I think we should clarify what’s kind of state is that we are talking first before making any assertions.
What we should avoid in Segments Routing panoramic view is that we shouldn’t introduce the per-flow state within the network.
Consider the different multicast services require different multicast trees, you need to store different “Replications Segments” for different multicast services.
Isn’t this per-flow state?
Based on the same principle, the Binding-SID introduces also some kind of per-flow state in the network——if every different flow needs to take different paths at the advertising node——In such case, you need to keep many or per-flow Binding-SID at the advertising node. It violates certainly the dogmas of segment routing.
The other SIDs(or that defined in RFC8986)that you mentioned are the states about the action of the related segment value and they are not Path-related, we can omit them.
In summary, the “Replication Segment SID” that introduced in this draft has the similar effects that the MPLS label derived from the various distribution protocol of multicast VPN solutions. Will you also call these MPLS states within the transit nodes as segment routing based?
Aijun Wang
China Telecom
On Dec 11, 2022, at 08:02, Robert Raszuk <rob...@raszuk.net> wrote:
Joel,
Yes IMO your understanding is correct.
That does also mean that anyone who is making assertions that the subject document is introducing per flow "state" is just wrong or simply does not understand SR dogmas.
Speaking personally, my understanding of the "stateless" aspect
of SR does not match what this email seems to describe.
SR is path stateless. There is no state related to specific
paths across the network. Any SID may be used, if it has relevant
meaning, in any path.
Advertising routers have internal state about what they mean when
the advertise SIDs.
Transit rotuers have state about where to forward packets based
on the current SID in the packet.
Binding SIDs have stored state about what stack of labels replace
the binding SID at the advertising router.
All these forms of state are considered by the community, as far
as I can tell, as acceptable and reasonable forms of state with
SR.
Personally, it seems to me that replication SIDs have much the
same kinds of state, and therefore fit well in the SR
architecture.
Yours,
Joel
On 12/9/2022 11:53 AM, Huaimo Chen
wrote:
Hi Everyone,
It seems that the
core value of segment routing is stateless (in the core of a
network). The document defines a new type of segment for Segment Routing
[RFC8402], called Replication
segment. Using Replication
segment is not stateless. This is not consistent with the
core value of segment routing. I oppose the progress of the
document.