Re: [Detnet-dp-dt] FW: New update of the draft available

[Loa Andersson <[e-mail]>]

Bala'zs,

OK - I think we are close to consensus here.

In line please.

On 2017-03-14 17:11, Balázs Varga A wrote:
> Hi Loa,
>
> Yes, I am fine with the tunneling, no issues on that. Now I see
> your point better.
>
> Your concern is regarding the control plane and not the data plane.

Well, almost - I'm concerned that for some control planes it is not
possible to distinguish between a node that act as P or PE when setting
up a PW. I it is my gut feeling that if we want to change the control
plane to do this it would be a major change.

> In my view request for PRER
Jouni calls the "Packet Replication and Elimination function (PREF)"
in the draft - should we stay with that terminology??

  should be (i) PW and (ii) node specific:

This a a big bite to bite of for example for LDP if we are talking
about an un-tunneled PW, if we are talking about a MS-PW with a
L-label (PSN tunnel) it is state of the art.

It might be that I misunderstood your earlier comment:

"I think the addition of the DetNet Flow-ID to the "PW Encapsulation
  header" changed our former discussions. If the DetNet Flow-ID is
  present, we can change the d-pw label during the transport like in
  case of a legacy MS-PW."

I took that to mean that we could change he PW label "during transport"
and thought you meant at P nodes also. Reading it more careful and
seeing "like in case of a legacy MS-PW" I think you meant that the
PW-label can be changed at every DA-S-PE node. And pushed/popped at the
DA-T-PE nodes.

If this is the case we are in agreement!

> - PRER points need proper planning before the setup of the DetNet-PW
> (which nodes should be selected for x-PE roles, already established
> overlay tunnel might be considered during the design also, etc.)

hmmm  - yes, but setting up an un-tunneled PW through PA-1/PE-B2, will
be non-deterministic, the control plane for one overlay network will
view PA-1/PE-B2 as a P node (un-capable of DetNet NSP and PREF), and
send the TLV/object that request DetNet NSP and PREF that is intended
for PE-A2. Since PA-1/PE-B2 are capable of DetNet NSP and PREF, since
it is a DA-A-PE in the other overlay network, it will try to act on
the  DetNet NSP and PREF (and very likely fail).

> - signaling for PW setup should contain the planning results

yes it does, only that this planning should include the L-label overlay.

>
> So in your example, when PE-A1 signals for the PW to PE-A2, the role
> of P-A1 should be clear (i.e. no PRER for PW between A1 and A2) and
> included in signaling.

Sure - but since PA-1/PE-B2 is capable of DetNet NSP and PREF, it can't
be allowed to see the signalling for PE-A2 or packets that are intended
for PE-A2. One reason we need the L-label.

> Simultaneously PE-B1 should signal explicitly that PE-B2 is a PRER
> point for the PW between PE-B1 and PE-B3.
Yes - in this case the MS-PW label is exposed, it is understood that
NSP and PREF is required, since the signaling is specifically to
PA-1/PE-B2 (e.g. tLDP) and when forwarding the L-label is php'ed or
popped.

/Loa
>
> Cheers
> Bala'zs
>
> -----Original Message-----
> From: Detnet-dp-dt [mailto:[e-mail]] On Behalf Of Loa Andersson
> Sent: Monday, March 13, 2017 4:05 PM
> To: Balázs Varga A <[e-mail]>om>; [e-mail]
> Subject: Re: [Detnet-dp-dt] FW: New update of the draft available
>
> Balázs,
>
> I'd not seen this before. I thought that we'd reached consensus on the need to tunneling across the P routers after the comments from Jouni and Yuanlong. Please see inline.
>
> On 2017-03-13 18:09, Balázs Varga A wrote:
>> Hi, mail below seems to be lost in hyperspace. I send it again, sorry
>> for possible duplicates ... Cheers Bala'zs
>>
>> -----Original Message-----
>> From: Balázs Varga A
>> Sent: Saturday, March 11, 2017 7:54 PM
>> To: 'Loa Andersson' <[e-mail]>nu>; [e-mail]
>> Subject: RE: [Detnet-dp-dt] New update of the draft available
>>
>> Hi Loa,
>>
> <snip>
>
>>> d-pw label with signaling that exposes the "request for DetNet
>>> NSP/FRER" to P nodes
>> Could You explain it more, I do not see your point. Do You propose that the L-label value would control whether a PW is a DetNet PW with PRER or not?
>>
>
> We discuss the "overlay network" in terms of P and PE nodes. The P node does not not do PRER, while the PE may do so if we tell them.
>
> However one node by serve as P in one overlay network, and as PE in another.
>
> Consider:
>
>                  +------+
>                  | PE-B1|
>                  +------+
>                     |
>                     v
>                  +------+
>                  | P-B1 |
>                  +------+
>                     |
>                     v
>                  +------+
>   +------+       | P-A1 |       +------+       +------+       +------+
>   | PE-A1|------>| PE-B2|------>| P-A2 |------>| P-A3 |------>| PE-A2|
>   +------+       +------+       +------+       +------+       +------+
>                     |
>                     v
>                  +------+
>                  | P-B2 |
>                  +------+
>                     |
>                     v
>                  +------+
>                  | PE-B3|
>                  +------+
>
>
> Ledgend: PE-A = DA-*-PE, PE node in network A
>           P-A  = P node network A
>           PE-B = DA-*-PE, PE node in network B
>           P-B  = P node network B
>
> Problem is signalling!
>
> If have the PW-Label un-tunneled PE-B will establish the unprotected PW to PE-B3, requesting PERF of the PERF capable nodes.
>
> Node PE-B2 will do PREF, which is fine, that is what it is supposed to do. PE-B3 will eliminate duplicates as it is supposed to do.
>
> PE-A1 will also try to set up an un-tunneled PW to PE-A2, PA-2, PA-3 and
> PE-A2 will do what they are supposed to do. The problem is that the node "P-A1" is also "PE-B2" and will understand the request for PERF, which it is not supposed to do.
>
> Our solution is to use the L-label to tunnel between DA-*-PE.
>
> /Loa
>
>
>
>> Thx & Cheers
>> Bala'zs
>>
>> -----Original Message-----
>> From: Detnet-dp-dt [mailto:[e-mail]] On Behalf Of
>> Loa Andersson
>> Sent: Saturday, March 11, 2017 3:38 AM
>> To: [e-mail]
>> Subject: Re: [Detnet-dp-dt] New update of the draft available
>>
>> Jouni,
>>
>> Mostly looks very good, the thing I don't understand is why you still say that the L-label is optional, it really isn't.
>>
>> - we need it to deliver the d-pw label unchanged
>> - since we can't guarantee that the "P" nodes are not able to do
>>    DetNet NSP, we can't set up the d-pw label with signaling that
>>    exposes the "request for DetNet NSP/FRER" to P nodes.
>> - needed for protection
>> - the end-to-end tunnel must be the innermost tunnel, carrying
>>    pw label
>>
>> There is one case, if two DA-*-PEs are immediately adjacent, there the L-lable will be an implicit NULL label and not appear in the stack, but for the control plane it is there.
>>
>> /Loa
>>
>> On 2017-03-11 07:25, Jouni Korhonen wrote:
>>> * Added Loa’s comments on the L-label.
>>> * Added Janos’ comments.
>>> * Added extended forwarder text.
>>> * Added (speculative text.. can be removed) D bit to flow-id word so
>>> that we can check in a ring case the direction of the flow (note this
>>> does not double history buffer space as claimed.. did not bother to
>>> fix that)
>>> * reworked the PW encapsulation pictures.
>>> * Added more content to DA-*-PE descriptions.
>>>
>>> - Jouni
>>>
>>>
>>
>

-- 


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