In the shared memory case, indeed, it is very reasonable to look at the
proxy as an arm of the function.
In SFC, we do look at it (the proxy as an arm of the SF) that way more
generally. We can do that because the SFC architecture calls that out.
Given that the other cases do involve network transfer of packets, and
thus are visible, and given that neither the SR architecture, the MPLS
architecture, nor the SR architecture call out such things, it seems
that we need to be explicit. To be clear, I think it is a very
reasonable approach. Much mroe reasonable, for example, than NAT in
its various incarnations. All I am asking is that we state explicitly
that we understand that as an architectural component this violates
existing rules, and does so over a limited scope so as to preserve the
desired / needed behavior over the larger scope. The one thing we may
need to be clear about is the degree of proximity required between the
SF and its serving proxy. If we treat this as legitimate arbitrary
networking, it gets out of hand.
Yours,
Joel
On 7/26/2020 3:33 PM, Francois Clad (fclad) wrote:
Hi Joel,
Thank you for your email.
A proxy and its associated SF can be seen from the network as a single entity:
a packet enters this entity from the network, gets processed by the SF, and
exits towards the network. The packet modifications that occur between the
entry and exit of this entity are compliant with existing standards.
Whatever happens between the proxy and SF is internal processing and invisible
to the network. However, a network operator or controller needs some
information about the proxy’s internal behavior to determine an appropriate
SID-list through the SF and possibly configure the proxy.
Cheers,
Francois
On 25/07/2020 20:23, "spring on behalf of Joel M. Halpern"
<spring-boun...@ietf.org on behalf of j...@joelhalpern.com> wrote:
<chair hat off for now; This issue may, depending upon resolution,
become a chair issue, in which case, I will look at it through a
different lens. Heck, I may even disagree with myself.>
Let me start by saying that I understand and support what the draft is
trying to do. While I like SFC, I am under no illusions that it is or
should be the only answer to service chaining / service programming.
Further, I understand what the proxies are for. They seem necessary.
To deploy this stuff, we have to be able to work with older equipment.
Proxies seem the best way to do so.
The document is even clear that proxy is a new kind of thing. Good.
In order to do its job, and as I read this document, the SR proxies (of
various kinds) violate the rules for MPLS processing, SRH processing,
and IPv6 processing at various points. They have to.
It seems to me that we need to accept this requirement, and state it
clearly. Most likely, this would suggest that we will want some form of
signoff from the MPLS and 6man working groups that these violations, for
these specific reasons, are acceptable to the community. Personally, I
would rather have the discussion soon, rather than pretending it is a
non-issue and having the discussion during IETF last call.
Maybe I am misreading, and things are less conflicted. That would be
great.
Yours,
Joel
<chair hat returning to wherever it belongs.>
_______________________________________________
spring mailing list
spring@ietf.org
https://www.ietf.org/mailman/listinfo/spring
_______________________________________________
spring mailing list
spring@ietf.org
https://www.ietf.org/mailman/listinfo/spring