an example in my comments below: > > > > > >> > [LES:] Consider the following simple example. > > > >> > > > > >> > Node A needs to send 10 sub-TLVs about a particular object – > > > >> > requiring more than 255 bytes to be sent. > > > >> > > > > >> > Some nodes in the network do not support reception of more than > 255 > > > >> > bytes/object. Consider two such nodes. > > > >> > > > > >> > Node B, based on the local configuration, needs to be able to > receive > > > >> > sub-TLVs 1,3,5,7,9 from A. > > > >> > > > > >> > Node C, based on local configuration, needs to be able to receive > > > >> > sub-TLVs 2,4,6,8,10 from A. > > > >> > > > > >> > > > > >> > > > > >> > There is no way that A can advertise all 10 sub-TLVs in a way which > > > >> > allows both B and C to correctly process the sub-TLVs they require. > > > >> > > > > > Les > > > > > -----Original Message----- > > > From: Christian Hopps <cho...@chopps.org> > > > Sent: Tuesday, March 28, 2023 9:52 AM > > > To: Les Ginsberg (ginsberg) <ginsb...@cisco.com> > > > Cc: Christian Hopps <cho...@chopps.org>; Huaimo Chen > > > <huaimo.c...@futurewei.com>; draft-chen-lsr-isis-big- > tlv.auth...@ietf.org; > > > lsr@ietf.org > > > Subject: Re: [Lsr] Comments on draft-chen-lsr-isis-big-tlv-00 > > > > > > > > > "Les Ginsberg (ginsberg)" <ginsb...@cisco.com> writes: > > > > > > > Chris - > > > > > > > > Please see inline - I'll try to conform to your request about ">>>" > quoting - > > > > but given that this style does not identify who made the comment, I > have > > > found > > > > in the past that this style becomes very hard to follow after a couple > > > > of > > > > replies. > > > > Though perhaps that could be said of any style. 😊 > > > > > > Well in the ">>>" style my text that you were quoting would have been > > > > > > "> like this" > > > > > > and yours would not have anything preceding it.. like mine is here. > > > > > > anyway, it's a losing battle against html I typically just load these > > > email > into > > > chrome when I need to read them.. > > > > > > >> -----Original Message----- > > > >> From: Christian Hopps <cho...@chopps.org> > > > >> Sent: Tuesday, March 28, 2023 7:27 AM > > > >> To: Les Ginsberg (ginsberg) <ginsb...@cisco.com> > > > >> Cc: Huaimo Chen <huaimo.c...@futurewei.com>; draft-chen-lsr-isis- > big- > > > >> tlv.auth...@ietf.org; lsr@ietf.org > > > >> Subject: Re: [Lsr] Comments on draft-chen-lsr-isis-big-tlv-00 > > > >> > > > >> > > > >> Hi, > > > >> > > > >> So I agree that using this new container TLV along with old TLVs > doesn't > > > help. > > > >> > > > > > > > >>>[LES:] Good - we agree. > > > > > > > >> However, it is worth nothing that if *only* the container TLV was used > > > (i.e., > > > >> once a TLV became too large it would be removed and placed inside > > > >> container TLVs) then it would actually represent a safer way to deploy > this > > > >> "multiple tlv" functionality. > > > >> > > > >> If the container only was used, then only routers that understood > would > > > be > > > >> able to use *any* of the TLV data. This would actually solve the > problem > > > of > > > >> "newly inserted legacy router brings everything back down" that using > a > > > >> required capability bit being set on all routers has. > > > >> > > > >>>[LES:] I don't agree - and here is why. Let's use the example of > Neighbor > > > TLVs. > > > >>>With what you propose, when a router starts using the container TLV, > > > those routers who don’t support/understand it would simply not be > aware > > > of the advertisement at all. > > > >>>This would result in inconsistent routing calculations on different > routers > > > leading to loops/blackholes. > > > >>>Hardly a benign impact. > > > > > > You're right, not sure why I thought new routers would know that old > routers > > > weren't acting on the container TLV. > > > > > > However, that is the missing piece, so it works if we also add a > > > capability > bit. > > > If we have the capability bit you now know which routers are processing > the > > > container TLV and which aren't. That should be enough info to route > > > correctly. > > > > > > >>>There is no free lunch here. No matter what encoding scheme you > come > > > up with, unless all routers in the network understand it, things are going > to > > > break. > > > > > > > > > > Using a container TLV *and* a capability bit is not a free lunch, but it > should > > > work to allow incremental deployment safely. If that's something we > want as > > > a WG. > > > > > > Thanks, > > > Chris. > > > [as wg-member] > > > > > > >> This later issue with the capability bit is why no-one wanted to use a > > > >> it, > > > and > > > >> why we currently have this very sub-optimal "solution" of "just do it > and > > > >> hope it works". > > > > > > > >>>[LES:] Folks (like me) who implemented MP for TLVs like > Neighbor/Prefix > > > were > > > >>> following established practice for the protocol i.e., there are > > > >>> multiple > > > >>> cases where this behavior is explicitly specified (please see MP draft > for > > > a > > > >>> list) > > > >>>So it made sense to use the same mechanism for other TLVs. > > > >>>We are not naïve - we understood very well that if not all routers in > the > > > network supported at least reception of MP TLVs that there would be > > > deployment issues. > > > >>>That is why I am working with enthusiasm on the MP draft. > > > > > > > > Les > > > > > > > >> > > > >> Thanks, > > > >> Chris. > > > >> [as wg-member] > > > >> > > > >> > > > >> P.S. the quoting style used in this thread is fabulously hard to > > > comprehend in > > > >> a text based email client.. What's wrong with good old ">>>" quoting > style > > > >> anyway? > > > >> > > > >> > > > >> "Les Ginsberg (ginsberg)" <ginsberg=40cisco....@dmarc.ietf.org> > > > writes: > > > >> > > > >> > Huaimo – > > > >> > > > > >> > > > > >> > > > > >> > Please see inline. > > > >> > > > > >> > > > > >> > > > > >> > From: Huaimo Chen <huaimo.c...@futurewei.com> > > > >> > Sent: Sunday, March 26, 2023 3:41 AM > > > >> > To: Les Ginsberg (ginsberg) <ginsb...@cisco.com>; lsr@ietf.org; > > > >> > draft-chen-lsr-isis-big-tlv.auth...@ietf.org > > > >> > Subject: Re: Comments on draft-chen-lsr-isis-big-tlv-00 > > > >> > > > > >> > > > > >> > > > > >> > Hi Les, > > > >> > > > > >> > > > > >> > > > > >> > Thanks much for your comments. > > > >> > > > > >> > My responses are inline below with HC. > > > >> > > > > >> > > > > >> > > > > >> > Best Regards, > > > >> > > > > >> > Huaimo > > > >> > > > > >> > > > > >> > > > > >> > From: Les Ginsberg (ginsberg) <ginsb...@cisco.com> > > > >> > Sent: Thursday, March 23, 2023 3:35 AM > > > >> > To: lsr@ietf.org <lsr@ietf.org>; > > > >> > draft-chen-lsr-isis-big-tlv.auth...@ietf.org < > > > >> > draft-chen-lsr-isis-big-tlv.auth...@ietf.org> > > > >> > Subject: Comments on draft-chen-lsr-isis-big-tlv-00 > > > >> > > > > >> > > > > >> > > > > >> > Folks - > > > >> > > > > >> > > > > >> > > > > >> > This draft proposes a new way to handle advertisement of more > than > > > >> > 255 bytes of information about a given object. > > > >> > > > > >> > It defines a new "container TLV" to carry additional information > > > >> > about an object beyond the (up to) 255 bytes of information > > > >> > advertised in an existing TLV. > > > >> > > > > >> > > > > >> > > > > >> > The draft is defining a solution to a problem which has already been > > > >> > addressed without requiring any protocol extensions. > > > >> > > > > >> > [HC]: It seems that a protocol includes a set of procedures. Would > > > >> > you mind telling me which existing protocols can be used to resolve > > > >> > the problem without requiring any protocol extensions? > > > >> > > > > >> > > > > >> > > > > >> > [LES:] Please read draft-pkaneria-lsr-multi-tlv-02 carefully. > > > >> > > > > >> > Section 1 documents that there are existing RFCs which explicitly > > > >> > state that multiple TLVs for the same object are allowed to be sent. > > > >> > > > > >> > What the draft goes on to discuss is the use of the same mechanism > > > >> > (sending multiple TLVs for the same object) in cases where existing > > > >> > RFCs have not explicitly stated this behavior. > > > >> > > > > >> > > > > >> > > > > >> > It is also a fact that there are multiple implementations from > > > >> > multiple vendors already shipping that utilize this mechanism for > > > >> > TLVs such as Neighbor and Prefix reachability. > > > >> > > > > >> > > > > >> > > > > >> > The existing solution - discussed in > https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatat > racker.ietf.org%2Fdoc&data=05%7C01%7Cbruno.decraene%40orange.com% > 7C0e548b0a0da64bff39f708db2fb1b252%7C90c7a20af34b40bfbc48b9253b6f5 > d20%7C0%7C0%7C638156212465727775%7CUnknown%7CTWFpbGZsb3d8eyJ > WIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D% > 7C3000%7C%7C%7C&sdata=Lrj1P%2BYnTCR0rwJRbNNpVWxbjpVX5Muq4v8A > W802WPc%3D&reserved=0 > > > >> > /draft-pkaneria-lsr-multi-tlv/ has already been successfully > > > >> > implemented and deployed by multiple vendors. > > > >> > > > > >> > [HC]: You are a co-author of this draft, called a first draft for > > > >> > resolving the problem on big TLVs. This first draft contains some > > > >> > protocol extensions. If there is a solution for the problem without > > > >> > requiring any protocol extensions, then why do you as a co-author > > > >> > work on the first draft with protocol extensions? > > > >> > > > > >> > > > > >> > > > > >> > [LES:] There are no protocol extensions defined in > > > >> > draft-pkaneria-lsr-multi-tlv-02 (please see the statement in the > IANA > > > >> > section). The draft has been written to clarify existing behavior and > > > >> > to discuss best deployment practices in cases where not all > > > >> > implementations support reception of multiple TLVs for a given > > > >> > object. > > > >> > > > > >> > > > > >> > > > > >> > The definition of a second solution to the problem is not needed - > > > >> > and in fact further complicates both implementation and > deployment. > > > >> > Should the existing solution be used? Should the new solution be > > > >> > used? What is the state of support by all nodes in the network for > > > >> > each solution? > > > >> > > > > >> > [HC]: It seems better to merge the two drafts (i.e., the first draft > > > >> > and the second draft defining container TLV) into one. > > > >> > > > > >> > > > > >> > > > > >> > [LES:] This would the worst possible outcome. > > > >> > > > > >> > It would define two mechanisms for sending more than 255 bytes of > > > >> > information about an object. > > > >> > > > > >> > This would require implementations to support two different > > > >> > mechanisms for advertising the same information – also requiring > the > > > >> > ability to control which mechanism should be used in a given > > > >> > deployment and even raising the possibility that both forms would > > > >> > need to be sent in parallel. This adds unnecessary complexity to > > > >> > implementations. > > > >> > > > > >> > > > > >> > > > > >> > For operators deploying features+scale which require such support, > > > >> > they would now have to identify not only whether all > implementations > > > >> > in their deployment support sending/receiving more than 255 bytes/ > > > >> > object, but also which form of advertisement is supported – further > > > >> > complicating deployment considerations. > > > >> > > > > >> > > > > >> > > > > >> > And since there are explicit statements requiring the current form of > > > >> > advertisement to be used for some TLVs, behavior would potentially > > > >> > differ on a per TLV basis. > > > >> > > > > >> > > > > >> > > > > >> > The motivation for the new solution seems to be the notion that it > > > >> > supports partial deployment. Text in > https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww > .ietf.org%2Farchive%2Fid%2F&data=05%7C01%7Cbruno.decraene%40orang > e.com%7C0e548b0a0da64bff39f708db2fb1b252%7C90c7a20af34b40bfbc48b9 > 253b6f5d20%7C0%7C0%7C638156212465727775%7CUnknown%7CTWFpbGZs > b3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn > 0%3D%7C3000%7C%7C%7C&sdata=hvQfndkpApbJFtWqufc0t%2F9RrEDEAqd > AvzuMBoo%2FTh4%3D&reserved=0 > > > >> > draft-chen-lsr-isis-big-tlv-00.html#name-incremental-deployment > > > >> > states: > > > >> > > > > >> > > > > >> > > > > >> > "For a network using IS-IS, users can deploy the extension for big > > > >> > TLV in a part of the network step by step. > > > >> > > > > >> > The network has some nodes supporting the extension (or say new > > > nodes > > > >> > for short) and the other nodes not > > > >> > > > > >> > supporting the extension (or say old nodes for short) before the > > > >> > extension is deployed in the entire network." > > > >> > > > > >> > > > > >> > > > > >> > This suggests the authors believe that a network can function with > > > >> > some nodes using all of the advertisements and some nodes using > only > > > >> > the legacy advertisements, but this is obviously false. > > > >> > > > > >> > Fundamental to operation of a link state protocol is that all nodes > > > >> > in the network operate on identical LSPDBs. > > > >> > > > > >> > The suggestion that features will work correctly when some nodes > use > > > >> > attributes advertised in legacy TLVs and the new container TLV while > > > >> > some nodes use only the attributes advertised in legacy TLVs is > > > >> > simply incorrect. > > > >> > > > > >> > [HC]: Every node in the network has the same LSPDB. The new > nodes > > > >> > understand the new container TLVs and may use them. The old > nodes > > > do > > > >> > not understand them and do not use them. > > > >> > > > > >> > > > > >> > > > > >> > [LES:] Consider the following simple example. > > > >> > > > > >> > Node A needs to send 10 sub-TLVs about a particular object – > > > >> > requiring more than 255 bytes to be sent. > > > >> > > > > >> > Some nodes in the network do not support reception of more than > 255 > > > >> > bytes/object. Consider two such nodes. > > > >> > > > > >> > Node B, based on the local configuration, needs to be able to > receive > > > >> > sub-TLVs 1,3,5,7,9 from A. > > > >> > > > > >> > Node C, based on local configuration, needs to be able to receive > > > >> > sub-TLVs 2,4,6,8,10 from A. > > > >> > > > > >> > > > > >> > > > > >> > There is no way that A can advertise all 10 sub-TLVs in a way which > > > >> > allows both B and C to correctly process the sub-TLVs they require. > > > >> > > > > >> > > > > >> > > > > >> > Network functionality is compromised. > > > >> > > > > >> > > > > >> > > > > >> > It is true that even with the existing solution unless all nodes are > > > >> > capable of processing more than 255 bytes of information/object > > > >> > network functionality will be compromised. That is exactly what > > > >> > motivated the writing of draft-pkaneria-lsr-multi-tlv. > > > >> > > > > >> > But your proposal does nothing to make that requirement easier to > > > >> > address. It in fact makes implementation/deployment even more > > > >> > difficult – as I have described above. > > > >> > > > > >> > > > > >> > > > > >> > It is also important to also state that the advertisement of more > > > >> > than 255 bytes of information is driven by configuration – not a > > > >> > protocol implementation choice. Suppressing advertisement of > some of > > > >> > the configured information also does not result in a working > network. > > > >> > > > > >> > > > > >> > > > > >> > In short, there is no positive value from the proposed extension – > > > >> > and it does harm by further complicating implementations and > > > >> > deployments. > > > >> > > > > >> > [HC]: The second draft defines a general mechanism for resolving > the > > > >> > problem. It is backward compatible and simple. It does not do any > > > >> > harm. > > > >> > > > > >> > > > > >> > > > > >> > [LES:] You are proposing a second solution for a problem that has > > > >> > already been solved. In doing so you are introducing new problems > and > > > >> > not solving any existing issues. Saying this “does no harm” is > > > >> > clearly false. > > > >> > > > > >> > > > > >> > > > > >> > Les > > > >> > > > > >> > > > > >> > > > > >> > > > > >> > > > > >> > The draft should be abandoned. > > > >> > > > > >> > > > > >> > > > > >> > Les > > > >> > > > > >> > > > > >> > > > > >> > > > > >> > > > > >> > _______________________________________________ > > > >> > Lsr mailing list > > > >> > Lsr@ietf.org > > > >> > > https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww > .ietf.org%2Fmailman%2Flistinfo%2Flsr&data=05%7C01%7Cbruno.decraene% > 40orange.com%7C0e548b0a0da64bff39f708db2fb1b252%7C90c7a20af34b40b > fbc48b9253b6f5d20%7C0%7C0%7C638156212465727775%7CUnknown%7CTW > FpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJX > VCI6Mn0%3D%7C3000%7C%7C%7C&sdata=3L81UcaHjh7LdY1cfPUqNslBunb8 > dqg9lv71mfrj7rg%3D&reserved=0 > > > > Orange Restricted > > __________________________________________________________ > __________________________________________________________ > _____ > > 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. _______________________________________________ Lsr mailing list Lsr@ietf.org https://www.ietf.org/mailman/listinfo/lsr
