Re: CRH is not needed - Re: How CRH support SFC/Segment Endpoint option?

[Tom Herbert <[e-mail]>]

On Tue, May 26, 2020 at 10:27 AM James Guichard
<[e-mail]> wrote:
>
> Hi Andrew,
>
>
>
> Some of your comments about SRv6 are a little subjective.
>
>
>
> You say SRv6 is complex and yet in its basic form it is simply a list of IPv6 addresses carried in a routing header which does not seem too difficult to understand especially in light of the fact that with CRH I have essentially the same thing albeit with another level of indirection to resolve the mappings.

Jim,

I agree with the assessment that SRv6, specifically SRH, is complex.
It is more than simply a list of IPv6 addresses-- the protocol format
includes Flags fields, Tags, Last Entry, and TLVs one of which is
HMAC. Those "features" do not make for a simple dataplane protocol
that is amendable hardware and working through those was one of the
reasons it took so long to get SRH through 6man. Had SRH be defined as
just a list of IPv6 addresses and nothing else, I believe these
conversations would be quite different.

Tom

> You say SRv6 has massive overhead and yet there are several proposals in which that overhead can be compressed without having to implement a mapping system to resolve the indirection.
> The overloading the IPv6 address space comment seems at best subjective given that the IPv6 address in the DA field for SRv6 on the wire looks no different than any other IPv6 packet; the fact that some router in the network locally defines a set of instructions to execute upon receipt of a packet using that IPv6 address is a local affair and of no particular concern to any other router. The same logic applies to things like policy-based routing where a router can implement some policy to direct a packet on a different path that no router upstream is aware of.
> Given the long and quite exhausting email thread on violation of RFC8200 it is apparent that not everyone agrees with you on this point and in fact I would go as far as to say the current text is pretty clear from an English language point of view that I find it interesting how many different interpretations have sprung up.
> Network programming is a tool (that you are free to use or not) to enable innovation through locally defined instructions; again, how is this over engineered and if it is over engineered how can it also be inflexible? Seem to me that the complete opposite is true.
>
>
>
> Thanks!
>
>
>
> Jim
>
>
>
> From: spring <[e-mail]> On Behalf Of Andrew Alston
> Sent: Tuesday, May 26, 2020 7:59 AM
> To: Mach Chen <[e-mail]>; Zafar Ali (zali) <zali=[e-mail]>; Ron Bonica <rbonica=[e-mail]>; Chengli (Cheng Li) <[e-mail]>; 6man <[e-mail]>; [e-mail]
> Subject: Re: [spring] CRH is not needed - Re: How CRH support SFC/Segment Endpoint option?
>
>
>
> Speaking as an operator that operates in 15 countries – yes – yes and yes again.
>
>
>
> SRv6 will never find a home on our network – it is complex, it has massive overhead, it overloads the address space in ways that make me cringe, it currently (in my view) violates RFC8200, the network programming draft is so overengineered that it is entirely inflexible – where as CRH – is simple, it gives me exactly what we need, it is a building block to many other things in the pipe line, and it has been tested and is functional.
>
>
>
> I will not begrude anyone who wants to run SRv6 – each to his own – but as an operator – I 100% want CRH – and I 100% do not believe that SRv6 is in any way shape or form an alternative to it – or something that I will ever run across any of the countries we operate in.
>
>
>
> Thanks
>
>
>
> Andrew
>
>
>
> Do we really want this?
>
>
>
> My two cents.
>
>
>
> Best regards,
>
> Mach
>
>
>
> From: ipv6 [mailto:[e-mail]] On Behalf Of Zafar Ali (zali)
> Sent: Tuesday, May 26, 2020 3:02 PM
> To: Ron Bonica <rbonica=[e-mail]>; Chengli (Cheng Li) <[e-mail]>; 6man <[e-mail]>; [e-mail]
> Subject: CRH is not needed - Re: How CRH support SFC/Segment Endpoint option?
>
>
>
> Hi,
>
>
>
> It appears that some may have misunderstood the SRv6 solution and invented CRH.
>
> It is good to clarify these points.
>
>
>
> SRv6 offers the possibility to combine underlay and overlay instructions in a single SRH.
>
> However,
>
> ·         This is entirely optional
>
> ·         If one would like to spread the source routed policy between multiple extension headers, one can use SRv6 to do this
>
> o   SRH to hold the topological endpoints
>
> o   Any combination of other extension headers to hold VPN and/or Service information. For example, SRH works seamlessly with NSH as documented in a WG document https://tools.ietf.org/html/draft-ietf-spring-nsh-sr-02.
>
>
>
> Claiming that a new data plane is needed to achieve this separation is false.
>
>
>
> Claiming that CRH is needed to decrease the overhead of source-routed policy in IPv6 is incorrect, too. Many members of the SPRING working group have produced documents to extend the SRv6 solution for the specific purpose of minimizing the MTU overhead and/or supporting very long SID-lists on legacy hardware.
>
>
>
> Also, CRH is just a re-engineering of SR-MPLS Data Plane with IPv6 Control Plane [RFC8663] and RFC8663 is an already productized, deployed, proven, and standardized solution.
>
>
>
> 6man took 6 years to define SRH. The 6man WG spent a lot of efforts (1000s of email, dozens of document revision, dozens of IETF presentations, control plan work that is adopted by multiple workgroups, etc.).
>
>
>
> There is no need to define a new data plane, new control plane and associated management plane for the same purpose the IETF across multiple areas has worked for years.
>
>
>
> Thanks
>
>
>
> Regards … Zafar
>
>
>
>
>
> From: ipv6 <[e-mail]> on behalf of Ron Bonica <rbonica=[e-mail]>
> Date: Friday, May 22, 2020 at 10:17 AM
> To: "Chengli (Cheng Li)" <[e-mail]>, 6man <[e-mail]>, "[e-mail]" <[e-mail]>
> Cc: "[e-mail]" <[e-mail]>
> Subject: RE: How CRH support SFC/Segment Endpoint option?
>
>
>
> Cheng,
>
>
>
> The sole purpose of a Routing header is to steer a packet along a specified path to its destination. It shouldn’t attempt to do any more than that.
>
>
>
> The CRH does not attempt to deliver service function information to service function instances. However, it is compatible with:
>
>
>
> -          The Network Service Header (NSH)
>
> -          The Destination Options header that precedes the Routing header
>
>
>
> Both of these can be used to deliver service function information to service function instances.
>
>
>
>                                                                                                                      Ron
>
>
>
>
>
>
>
> Juniper Business Use Only
>
> From: Chengli (Cheng Li) <[e-mail]>
> Sent: Friday, May 22, 2020 2:56 AM
> To: 6man <[e-mail]>; [e-mail]; Ron Bonica <[e-mail]>
> Cc: [e-mail]
> Subject: How CRH support SFC/Segment Endpoint option?
>
>
>
> [External Email. Be cautious of content]
>
>
>
> Hi Ron,
>
>
>
> When reading the CRH draft, I have a question about how CRH support SFC?
>
>
>
> For example, we have a SID List [S1, S2, S3, S4, S5], and S3 is a SFC related SID, how to indicate that? By PSSI? [1]
>
>
>
> But how to know which segment endpoint node/egress node should process this PSSI? At the beginning of the SRm6 design, this is described in [2]. But you deleted the containers [2].
>
>
>
> Without that, I don’t really understand how SFC can be supported.
>
>
>
>
>
> Best,
>
> Cheng
>
>
>
>
>
>
>
> [1]. https://tools.ietf.org/html/draft-bonica-spring-sr-mapped-six-01#section-4.1
>
> [2]. https://tools.ietf.org/rfcdiff?url2=draft-bonica-6man-seg-end-opt-04.txt.
>
>
>
>
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------