Hello Fabien, 

I agree with you, but this knowledge is only needed in a network with
multi-switching capability nodes. Running such a network is a Multi
Layer Network (MLN) problem. So it would be necessary to know the
capabilities of the interfaces of the nodes in the network, their
adaptation capabilities between switching types, and the LSP type.

Although I would not think that for the path computation it is needed to
know that the transit nodes support the same switching capabilities.
What you need to know is that you could build a technology tunnel (an
FA-LSP) over the nodes that do not support your end to end LSP switching
type. This is what MLN is supposed to do for you. The PCE could then
make an end to end path that includes other switching capabilities
provided that a trigger for the FA-LSP will be set up.

You might want to look at Eiji's two PCE MLN drafts. Hopefully, Eiji's
PCEP extensions will give everyone what they need.

Thanks,
Dan

-----Original Message-----
From: fabien.verhaeghe [mailto:[EMAIL PROTECTED] 
Sent: 03 April 2007 12:08
To: Tomohiro Otani; Adrian Farrel
Cc: [EMAIL PROTECTED]
Subject: Re: [Pce] Switching type Constraint in PCEP

Hi Tomo,

I read your draft pointed by Dan that indeed contains some valuable 
information.
Especially in section 2.2:

"For the simplicity of the analysis in path consideration, the below
   basic assumptions are made when the LSP is created.

       (1) Switching capabilities (SC) of outgoing links from the
           ingress and egress nodes (link1-2 and link4-3 in Figure 1)
           must be consistent with each other.
       (2) SC of all transit links including incoming links to the
           ingress and egress nodes (link2-1 and link3-4) should be
           consistent with switching type of a LSP to be created.
       (3) Encoding-types of all transit links should be consistent
           with encoding type of a LSP to be created."

Point 2 and 3 seems to imply that a PCE needs to know both the switching
and 
encoding type
of the LSP.
To me the information would typically be provided by the PCC (which
knows 
what kind of LSP is being established)
using PCEP.
Am I misunderstanding something or do you agree?

Regards
Fabien

----- Original Message ----- 
From: "Tomohiro Otani" <[EMAIL PROTECTED]>
To: "Adrian Farrel" <[EMAIL PROTECTED]>
Cc: "fabien.verhaeghe" <[EMAIL PROTECTED]>; "LE ROUX 
Jean-Louis RD-CORE-LAN" <[EMAIL PROTECTED]>; "Dan Li" 
<[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
Sent: Friday, March 30, 2007 1:44 AM
Subject: Re: [Pce] Switching type Constraint in PCEP


> Hi Fabien and Adrian,
>
> Looking at the discussion, path computation itself and a protocol
> staff such as PCEP are a different thing.  I believe, touch upon
> by Dan, that our cspf draft is a kind of guideline how to consider
> GMPLS-specific constrains including encoding type for path
computation.
> I hope that the cspf draft help Fabien to understand parameters.
>
> If you need any help for GMPLS specific extensions, please let me
know.
>
> With best regards,
>
> Tomo
>
>
>
> Adrian Farrel wrote:
>> Hi Fabien,
>>
>>> Reading the generic and inter layer requirement documents it is not
>>> clear to me if Switching Type is really specific to inter layer
>>> application.
>>> As far as I understand inter-layer requirement refers to path
>>> computation which result path may comprise multiple layer
technologies.
>>> Though I understand Switching type is required in this context, it
is
>>> also
>>> required for mono layer path computation in a multi-layer
environment
>>> (e.g. to computate a Lambda path using a TE Database that contains
>>> also TDM TE Links).
>>> So it is more a GMPLS specific requirement to me.
>>
>> I think you are correct. That is, the switching type constraint is
>> required in the multi-layer context, but it may have wider
applicability.
>>
>> Provided that the multi-layer work progress quite quickly, I would
>> suggest that it is convenient to group all of the protocol extensions
>> together. You can then select from the set of optional objects as
>> necessary for your application. I would hope that you would provide
your
>> review input to the multi-layer work to ensure that the protocol
>> extensions give you what you need.
>>
>> On the other hand, if the multi-layer work seems to get stuck or held
>> up, then I would understand you pushing to get the "GMPLS-specific"
>> extensions done separately.
>>
>>> Acually it is not clear to me what is the scope of generic and
>>> "application specific" requirements. Generic Requirement RFC4657
>>> section 5.1.16
>>> mention Switching and encoding type as "GMPLS specific
requirements".
>>> Does it mean "GMPLS specific" is "considered as generic?
>>
>> Hmmm.
>> This is a line we have not succeeded in drawing very clearly.
>>
>> However, where we seem to have arrived is that the PCEP I-D contains
the
>> protocol definition and "core" functional units. Other units that
were
>> identified in 4657 (such as XRO, objective funcions, ...) are going
into
>> seperate I-Ds. Following that model, the GMPLS-specific work
definitely
>> goes into a separate draft. As above, I am comfortable for that to be
in
>> the multi-layer I-D (perhaps we re-name to "GMPLS and multi-layer"?)
to
>> help keep the number of I-Ds manageable.
>>
>>> Concerning the necessity of "encoding type" in PCEP I have to admit
>>> that I'm not sure to correctly understand the meaning of this
>>> parameter in the GMPLS
>>> architecture :-).
>>
>> I think this puts you in the majority!
>>
>>> What I understand at least is that it is both an interface and
>>> an LSP property, so from this it seems to be required in PCEP.
>>> I would be interested to see discussion about encoding type
>>> requirement in PCEP cause it may clarify the encoding type meaning
to
>>> me ;-).
>>
>> So my view (other will surely correct me :-) is that encoding type is
>> only needed where the signal is unpacked for some reason. Typically,
>> this only happens at end points, so the PCE function is to ensure
that
>> the destination interface is capable of decoding the signal and
>> delivering the data. In general, it has been considered that the
>> destination capabilities are either known to the ingress (guaranteed
LSP
>> setup success - you wouldn't even bother trying to set up an LSP to
>> someone you knew couldn't understand you) or are not known in which
case
>> the LSP will either succeed or fail (but no amount of re-routing will
>> help). Now, it *is* possible that different interfaces on the egress
>> have different levels of support for encoding types, but this has
>> generally been considered as very advanced and not worth optimising
for.
>>
>> However, recent discussion has suggested that in certain
technologies,
>> signal regeneration also requires knowledge of the encoding type.
This
>> would certainly expand the requirement for PCE to know about encoding
>> type. I am not sure about the hardware specifics here and perhaps
>> someone can enlighten us.
>>
>> Cheers,
>> Adrian
>>
>>
>> _______________________________________________
>> Pce mailing list
>> [EMAIL PROTECTED]
>> https://www1.ietf.org/mailman/listinfo/pce
>>
>
> 


_______________________________________________
Pce mailing list
[EMAIL PROTECTED]
https://www1.ietf.org/mailman/listinfo/pce

_______________________________________________
Pce mailing list
[EMAIL PROTECTED]
https://www1.ietf.org/mailman/listinfo/pce

Reply via email to