Re: [OSPF] Regarding draft-ppsenak-ospf-te-link-attr-reuse-00

[Peter Psenak <[e-mail]>]

Hi Chris,

please see inline:

On 10/22/15 17:00 , Chris Bowers wrote:
> Peter,
>
> I would suggest making the text of the draft more explicit about the conditions under which a given link and set of attributes should be included in the TE Opaque LSA or the Extended Link Opaque LSA.  RFC3630 is subject to interpretation on its own, and since it was written before the existence of the Extended Link Opaque LSA, it is not self-evident how to interpret it with respect to using this new LSA.   Clarifying the proposed rules for use of the TE Opaque LSA or the Extended Link Opaque LSA without relying on interpretations of 3630 will be helpful.  It will help the WG evaluate the proposal overall and determine what, if any, backwards compatibility issues this proposal may cause with existing implementations.  It may also help future implementers avoid interoperability and backwards compatibility issues.

Sure, we can add more text in the draft.

What I have a problem with is that you are questioning RFC3630 and claim 
that it subject to interpretation. I don't think that is the case. 
RFC3630 defines TE Opaque LSAs as a mechanism to describe TE topology. 
There is nothing left that is subject to interpretation. If we do not 
have an agreement on this part, there is no text in the draft that is 
going to make things clear.


>
> As a concrete example, I think it would be useful to explicitly address the case of how to advertise a link that only supports LDP in the text of the draft.   Below is an example of a format that would clarify this.    From the response to my question below regarding LDP, I assume that a link that only supports LDP signaling and not RSVP-signaling would not be advertised in the TE Opaque LSA.  However, I am honestly not positive that this is what is intended.

LDP being enabled on link is orthogonal to link being part of the TE 
topology. Implementation can choose what makes link part of TE topology, 
some may have explicit command on a per interface basis, some may do it 
based on some other feature being enabled on the link. No matter what 
mechanism they use it is a local decision on the box that makes the link 
part of TE topology and triggers the advertisement of the TE Opaque LSA 
for such link. What is important is that as soon as the TE Opaque LSA is 
flooded all other nodes in the area will assume they can use such link 
for traffic engineering and use it during the TE cSPF.

>
> Format of proposed clarifying text:
> ------------------
>
> A link MUST NOT be advertised in the TE Opaque LSA under the following conditions:
>
> 1) The link does not support RSVP-TE signaling.

one can do traffic engineering without RSVP. RFC3630 does not mandate 
the usage of RSVP.

>
> 2) Another condition...
>
> A link MAY be advertised in the TE Opaque LSA under the following conditions:

the only condition is that user decided to make the link part of the 
traffic engineering topology, regardless of what is the local mechanism 
on the box to make that happen.

>
> 1) Another condition ...
>
> A link MUST NOT be advertised in the Extended Link Opaque LSA under the following conditions:

I do not see any reason to specify why a link MUST NOT be advertised in 
Extended Link Opaque LSA.

TE Opaque LSA and Extended Link Opaque LSA are orthogonal to each other. 
TE Opaque LSA is dedicated for traffic engineering and must not be used 
for anything else.

thanks,
Peter

>
> 1) Some other condition ....
>
> Thanks,
> Chris
>
>
>
> -----Original Message-----
> From: Peter Psenak [mailto:[e-mail]]
> Sent: Wednesday, October 21, 2015 3:24 PM
> To: Chris Bowers <[e-mail]>; Acee Lindem (acee) <[e-mail]>; Shraddha Hegde <[e-mail]>; OSPF WG List <[e-mail]>
> Subject: Re: [OSPF] Regarding draft-ppsenak-ospf-te-link-attr-reuse-00
>
> Hi Chris,
>
> On 10/21/15 21:44 , Chris Bowers wrote:
>> Peter,
>>
>> RFC3630 does not appear to restrict the use of the attributes it defines.   The term "TE extensions" may seem to imply some restriction, but the Applicability section of RFC3630 explicitly addresses this potential interpretation by saying that a more accurate designation is "extended link attributes".
>>
>> 1.1.  Applicability
>>
>>      Many of the extensions specified in this document are in response to
>>      the requirements stated in [5], and thus are referred to as "traffic
>>      engineering extensions", and are also commonly associated with MPLS
>>      Traffic Engineering.  A more accurate (albeit bland) designation is
>>      "extended link attributes", as the proposal is to simply add more
>>      attributes to links in OSPF advertisements.
>
> RFC3630 says:
>
>      The extensions provide a way of describing the traffic engineering
>      topology (including bandwidth and administrative constraints) and
>      distributing this information within a given OSPF area.  This
>      topology does not necessarily match the regular routed topology,
>
> above clearly indicates that if the link is advertised in TE Opaque LSA, it is part of the TE topology, otherwise it is not. That restricts the usage of the TE Opaque LSA to the links that are part of the TE topology.
>
>>
>> -------
>> Also, the response below uses the term "TE-enabled" which along with "TE-application" does not appear to have a precise definition in draft-ppsenak-ospf-te-link-attr-reuse-00.   Based on RFC 3630, it seems reasonable to say that a link is "TE-enabled" if the link is advertised in the TE Opaque LSA.  I don't think this is the meaning you intend, so to avoid confusion, I will use the term "RFC-3630-TE-enabled" to mean that the link is advertised in the TE Opaque LSA defined in RFC 3630.
>>
>> So can you clarify what "TE-enabled" or a "TE-application" means in your document?  I assume that it should mean that MPLS is enabled, but it is actually not clear to me if just having LDP-enabled on a link would qualify as being "TE-enabled" or not.
>
> TE-enabled means the link is part of the traffic engineering topology as
> described by RFC3630.
>
> thanks,
> Peter
>
>>
>> Thanks,
>> Chris
>>
>>
>> -----Original Message-----
>> From: Peter Psenak [mailto:[e-mail]]
>> Sent: Wednesday, October 21, 2015 12:40 PM
>> To: Chris Bowers <[e-mail]>; Acee Lindem (acee) <[e-mail]>; Shraddha Hegde <[e-mail]>; OSPF WG List <[e-mail]>
>> Subject: Re: [OSPF] Regarding draft-ppsenak-ospf-te-link-attr-reuse-00
>>
>> Hi Chris,
>>
>> On 10/21/15 19:20 , Chris Bowers wrote:
>>> In my opinion the backwards compatibility problems introduced by this
>>> proposal outweigh potential gains.
>>
>> there is no backwards compatibility problem with the draft.
>>
>>>
>>> As a concrete example, there is at least one existing implementation
>>> of remote LFA where policy is used to select a backup tunnel that does
>>> not share an SRLG with the failed link.  This SRLG information is
>>> carried in the TE Opaque LSA.
>>
>> that is fine, you are free to do that if the link is TE enabled, there is no problem. If the link is not TE enabled and you use TE Opaque LSA to flood the SRLG data for such link, you are going against the current specification. There is no way to do that today, because any router that would receive such TE Opaque LSA must assume such link is TE enabled.
>>
>>>
>>> As it currently reads, I think the proposal in
>>>     draft-ppsenak-ospf-te-link-attr-reuse has the potential to break
>>> existing standards-compliant implementations.
>>
>> I don't believe so.
>>
>>>
>>> I might be OK with having the proposal only apply to sub-TLVs  that
>>> get defined in the future.  However, I think that taking TLVs that were
>>>     standardized over ten years ago, and selectively moving them or
>>> copying them to a different LSA based on a set of rules that is
>>> subject to interpretation is going to create confusion and
>>> interoperability headaches.
>>
>> What we propose is the way to advertise link attributes without making the link part of TE topology. We simply do not have a way to do that today. I do not see any problem in doing so, because we do not change anything on the TE Opaque LSA side, we are defining something new.
>>
>> thanks,
>> Peter
>>
>>>
>>> Chris
>>>
>>> *From:*OSPF [mailto:[e-mail]] *On Behalf Of *Acee Lindem
>>> (acee)
>>> *Sent:* Wednesday, October 21, 2015 6:48 AM
>>> *To:* Shraddha Hegde <[e-mail]>; OSPF WG List
>>> <[e-mail]>
>>> *Subject:* Re: [OSPF] Regarding
>>> draft-ppsenak-ospf-te-link-attr-reuse-00
>>>
>>> Hi Shraddha,
>>>
>>> *From: *OSPF <[e-mail] <mailto:[e-mail]>> on
>>> behalf of Shraddha Hegde <[e-mail]
>>> <mailto:[e-mail]>>
>>> *Date: *Wednesday, October 21, 2015 at 1:20 AM
>>> *To: *OSPF WG List <[e-mail] <mailto:[e-mail]>>
>>> *Subject: *[OSPF] Regarding draft-ppsenak-ospf-te-link-attr-reuse-00
>>>
>>>       Hi All,
>>>
>>>       draft-ppsenak-ospf-te-link-attr-reuse-00 proposes moving and/or
>>>       copying TLVs from the TE Opaque LSA to the Extended Link Opaque LSA.
>>>       The draft lists the problems that the draft is trying to solve.  I
>>>       have reproduced that list of problems below, with each problem
>>>       followed by what I believe to be a better and simpler solution.
>>>
>>>           1.  Whenever the link is advertised in a TE Opaque LSA, the
>>> link
>>>
>>>               becomes a part of the TE topology, which may not match IP
>>> routed
>>>
>>>               topology.  By making the link part of the TE topology,
>>> remote
>>>
>>>               nodes may mistakenly believe that the link is available
>>> for MPLS
>>>
>>>               TE or GMPLS, when, in fact, MPLS is not enabled on the link.
>>>
>>>       To address this issue, we simply need to define a new sub-TLV in the
>>>       TE Link LSAto say whether MPLS/GMPLS/RSVP is enabled on the link
>>>       instead of moving the TLVs around into different LSAs.
>>>
>>>           2.  The TE Opaque LSA carries link attributes that are not
>>> used or
>>>
>>>               required by MPLS TE or GMPLS.  There is no mechanism in TE
>>>       Opaque
>>>
>>>               LSA to indicate which of the link attributes should be
>>> passed to
>>>
>>>               MPLS TE application and which should be used by OSPFv2 and
>>> other
>>>
>>>               applications.
>>>
>>>       OSPF database is a container and OSPF can use any of the LSAS for
>>>       its own use including the TE LSAs.As far as the TE database goes, it
>>>       contains data from TE LSAs as well as non-TE LSAs (Network LSA)
>>>       today so thereasoning described here doesn't make sense.
>>>
>>>           3.  Link attributes used for non-TE purposes is partitioned
>>> across
>>>
>>>               multiple LSAs - the TE Opaque LSA and the Extended Link
>>> Opaque
>>>
>>>               LSA.  This partitioning will require implementations to
>>> lookup
>>>
>>>               multiple LSAs to extract link attributes for a single
>>> link,
>>>
>>>               bringing needless complexity to the OSPFv2 implementations.
>>>
>>>       There will be nodes in the network which will run older software
>>>       which send these attributes via TE LSAs so the problem of looking
>>>       into the TE LSAs for TE relatedinformation doesn't get solved with
>>>       this draft.  Rather it makes it more complicated. With this draft,
>>>       the multiple LSA lookup will only increase.An implementation will
>>>       first have to find if Extended link LSA contains the required info,
>>>       if not it will need to lookup the info in TE.LSA.
>>>
>>> The applications using the TE parameters for non-TE use-cases will use
>>> the OSPF Prefix/Link attributes for these use cases. Hence, there is
>>> no requirement to lookup the LSAs in multiple places. Backward
>>> compatibility will be covered in the specifications of these applications.
>>>
>>> Thanks,
>>>
>>> Acee
>>>
>>>       Looking up multiple LSAs for information is an implementation issue
>>>       and I am sure there will be implementations that will handle this
>>>       gracefully so that it doesn't cause
>>>
>>>       delays in critical paths. It doesn't seem reasonable to come up with
>>>       protocol extensions to solve implementation issues.
>>>
>>>       Rgds
>>>
>>>       Shraddha
>>>
>>>
>>>
>>> _______________________________________________
>>> OSPF mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/ospf
>>>
>>
>> .
>>
>
> .
>