Re: [Idr] WG adoption call for draft-abraitis-bgp-version-capability-08, to end September 25

["Jakob Heitz (jheitz)" <[e-mail]>]

Hey Robert,

That's a super cool idea. I like it.

Regards,
Jakob.

From: Idr <[e-mail]> On Behalf Of Robert Raszuk
Sent: Thursday, October 29, 2020 4:31 AM
To: Jeffrey Haas <[e-mail]>
Cc: John Scudder <jgs=[e-mail]>; IDR List <[e-mail]>; Donatas Abraitis <[e-mail]>
Subject: Re: [Idr] WG adoption call for draft-abraitis-bgp-version-capability-08, to end September 25

Just to clarify one point ... I was not seeing a need to use domain names. Just either IP address of bgp peer or implicitely use bgp peer's address.

https url would be actually pretty short ... just the path to the opaque text.

I am more interesting on your view of how to do NSR/ISSU with current proposal.

Cheers,
R.



On Thu, Oct 29, 2020, 12:18 Jeffrey Haas <[e-mail]<mailto:[e-mail]>> wrote:
Robert,

On Thu, Oct 29, 2020 at 09:22:29AM +0100, Robert Raszuk wrote:
> > - I think requiring extended optional params is a good idea for this.  It
> >   mitigates the necessity for having to do do the math to squeeze stuff in
> >   or not.
>
> How about we would just carry a fixed size URL reference to this
> effectively static and opaque to the bgp protocol information instead of
> actual text string ?

I think this is problematic in a few senses:
- Sure, you could take this idea to the extreme that we'll just have a
  single four-byte field with a FCFS registry that everyone uses and has
  private space for a local registry.  And people would hate that.  You're
  devolving to pre-registration for something that may change frequently.
- Sure, you could just reserve char version[64] in the structure, but domain
  names may vary in length.  And when you move to punycode i18n domains,
  this could be even messier.  See prior issues with RFC 8203.
- I strongly expect that some operators will want to stick in their own
  strings here.  "Role" potentially in combination with "version".

> IMO anything which is static and is not needed for protocol operations,
> best path selection etc ... should be passed as a pointer.
>
> Fetching such string could be done in spare CPU times well before need to
> locally present it or at the run time when someone executes a show cmd or
> other form of query api.

Which is really a lot of typing to say "we're not exchanging this out of
band ... why?"  Which is still a legitimate argument.  It's why I think the
use case, although slightly helpful, has a lot of weaknesses.

The one slight boost I give the core use case that we're regularly seeing in
data center cases is protocol stacks are being spun up with very little
additional components. This provides a push to consolidate channels for
sending critical information.

-- Jeff