Re: [Detnet-dp-dt] Using MS-PW concept for the d-pw

[Loa Andersson <[e-mail]>]

Folks,

I think every step to try summarize is good.

I still miss some corner stones and some context.

If we want to design existing control planes out, we can almost
anything.

If we want design existing control planes in, we have to consider
what they'll do and take that into account.

I think, but is not sure, that we now have both "god box" and LDP,
RSVP-TE and segment routing in scope. I'd like to keep it that way.

If we can agree on that we can start evaluating the different proposal 
against this back ground, or whatever control we think is appropriate.

/Loa

On 2017-02-25 06:44, Jouni Korhonen wrote:
> Folks,
>
> Let’s try to get to a rough agreement here. We are going around very similar approaches. I would like to progress with the draft text further where I am now. There’s other stuff to dig into still.
>
> We seem have three candidates on table (all based on MS-PWs):
>
> 1) Global d-pw labels, L-Labels necessary (Jouni)
> 2) Per PW d-pw labels, virtual labels (or local-id), no L-Labels necessary (Balazs + Jouni)
> 3) Per PW d-pw labels, virtual labels/queue (or local-id), L-Labels (Yuanlong)
>
> Alternatives 2 and 3 are very close. The difference in elimination whether it uses a “virtual label” or a “internal port/queue” that achieves the same is technically minor. For clarity and documentation purpose a virtual label or node-id is probably cleaner. The difference to me between 2 and 3 is how the replication is done. Alternative 2 swaps d-pw labels only after the replication. Alternative 3 pushes L-labels in addition to d-pw label swaps. I would say it is safe to merge alternatives 2 and 3, and let the L-label layer to be an operational decision.
>
> Alternative 1 is more straight forward from the hw point of view than 2 and 3. No label swaps for the elimination and replication. Replication also pushes L-labels. HOWEVER, I am not sure how much headache it introduces to the control plane. Someone more knowledgeable in that domain than me should bring facts in (e.g., lecture us of RFC4447, 5003 and some 5036). With a static control plane (“god box”/PCE approach) one does not care since everything is just programmed from a single point that has a full view of the topology, every flow and label in use. Dynamic control plane (LDP, BGP, etc) is a different beast. However, whatever we do some work is needed on the control plane protocols i.e., bringing in the concept of the DetNet flow over MS-PW, replication and elimination. That may be as simple as new PW Types, Status Codes, etc, but can also be more..
>
> Any preferences? I am actually OK with both 1 and 2 (with optional L-labels). My decision between the two will eventually be impacted by the control plane complexity (the analysis is still to-do).
>
>
> - Jouni
>
>

-- 


Loa Andersson                        email: [e-mail]
Senior MPLS Expert                          [e-mail]
Huawei Technologies (consultant)     phone: [phone]