Hi Ketan, The draft version from previous mail is submitted now. I'll update draft with those minor updates.
For last opened comment: KT> I am afraid this is not very clear. Are you saying that this document is not just introducing a new A flag but also a fixed 32 bit container that is always present in the SR-ERO object (when the algo capability is advertised)? I just saw that those fields are not optional in the diagram. Then I am confused why the size of the object would change depending on the value of the A field (see text at the end of section 4.2). <S> SR-ERO Subobject is extended with 4 bytes to keep length aligned to multiple of 4. Those 4 bytes are present if needed for any extension included in them based on flags (not globally to all instances based on capability advertised). For now, it will be included if A flag is set in specific SR-ERO subobject. In the future if some other draft will add another field in "Reserved" space and new flag, then same 4 bytes can be included as well (including Algorithm field). That's why we are saying in the draft even now that PCEP speaker will ignore value of Algorithm field if A flag was not set (even if capability was advertised). We are trying to make sure that if in the future somebody will add new extensions, which can (but does not have to combined with extensions from this draft), then we will encode subobject in optimal way, so something like: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |L| Type=36 | Length | NT | Flags |N|A|F|S|C|M| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | SID (optional) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // NAI (variable, optional) // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reserved | NEW FIELD | Algorithm | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Instead of doing sub-optimal encoding like this: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |L| Type=36 | Length | NT | Flags |N|A|F|S|C|M| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | SID (optional) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // NAI (variable, optional) // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reserved | Algorithm | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reserved | NEW FIELD | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Especially since ERO subobjects can be encoded a lot of times per LSP (multiple hops in single ERO, possibly repeated in RRO, multiple segment-lists, reverse ERO,…), so optimality of encoding is more important than for example encoding of constraints. (We can still start discussion in mailing list about future extensions, but encoding above is at least keeping doors opened even if we decide to use sub-TLV or any other solution.) Regards, Samuel -----Original Message----- From: Ketan Talaulikar <[email protected]<mailto:[email protected]>> Sent: Friday, June 20, 2025 5:54 PM To: Samuel Sidor (ssidor) <[email protected]<mailto:[email protected]>> Cc: [email protected]<mailto:[email protected]>; [email protected]<mailto:[email protected]> Subject: Re: AD review of draft-ietf-pce-sid-algo-19 Hi Samuel/co-authors, Thanks for taking the time to discuss and incorporate the suggestions/feedback. I would also suggest posting this update as an "intermediate update" so the WG also gets the time to review and (if necessary) follow-up. We are almost done anyway. Please check inline below for follow-ups/clarifications. There is really only one major item where I am still not clear (related to the Algo field in the SR-ERO). Also you can consider points where I have not responded as addressed (either by the proposed changes or responses). On Tue, Jun 17, 2025 at 8:48 AM Samuel Sidor (ssidor) <[email protected]<mailto:[email protected]>> wrote: > Hi Ketan, > > > > Please find inline responses <S> and version updated based on > discussion with other co-authors (thanks a lot to PSF and Andrew for great > discussion). > > > > Thanks, > > Samuel > > > > Hi Authors/WG, > > Thanks for your work on this important document that tries to leverage > IGP technologies like FlexAlgo in PCE and builds on top for delivering > new use-cases. > > I would like to start with a high-Level overview of the specifications > based on my reading and some questions/suggestions. Please correct me > if my understanding is wrong. > > 1) There are 3 main types of extensions: > > a) Extensions to Metric Objects: these aren't specific to SR path types, > they aren't specific to FlexAlgo, they are more general. I suggest that > this be brought out in the Abstract and Introduction sections. > Also, that > the specifications for this be brought out as a top-level section in the > document. Please consider this as a major editorial comment. > > <S> Both abstract and introduction updated. > > b) Algo support in SR-ERO/RRO: this is purely a signaling extension to > convey additional algo information and unrelated to any computational > enhancements to the PCE functionality. E.g., it could be used for say a > PCE-initiated explicit path. Would be great if this could be called out > and the document sections were re-ordered accordingly. > > <S> Ack, moved to separate section. > > c) Algorithm Constraint: this is getting algo (both FlexAlgo and any > other specific IGP algo) related aspects into the path computation. This > now enables the conveying of IGP aspects (e.g., for FlexAlgo the > optimization metric and other constraints from the FAD) in PCEP. > > Both (b) and (c) are covered by the same new capability being introduced > and specific to the SR path setup types. Ideally, it would have > been nice > to use different capabilities, but I guess (b) is really a small > add-on and > hence covered together with (c). I hope there is no one that tries to > implement (b) but not (c). > > <S> Correct, it is small enough to create separate capability, but we > are open to change it if there anybody really plans to implement subset only. > > 2) The (1)(c) functionality can be further split into two major types: > > a) FlexAlgo related (algo 128-255): this requires that the PCE is aware > of IGP signaling information related to FlexAlgo (such as FAD, algo > participation, metric types including generic metric, ASLA, etc.). > It would > be good to call out that it is expected that this topology information > is conveyed to the PCE via existing mechanisms (viz. IGPs or BGP-LS). > > <S> Added explicit statement to “5.2.2. Path Computation for Flexible > Algorithms” section. > KT> Thanks. Please update the reference to BGP-LS to RFC 9552. > > b) non-FlexAlgo related (algo 0-127): Of these algos, only default (0) > and strict-SPF (1) are specified. Since the rest is undefined, they > cannot > be supported in PCEP as well - this needs to be called out along > with the > necessary error handling, if those are received. Further, default > (0) has > been already in use in PCEP for ages now - this needs to be called out
_______________________________________________ Pce mailing list -- [email protected] To unsubscribe send an email to [email protected]
