Re: [OSPF] FW: New Version Notification for draft-ietf-ospf-prefix-link-attr-00.txt

[Peter Psenak <[e-mail]>]

Hi Tony,

please see inline:

On 9/4/14 21:02 , A. Przygienda wrote:
> On 09/04/2014 12:24 AM, Peter Psenak wrote:
>> Hi Tony,
>>
>> please see inline:
>
> Hey Peter, same
>
>>
>> On 9/3/14 18:13 , A. Przygienda wrote:
>>>
>>>
>>> It's also wise to add 'if the same extended prefix TLV (i.e. for same
>>> prefix) is seen twice in same opaque LSA only use the first and force
>>> people to put all sub-tlvs into a single tlv.
>>
>> it's kind of obvious, but we can add a text to be sure.
>
> As I see it, the purpose of an RFC is to assure interoperability. If
> there are two
> interpretations available of the same packet that will make two
> solutions not interoperable,
> the standard has to mandate the correct one.

ok, we will add a text to make it clear.

>
>>
>>>
>>> "OSPFv2 requires functional extension beyond what can readily be done
>>> with the fixed-format Link State Advertisements (LSAs) as described
>>> in RFC 2328 [OSPFV2]. This document defines OSPF opaque LSAs based
>>> on Type-Length-Value (TLV) tuples that can be used to associate
>>> additional attributes with advertised prefixes or links."
>>>
>>> That's strongly suggestive of "you advertise normal LSA and then you use
>>> opaque to 'extend' it" . Extended extends and is not independent from
>>> original stuff in normal use of English idiom.
>>
>> because existing LSA are not extensible, any additional link/prefix
>> attributes must be advertised in a separate LSAs. We called it
>> "extended", but it does not mean that existence of the extended LSA
>> require existence of the base LSA. I had a text in the original SR
>> draft around that, we can add it here too.
>
> yepp, once it spelled out, it's obvious but the first conclusion one
> jumps to is that they are inter-dependent otherwise.

ok, we will add a text to clarify.

>
>>
>>
>>> .,..
>>> the area that do not support opaques, an additional text is needed
>>> saying  "never compute through routers without opaques using information
>>> in opaques that influences reachability or metrics". Actually, it's
>>> worse since if a router without opaques forwards through a router
>>> advertising opaques influencing metrics (for sake or simplicity), you
>>> cannot start using opaques to keep on forwarding.  In case that all
>>> doesn't parse, I can do a simple looping example.
>>
>> we are not advertising any metrics in the extended LSAs, so I do not
>> see why we would have to deal with that case here. If one day someone
>> wants to do that, then it would have to be covered in a separate draft
>> and we would deal with it at that time.
>
> agreed here. you're correct. That would be overspecification in this draft
>
>>
>>> OK, I still don't understand 'WHICH flooding scope', the  type-3 vs.
>>> type-5 one or the 'opaque flodding scope one'.  I think you mean 'all
>>> prefixes in the same opaque must have the same OSPF-route-type
>>> [0+unevens]'  but the text is really hard to parse here.
>>>
>>> Acee, you mean remove the restriction ?  I see how some implementations
>>> may benefit from cramming all type-3 into the same opaque but I would
>>> also suggest to not restrict this and remove the text.
>>
>> the restriction can not be removed. All the Extended Prefix TLVs that
>> you put in a singe Extended Prefix LSA will only use a flooding scope
>> of the Extended LSA itself.
>
> I still think it's not clear what the original paragraph means.

prefixes can have different flooding scope. Today intra-area and 
inter-area prefixes have area flooding scope and external prefixes have 
domain wide flooding scope. You can not mix prefixes of different 
flooding scope in a single Extended Prefix LSA, because Extended Prefix 
LSA can only have a single flooding scope.

>
>>
>>> yes, and again, I think 'extended' is an ill chosen name maybe if that's
>>> the intention of the draft.
>>>
>>> BTW, what is the plan if all the TLVs blow out the MTU of the opaque ?
>>
>> you can generate many Extended Prefix or Extended Link LSAs, so there
>> is no issue.
>
> ack
>
>>
>>> I'm missing something ?   RFC5250 is not describing how the instance
>>> (they call it Opaque ID) is supposed to be generated (it references
>>> appendix A referencing section 7 which has nothing in it about this).
>>> I'm divining the instance allows to have multiple opaques of the same
>>> type which leads to bunch interesting discussions
>>>        * what happens when the MTU gets blown & TLVs are shifted into
>>> next 'fragment' in ISIS speak [with all the restrictions how the TLVs
>>> can migrate between fragments, ISIS guys suffered this one [as did PNNI]
>>> @ length & will surely lend a helping hand in such a case].
>>
>> TLVs can come and go and move from one Extended Opaque LSA to the
>> other. Implementation needs to track these changes and deal with them.
>
> I would let the ISIS guys chime in here. They have tons experience with
> TLVs sliding fragments and what kind of implementation/operation pain
> that can cause.
>
>>
>>
>>>        * what happens if the same extended prefix TLV shows up twice in
>>> two different opaques of same type ?  You use the one that is in lower
>>> opaque ID ?
>>
>> I would call it a bug on the originator side.
>
> yes, but still, the draft should mandate what the desired behavior is @
> this point in time otherwise some will use the lower ID, some higher and
> some none of the above & the implementations will not interop ?

ok, we will add a tiebreaker there.

thanks,
Peter
>
> --- tony
>
> _______________________________________________
> OSPF mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/ospf
> .
>