I am not asking how to deal with the fact that the mode cannot manage
different flavors using just information that is in the packet. I am
pointing that out and you are trying to avoid to acknowledge that the
problem exists. That's your choice.

Regards,
Greg

On Tue, Oct 5, 2021, 15:07 Robert Raszuk <[email protected]> wrote:

> To me the easiest option here is to simply configure on each node selected
> compression schema for the domain. Do you see anything wrong with it ?
>
> On Wed, Oct 6, 2021 at 12:05 AM Greg Mirsky <[email protected]> wrote:
>
>> Hi Robert,
>> sorry,  but it doesn't seem to address my concern. My question is not
>> about mixing compression flavors in the same SRH (that is an interesting
>> case of its own). I am asking how a node that supports all the flavors
>> defined in the draft would parse an SRv6 packet with compressed SIDs in the
>> SRH. The impression I've got so far, is that is not possible for a node to
>> process a SID correctly without preconditions of "planning". In other
>> words, a controller constructs the list on the assumption that each node
>> supports one and only one flavor of CSID compression. Thus, I can conclude,
>> that the defined flavors are mutually exclusive and thus are different data
>> plane techniques of the SRv6 SID compression.
>>
>> Regards,
>> Greg
>>
>> On Tue, Oct 5, 2021, 14:41 Robert Raszuk <[email protected]> wrote:
>>
>>> Greg,
>>>
>>> SRH should have an equal size SIDs. That notion applies to compress
>>> SIDs. Mixing multiple flavors in a single domain node to node seems of no
>>> use to me. Within your domain you are subject to the domain
>>> architecture which is the key factor what compression scheme is chosen.
>>>
>>> Across domains (say you own N domains) the compressed SID size may vary.
>>>
>>> Does this answer your and maybe Ron's question ?
>>>
>>> Thx,
>>> R.
>>>
>>>
>>> On Tue, Oct 5, 2021 at 11:36 PM Greg Mirsky <[email protected]>
>>> wrote:
>>>
>>>> Hi Robert,
>>>> as I understand it, you believe everything that is written in the
>>>> draft. I hope you can help me find an answer to one simple question:
>>>>
>>>> Can a node that supports this draft in its entirety, i.e., supports all
>>>> "flavors" defined in the document, process received SRv6 packet with the
>>>> SRH encoded according to the specification?
>>>>
>>>> So far, the proponents of the draft referred to "planning" how flavors
>>>> of SRv6 SID compressed. To the best of my understanding, that is is a clear
>>>> demonstration of the incompatibility between flavors defined in the CSID
>>>> draft. Regardless of what is written in it.
>>>>
>>>> Regards,
>>>> Greg
>>>>
>>>> On Tue, Oct 5, 2021 at 1:24 PM Robert Raszuk <[email protected]> wrote:
>>>>
>>>>> Ron & SPRING WG chairs,
>>>>>
>>>>> Through this discussion we first have seen a debate if we need one or
>>>>> more data planes to compress SIDs in SRv6. WG clearly stated we need one.
>>>>>
>>>>> Following that we have observed a first terminology shift to see if
>>>>> asking how many solutions should be supported will work any better. To 
>>>>> that
>>>>> many WG members clearly stated that they support one solution.
>>>>>
>>>>> Well please notice that the draft in question in its introduction
>>>>> states:
>>>>>
>>>>> Abstract
>>>>>
>>>>>    This document defines a compressed SRv6 Segment List Encoding in the
>>>>>    Segment Routing Header (SRH).  *This solution* does not require
>>>>> any SRH
>>>>>    data plane change nor any SRv6 control plane change.  *This
>>>>> solution*
>>>>>    leverages the SRv6 Network Programming model.
>>>>>
>>>>> So based on my understanding of English the entire draft talks about a
>>>>> single solution.
>>>>>
>>>>> Then suddenly a new question popped up: how many behaviours are
>>>>> acceptable.
>>>>>
>>>>> I bet number of folks including myself said "one" keeping in mind
>>>>> previous discussions and the definition of "one" meaning based on the SRv6
>>>>> data plane in compliance to [RFC8402], [RFC8754] and [RFC8986].
>>>>>
>>>>> Interestingly enough the draft in question defines not behaviours but
>>>>> flavors as new variants of the already defined behaviors in Standards 
>>>>> Track
>>>>> RFCs. Namely it defines:
>>>>>
>>>>> 4.1.  NEXT-C-SID Flavor
>>>>> 4.2.  REPLACE-C-SID Flavor
>>>>>
>>>>> The newly defined behaviour End.XPS is optional.
>>>>>
>>>>> So if there is anything to ask here is to check if WG is ok with two
>>>>> flavors or not. I do not recall that question has ever been asked formally
>>>>> during the WG adoption call.
>>>>>
>>>>> With that let's note that optimal compressed SID size may be different
>>>>> network to network. One size does not fit all. Draft says:
>>>>>
>>>>> 6.1.  C-SID Length
>>>>>
>>>>>    The NEXT-C-SID flavor supports both 16- and 32-bit C-SID lengths.  A
>>>>> *   C-SID length of 16-bit is recommended.*
>>>>>
>>>>>    The REPLACE-C-SID flavor supports both 16- and 32-bit C-SID lengths.
>>>>> *   A C-SID length of 32-bit is recommended.*
>>>>>
>>>>> While I personally think 8-bit should be an option, if we choose a
>>>>> single flavor we will introduce suboptimality for no good reason. Hardware
>>>>> capable of supporting any flavor clearly can do LPM on locator. Also
>>>>> hardware capable of supporting one flavor can support few other flavors as
>>>>> this is pretty much just an offset game.
>>>>>
>>>>> Kind regards,
>>>>> Robert
>>>>>
>>>>>
>>>>>
>>>>> On Tue, Oct 5, 2021 at 9:43 PM Ron Bonica <rbonica=
>>>>> [email protected]> wrote:
>>>>>
>>>>>> Pablo,
>>>>>>
>>>>>>
>>>>>>
>>>>>> Ae you sure? Please look at the question as Joel asked it (
>>>>>> https://mailarchive.ietf.org/arch/msg/spring/nS2gnQ_jxvpbmcxs_d3JAbUCT1I/
>>>>>> ).
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> Ron
>>>>>>
>>>>> _______________________________________________
>>>>> spring mailing list
>>>>> [email protected]
>>>>> https://www.ietf.org/mailman/listinfo/spring
>>>>>
>>>>
_______________________________________________
spring mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/spring

Reply via email to