[spring] PHP - Deep Listening (was: RE: Separating issues)

[Ron Bonica <[e-mail]>]

Suresh,

Fair enough. Let's review PHP, with open minds, and see if the motivation merits the proposed behavior....

Assume that an SRv6 node receives the following packet:


  *   IPv6 header. Destination Address == LOCATOR:0x0002 (0x0002 indicates that this is and end with PSP)
  *   IPv6 header. Next Header == Routing Header
  *   Routing header. Routing Type  == SRH
  *   Routing header. Segments Left == 1

According to the NP draft, the nodes should:


  *   Copy SID[0] to IPv6 header. Destination Address
  *   Update IPv6 header. Next Header
  *   Update IPv6 header, Payload Length
  *   Remove the Routing Header

Now that we understand how PHP works, we can compare PHP to a more orthodox alternative. That is, to decrement Segments Left (to 0) and forward the packet without PHP. Downstream nodes SHOULD ignore the SRH, because Segments Left is equal to 0.

The following are possible benefits of PHP:


  *   To save bandwidth on the final segment
  *   To optimize for ASIC some particular ASIC on the destination node, possibly mitigating the effect of a very large SRH
  *   Others?

Do either of these motivations merit the variation from the more orthodox IPv6 processing?

                                                                Ron




Juniper Business Use Only
From: Suresh Krishnan <[e-mail]>
Sent: Saturday, December 7, 2019 2:14 AM
To: 6man <[e-mail]>; SPRING WG <[e-mail]>
Cc: Brian E Carpenter <[e-mail]>; Andrew Alston <[e-mail]>; Ole Troan <[e-mail]>; Ron Bonica <[e-mail]>; [e-mail]; rtg-ads <[e-mail]>; Bob Hinden <[e-mail]>; Fernando Gont <[e-mail]>
Subject: Separating issues (was Re: [spring] We don't seem to be following our processes (Re: Network Programming - Penultimate Segment Popping))

(Apologies up front. I am about to get on a 10 hr flight and will be unable to respond for at least that period)

Hi all,
  Picking the last message in the thread to reply to. It looks to me that there are at least two different (but related) issues being discussed here

a) Spring SRv6 NP behavior (related to the WGLC of that draft)
b) The Header insertion drafts and how to deal with them

I really think that

a) should preferably stay in the spring ML and a pointer to the discussion sent to the 6man mailing list would be in order
b) should preferably stay in the 6man ML

I think the communities for the two drafts are different and I think the discussions can be more focused if the issues are addressed by the relevant wgs. As to what happens if the spring draft hits the IESG and contains text in violation of RFC8200, I had already sent my thoughts about this back in September

https://mailarchive.ietf.org/arch/msg/spring/uA-WxxgBJeMu65SkrKCTL5BJMcU<https://urldefense.com/v3/__https:/mailarchive.ietf.org/arch/msg/spring/uA-WxxgBJeMu65SkrKCTL5BJMcU__;!8WoA6RjC81c!VVaoO_crkvFgYmFJ0jxbrktSCLgt3HBTWrc-PPq5J1uMrJpL9oObm6RakvUsS3u0$>

with the relevant text

"If a draft violates RFC8200 and it hits the IESG for evaluation, I will certainly hold a DISCUSS position until the violations are fixed."

*In my view*, the authors of the SRv6 NP draft have made an effort to address these violations by removing the header insertion from the draft. We can continue discussing whether penultimate hop popping constitutes a violation on the spring mailing list (I intend to respond to Fernando'a mail there).

Thanks
Suresh

On Dec 7, 2019, at 11:07 AM, Fernando Gont <[e-mail]<mailto:[e-mail]>> wrote:

On 6/12/19 23:47, Brian E Carpenter wrote:
Again, comment at the end...
On 07-Dec-19 14:37, Fernando Gont wrote:
On 6/12/19 22:15, Brian E Carpenter wrote:
[...]

and if such a thing is required, an update to RFC8200 should be done.

Why does that follow? Alternatively, draft-ietf-spring-srv6-network-programming could acknowledge that it deviates from RFC8200.

You can deviate from s "should", not from a "must". This is an outright
violation of a spec, rather than a mere "deviation".


Whether that's acceptable would be a question for the IETF Last Call rather than any single WG.

I would expect that a WG cannot ship a document that is violating an
existing spec, where the wg shipping the document is not in a position
of making decisions regarding the spec being violated.

That would be like a waste of energy and time for all.



At the moment, the draft only mentions RFC8200 in a context that discusses neither insertion nor removal of extension headers, which is beside the point. Like draft-voyer, if it describes a violation of RFC8200, shouldn't that be explicit in the text?

There's a lot of jargon in draft-ietf-spring-srv6-network-programming. I can't tell from the jargon whether "insert" means "insert on the fly" and whether "Pop the SRH" means "delete on the fly". Should those terms be clarified before the draft advances?

Well, if it's not clear to you, it would seem to me that the simple
answer would be "yes".

But if "insert" refers to the encapsulating node at the SR domain ingress, it's no problem, and if "pop" simply means doing normal routing header processing, it's no problem. It simply isn't clear in the text, at least not clear to me.

The fact that a folk that has been deeply involved with IPv6 cannot
unequivocally tell what they talking about should be an indication with
respect to how ready the document is to be shipped.

(pop when you are the destination but SL!=0 is essentially 'in the
network removal')

THanks,
--
Fernando Gont
SI6 Networks
e-mail: [e-mail]<mailto:[e-mail]>
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492