Re: [CCAMP] Objective function draft

[Igor Bryskin <[e-mail]>]

I completely agree with George. Signaling the objective function is no different from signaling EROs or affinities, this is just another service specific policy, an instruction to the network as to how the service needs to be routed across the network domain. I think Georges documents complement our (GMPLS-ENNI) solution. We have the same objective but target different groups of clients. Specifically, George's clients are those who cannot or won't deal with the routing information leaked by the network, but have certain preferences' as to how their services need to be routed and rely fully on the network to do the routing. Our clients are those that are capable and willing to process the network advertisements and to control the routing themselves.

Cheers,
Igor

-----Original Message-----
From: [e-mail] [mailto:[e-mail]] On Behalf Of George Swallow (swallow)
Sent: Tuesday, September 18, 2012 9:45 AM
To: Gert Grammel; Julien Meuric
Cc: [e-mail]
Subject: Re: [CCAMP] Objective function draft

Gert -

Agreed.  I was loose in my terminology.  The Objective function is information that is signaled to the routing function as a constraint.
There is a strong analogy and precedent for RSVP-TE/GMPLS of a loose-hop in an ERO.  Another example would be Resource Affinities signaled in the Session Attribute object.

...George

On 9/17/12 5:22 PM, "Gert Grammel" <[e-mail]> wrote:

>Hi George,
>
>The objective function is in the end a routing information. Mixing 
>routing and signaling in one protocol is something I don't feel 
>comfortable with.
>
>In other words, if routing is needed between client and server, UNI is 
>the wrong choice. ENNI should be considered instead and 
>Draft-beeram-ccamp-gmpls-enni would be a good starting point.
>
>
>Gert
>
>
>
>________________________________________
>From: [e-mail] on behalf of George Swallow (swallow)
>Sent: Monday, September 17, 2012 12:19:21 PM
>To: Julien Meuric
>Cc: [e-mail]
>Subject: Re: [CCAMP] Objective function draft
>
>Hi Julien -
>
>On 9/17/12 9:37 AM, "Julien Meuric" <[e-mail]> wrote:
>
>>Hi George.
>>
>>Sorry for the late response. You are right: the minutes are not enough 
>>to trace the full discussion (which we also resumed right after the 
>>meeting). Let us start by thanking Adrian (as AD? former PCE co-chair?
>>author of... ;-) ) for bringing the PCE-associated vocabulary to a 
>>common understanding.
>>
>>Actually my concern is sustained by 2 points:
>>
>>1- The scope of the draft is about giving control of the routing 
>>objective function to the client node facing a transport layer. I see 
>>already several existing solution to achieve it:
>>- a PCEP request from the signaling head node is an option (which is 
>>associated to the advertisement of the supported objectives in PCEP);
>>- building IGP adjacencies between client and transport edge nodes 
>>(a.k.a. "border model") is another one.
>>In this context, it do not think extending RSVP-TE for this kind of 
>>application is worth the effort, since the requirement can already be 
>>addressed.
>
>As I understand it, in the optical and OTN cases, the border model 
>would not be popular as in many organizations this crosses political 
>boundaries.
>
>The point of the draft is to keep the UNI implementation simple and not 
>require a PCEP on the uni-c or necessarily on the uni-n.  We will keep 
>the format aligned so if the UNI-N needs to make a request of a PCS, it 
>can do so rather simply.
>>
>>2- There are cases when previous options are ruled out of a given 
>>deployment. I do believe that it is not simply due to protocol 
>>exclusion, but rather to the fact that the SP wants transport routing 
>>decisions to remain entirely within the transport network (in order to 
>>fully leave the routing policy in the hands of people doing the layer 
>>dimensioning). Thus, I feel this trade-off in path selection tuning is 
>>rather unlikely to happen and I fear we may be talking about RSVP-TE 
>>over-engineering here.
>
>The idea is simply to allow the client to express its needs/wishes.  
>The UNI-N remains in control.  By policy it can use the objective 
>function or not.  Further if it does use the objective function and 
>fails to find a path it can either say that there was no path or it 
>proceed to setup what it can.
>
>>(That is also why I preferred to consider your I-Ds separately during 
>>the CCAMP meeting.)
>
>Agreed.  I will ask for separate slots.  The discussion at the end was 
>rather disjointed.
>
>>
>>However, my comments are mostly related to the client/transport 
>>relationship. If the I-D is extended to cover more use cases with 
>>wider scopes (Adrian has made interesting suggestions), turning the 
>>overlay interconnection into one among a longer list, then my 
>>conclusion may be different.
>
>I'm happy to widen the scope in this way.
>
>...George
>
>>Regards,
>>
>>Julien
>>
>>
>>Le 11/09/2012 21:28, George Swallow (swallow) a écrit :
>>> Julien -
>>>
>>> Reading the CCAMP notes (which capture little of the actual
>>> discussion) I see that there may have been a perception in the room 
>>> that PCE functionality at the UNI-N was assumed (actual or proxy).
>>>
>>> This is not the case. The reason for our draft is that with the UNI, 
>>> much of the functionality that resides at the headend is moved to 
>>> the UNI-N. So the UNI-C needs a way to express an objective function 
>>> even if there is no PCE.
>>>
>>> Operationally it seems burdensome to require a PCEP at the UNI-C and 
>>> a PCEP at the UNI-N, when all that is being done is enabling the 
>>> UNI-N to perform what the client would do if it were connected to 
>>> the network via a normal link.
>>>
>>> Do you still object to the draft?
>>>
>>> Thanks,
>>>
>>> ŠGeorge
>>
>>
>
>_______________________________________________
>CCAMP mailing list
>[e-mail]
>https://www.ietf.org/mailman/listinfo/ccamp
>
>

_______________________________________________
CCAMP mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/ccamp