Re: [Teas] ACTN update

Igor Bryskin <> Mon, 21 November 2016 15:14 UTC

Return-Path: <>
Received: from localhost (localhost []) by (Postfix) with ESMTP id E5492129AC9 for <>; Mon, 21 Nov 2016 07:14:46 -0800 (PST)
X-Virus-Scanned: amavisd-new at
X-Spam-Flag: NO
X-Spam-Score: -5.716
X-Spam-Status: No, score=-5.716 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, DC_PNG_UNO_LARGO=0.001, HTML_MESSAGE=0.001, RCVD_IN_DNSWL_MED=-2.3, RCVD_IN_MSPIKE_H3=-0.01, RCVD_IN_MSPIKE_WL=-0.01, RP_MATCHES_RCVD=-1.497, SPF_PASS=-0.001] autolearn=ham autolearn_force=no
Received: from ([]) by localhost ( []) (amavisd-new, port 10024) with ESMTP id MnaFvPAvQIGT for <>; Mon, 21 Nov 2016 07:14:43 -0800 (PST)
Received: from ( []) (using TLSv1 with cipher RC4-SHA (128/128 bits)) (No client certificate requested) by (Postfix) with ESMTPS id B128D1294DF for <>; Mon, 21 Nov 2016 07:14:41 -0800 (PST)
Received: from (EHLO ([]) by (MOS 4.3.7-GA FastPath queued) with ESMTP id DBB86835; Mon, 21 Nov 2016 15:14:37 +0000 (GMT)
Received: from ( by ( with Microsoft SMTP Server (TLS) id; Mon, 21 Nov 2016 15:14:36 +0000
Received: from ([]) by ([]) with mapi id 14.03.0301.000; Mon, 21 Nov 2016 07:14:30 -0800
From: Igor Bryskin <>
To: Daniele Ceccarelli <>, "Scharf, Michael (Nokia - DE)" <>, "TEAS WG (" <>
Thread-Topic: ACTN update
Date: Mon, 21 Nov 2016 15:14:29 +0000
Message-ID: <0C72C38E7EBC34499E8A9E7DD007863908F1EE5F@dfweml501-mbx>
References: <> <0C72C38E7EBC34499E8A9E7DD007863908F18AA2@dfweml501-mbx> <> <0C72C38E7EBC34499E8A9E7DD007863908F18B77@dfweml501-mbx> <> <> <>
In-Reply-To: <>
Accept-Language: en-US
Content-Language: en-US
X-MS-Has-Attach: yes
x-cr-puzzleid: {DB03234D-7DA9-4ECF-B697-7049530AF54F}
x-originating-ip: []
Content-Type: multipart/related; boundary="_004_0C72C38E7EBC34499E8A9E7DD007863908F1EE5Fdfweml501mbx_"; type="multipart/alternative"
MIME-Version: 1.0
X-CFilter-Loop: Reflected
X-Mirapoint-Virus-RAPID-Raw: score=unknown(0), refid=str=0001.0A020206.58330F5F.019B, ss=1, re=0.000, recu=0.000, reip=0.000, cl=1, cld=1, fgs=0, ip=, so=2013-06-18 04:22:30, dmn=2013-03-21 17:37:32
X-Mirapoint-Loop-Id: bfe9e99d81ebc69cdd62452480a83a55
Archived-At: <>
Subject: Re: [Teas] ACTN update
X-Mailman-Version: 2.1.17
Precedence: list
List-Id: Traffic Engineering Architecture and Signaling working group discussion list <>
List-Unsubscribe: <>, <>
List-Archive: <>
List-Post: <>
List-Help: <>
List-Subscribe: <>, <>
X-List-Received-Date: Mon, 21 Nov 2016 15:14:47 -0000


>From your slides:

*         XXXX (I don't have a name for that functionality) asks the MDSC if an underlay infrastructure meeting the requirements is available, if not create it.

You have to be more specific and describe what you mean by "underlay infrastructure". As mentioned in previous emails there are reasons as to why you cannot just say in one breath "The tunnels I refer to are exactly TE tunnels at any layer between 0 and 3"

L3 TE tunnels

L0/L1 TE tunnels

IP topology re-configuration

Not required


Tunnel termination points

PEs directly supporting L3 service

Any pair of PE-P, P-P, PE-PE (to be) adjacent in IP layer


Fine, could be tailored and dynamically (re-)adjusted to the supported L3 service needs

Coarse, implies unused resources

You also said:
"In theory it is possible to create a L3VPN directly on top of L0 or L1 tunnels but it is not an appealing use case..."

Well, but this would be not L3VPN, rather, much more expensive L3PN, right?

Furthermore, from you slides:

*         Customer asks for a L3VPN between CE1 and CE2 (or between AP1 and AP2 if supported) with TE constraints using L3SM model enhanced with TE (missing).

Why? What is missing? If you want a customer to be able to map a L3 service onto the provider network resources, the client can do it:

a)      Using TE topology model negotiate desired network abstraction in the form of one or more abstract topologies;

b)     Using the provided abstract topology (ies) compute desired placement and configure desired TE tunnels via TE tunnel model;

c)      Using L3 service model map a L3 service onto the tunnels.

All three models are undeniably necessary. But why do we have to integrate them into an IETF endorsed version of Unified Theory Of Everything (a.k.a. ACTN), rather than coming up with a simple Best Known Practices document? As far as I can recall, it was never the policy of IETF to produce such things. One reason for that is that folks solving their own real (rather than somebody else's hypothetical) problems will surely come up with better ways to mix-n-match the models than we can suggest, especially, when each of such models may/does have a very long list of augmentations.


From: Daniele Ceccarelli []
Sent: Friday, November 18, 2016 12:29 PM
To: Scharf, Michael (Nokia - DE); Igor Bryskin; TEAS WG (
Subject: RE: ACTN update

The entity that creates the tunnels is the MDSC. The MDSC is a functionality and it can be included into the service orchestrator or the network orchestrator.

If the service is between nodes belonging to the same domain it is obviously possible to do it directly via the domain controller or (if the operator wants to have only the Service/Network Orchestator as unique point to control the network) via the Service/Network Orchestrator that in this case would speak just with one PNC.

Regarding the layers in the slide I only considered one layer (MPLS) because the service is done at one layer. I have added a slide with a packet-otpical scenario but I think it is irrelevant wrt the service to tunnel binding.

Critiques to the slides are more than welcome but also proposals on how to improve them would be highly useful.


From: Scharf, Michael (Nokia - DE) []
Sent: venerdì 18 novembre 2016 11:57
To: Daniele Ceccarelli <>om>; Igor Bryskin <>om>; TEAS WG ( <>
Subject: RE: ACTN update

In case of L3VPNs and L2VPNs, a "service orchestrator" may not have to create tunnels (or pseudowires etc.). This can be a function of a "network orchestrator" or a "domain controller" (or however we call them(), or it can be a standalone function outside any "orchestrator".

I agree with Igor that the picture does not cover well the case of having different IP and optical controllers - simply because it only shows one layer.


From: Teas [] On Behalf Of Daniele Ceccarelli
Sent: Friday, November 18, 2016 12:32 AM
To: Igor Bryskin; TEAS WG (<>)
Subject: Re: [Teas] ACTN update

Hi Igor,

Some more thoughts in line in red.


From: Igor Bryskin []
Sent: venerdì 18 novembre 2016 03:14
To: Daniele Ceccarelli <<>>; TEAS WG (<>) <<>>
Subject: RE: ACTN update

Hi Daniele,

Please, see in-line.


From: Daniele Ceccarelli []
Sent: Thursday, November 17, 2016 12:29 PM
To: Igor Bryskin; TEAS WG (<>)
Subject: RE: ACTN update

Hi Igor,

I've used the L3SM example because it is the customer facing model and it is what we have on top of the MDSC and the orchestrator and, being on top of the MDSC, means that the sigle-multi/domain aspects are hidden (on top of the MDSC you don't care and you don't know if between CE1 and CE2 or between AP1 and AP2 there is one, two or N domains.

In the context of L3SM multi-domain service coordination is NOT mandatory. The L3SM can be used on top of the MDSC in case of multiple domains  and in case of single domain it can be used either on top of the MDSC or on top of the PNC, this is a deployment choice and I just used an example.

IB>> I agree with you, except that L3 service controller does not seat on top of MDSC as a direct client. The L3SM<=>MDSC relationship is indirect and much more complex. Also from your slide below it is hard to interpret that L3SM nulti0domain coordination is not mandatory (IMHO by the way it is simply not needed)
DC: I get your point. The picture seems to mandate a multi-domain coordination between the "service" boxes, which is not the case, that's correct. The message that it wants to convey are the following:

1.       Coordination between the MDSC and the Service function in the Service Orchestrator is needed (in order to be able to map a service against a tunnel or a VN). It applies to single or multi domain and it doesn't care if it is single or multiple domains

2.       The MDSC does not the service

3.       The service can be created also at Controller level (within its own domain) and it is not the PNC that is doing that.

Hence a Service function box is needed in all the boxes, but this does not mean that they coordinate among them.


The interaction between the overlay network (i.e. the services) and the underlay/infrastruscture (i.e. the VN, tunnels etc) is mandatory if you want to map a service against a given tunnel or set of tunnel.

IB>> It is not clear what do you mean by tunnels. Normally L3 service (e.g. L3 VPN) PEs are connected via a mesh of IP/MPLS layer  RSVP-TE tunnels and L3 service specific LSPs are nested inside them. In any case, the TE part does not go deeper than IP/MPLS layer, and said RSVP-TE tunnels are not re-configured necessarily with every addition of a new L3 service. So, direct interaction between L3 service and transport service controllers is not mandatory (especially if the two are owned by separate organizations and implemented by separate vendors).
DC: I don't disagree with that. It must be possible to do the service to tunnel binding (i.e. I want L3VPN X to use tunnel Y) or in the classic way with not such a binding (i.e. I want an L3VPN between A and B and I don't care which tunnel will be used).
The binding does not mandate to create a new tunnel for each service obviously. Existing ones can and must be reusable.

If you don't want to do so you don't need ACTN or SDN in the network and you can simply operate the network as you did yesterday, with the tunnels created by LDP and the unawareness of which service uses which is simply enough to kwno that a packet goes from a source to a destination and you don't care about the path, the constraints related to the service, and the capability to know which services are used by which tunnel.

IB>> The point is that L3 service SDN controller and transport service (T-SDN) controller are both needed, but the two do not have to be integrated into one solution. It should be possible for an operator to cherry pick the controllers (e.g. L3 service controller provided by JNPR and T-SDN controller provided by Huawei).
DC: again agree, what in the figure makes you think that this is not the case?


From: Igor Bryskin []
Sent: venerdì 18 novembre 2016 01:08
To: Daniele Ceccarelli <<>>; TEAS WG (<>) <<>>
Subject: RE: ACTN update


Why in your opinion L3SM should know or care whether the provider looks after a single- or multi-domain network? There is a strong case for multi-vendor multi-domain transport network, but this is not necessarily true for a network providing L3 services, which (in contrast to, say.  L0/L1 transport network) could be a multi-vendor single domain network. In other words, why in the context of L3SM multi-domain service coordination is necessary?
Another question: when and how L3 service controller interacts with a transport controller (providing transport services)?

These clarifications I think would be useful.


From: Teas [] On Behalf Of Daniele Ceccarelli
Sent: Wednesday, November 16, 2016 9:32 PM
To: TEAS WG (<>)
Subject: [Teas] ACTN update

Dear WG and ACTNers,

Please find attached some slides trying to clarify terminology, roles and functionalities in ACTN (in line with the discussion we had at the mic during the TEAS session). It provides an architectural explanation and a possible workflow description trying to see if it fits with the architecture.

My proposal is to:

1.       Add a section in the framework document to explain the relationship and the split of roles between PNC, MDSC and existing components like orchestrator and domain controller.

2.       Add a section to the "Applicability of YANG models for ACTN" to explain how this impacts the definition of the MPI and what is missing (i.e. TE-Service model). This will have impacts also on the PCEP applicability.

3.    Write a new document defining service mapping model that provides mappings across LxSM and TE-service model and that can be extend to support different service models (L3VPN/L2VPN/L1VPN, VTS, Transport Connectivity service, etc). Dhruv, Young and I are already working on this.

Opinions? Thoughts?