Jie,
see below.

On 02/25/2015 10:00 PM, Dongjie (Jimmy) wrote:
> Hi Elwyn,
> 
> Thanks for your prompt response. Attached is a new revision which reflects 
> your comments and suggestions, please let me know if this revision is OK to 
> progress, thanks.
> 
> Hi Lou,
> 
> Thanks for your suggestion.
> 
> In this new revision the text about valid ERO subobject types is revised a 
> bit, as I noticed the type value of "label" (3) is also less than 32, while 
> it may not be used to identify a loopback target.
> 
Why not?  This is a technology specific issue.  For example, one could
loop a specific lambda in certain optical switches by identifying a
specific lambda via a label.

> Currently the IANA allocation rules for 5-31 is not changed by this revision. 
> If you think this change is needed, I can update the draft before submission. 
> Thanks.
> 

I think the previous text should be used.

Thanks,
Lou

> Cheers,
> Jie
> 
>> -----Original Message-----
>> From: Elwyn Davies [mailto:[email protected]]
>> Sent: Thursday, February 26, 2015 2:14 AM
>> To: Dongjie (Jimmy); General area reviewing team
>> Cc: [email protected]
>> Subject: Re: Gen-art LC review of draft-ietf-teas-rsvp-te-li-lb-03
>>
>> Hi, Jie.
>>
>> I think we are almost done.
>>
>> There is a proposal for the specific types at the end.
>>
>> Cheers,
>> Elwyn
>>
>>
>>
>> On 25/02/2015 06:10, Dongjie (Jimmy) wrote:
>>> Hi Elwyn,
>>>
>>> Thanks a lot for your response. Please see my replies inline:
>>>
>>> Best regards,
>>> Jie
>>>
>>>>>> s3.2, para 2:
>>>>>> OLD:
>>>>>>       The ingress MUST ensure that the desired loopback mode is
>>>>>> strictly identified in the ERO.
>>>>>>
>>>>>> Am I right that "mode" is a typo?  Should it be "node" rather than
>>>>>> "mode? I couldn't see anything about "loopback modes" in RFC 5860
>>>>>> or RFC 6371. After much head scratching I came to the conclusion that
>> "node"
>>>>>> was the right answer...  If not then please explain where modes are
>> defined.
>>>>> I guess we may not call it a typo. The above sentence was added
>>>>> during the
>>>> WG review, the intention was to strictly identify the entity at which
>>>> the loopback is desired, which can be either a node or an interface of a
>> node.
>>>>> If we replace "mode" with "entity", do you think it can be clearer?
>>>>>
>>>>>> Continuing on the basis that the word s/b "node"...   I think this would
>>>>>> be clearer as follows:
>>>>>> NEW
>>>>>> The ingress node MUST ensure that the node where loopback is
>>>>>> intended to occur is marked as a strict hop (i.e., not a loose hop) in 
>>>>>> the
>> ERO.
>>>>> If you agree with the above change, in your suggestion we can also
>>>>> substitute
>>>> "node" with "entity".
>>>> That would help.  How about:
>>>> NEW (mk 2):
>>>>
>>>> The ingress node MUST ensure that the entity (node or interface), at
>>>> which loopback is intended to occur, is marked as a strict hop (i.e.,
>>>> not a loose hop) in the ERO type subobject.
>>> Agree with this description. Will update in the new revision.
>> Fine.
>>>
>>>>> The prior subobject of "ERO Hop Attributes subobject" would also be
>>>>> ERO
>>>> subobject which identifies the target entity of loopback. As
>>>> specified in the sentence below, the type value of the prior
>>>> subobject ensures the prior subobject would be an IP prefix or an
>>>> interface ID, which can identify the node or interface at which
>>>> loopback is desired. Note that the current description requires both
>>>> the "ERO Hop Attributes subobject" and the prior subobject have the L
>>>> bit set to 0 to ensure that the two subobjects together strictly specify a
>> loopback request on an specific entity (node or interface).
>>>>> We can rephrase to make it clearer.
>>>> I have looked into this issue a bit more deeply, and realize that my
>>>> problem is not really with this draft but, at least partly, with
>>>> draft-ietf-teas-lsp-attribute-ro-01.  Having gone back to RFC3209, I
>>>> understand that the subobjects in an ERO as defined there are all
>>>> specifiers for type of 'entity' that could be waypoints on the
>>>> explicit route.  There is no allowance for other types of subobject
>>>> (such as the attribute subobjects being introduced by
>>>> draft-ietf-teas-lsp-attribute-ro-01).  As introduced in RFC 3209,
>>>> both EROs and RROs contained only entity type subobjects, but RFC
>>>> 5420 changed this for *RROs only* by introducing RRO Attributes
>>>> subobjects which could be inserted between the type subobjects. Thus
>>>> as standardized currently, EROs aren't allowed any other sort of subobject
>> and there is no equivalent to the "Presence Rules" defined for RROs in RFC 
>> 5420
>> section 7.3.1.
>>>>
>>>> draft-ietf-teas-lsp-attribute-ro-01 updates the presence rules for
>>>> RROs to allow for the RRO Hop Attributes subobjects but the necessary
>>>> presence rules that would control where the ERO HOP Attributes could
>>>> be inserted are not properly defined in
>>>> draft-ietf-teas-lsp-attribute-ro-01.  Para 1 of s2.3 of
>>>> draft-ietf-teas-lsp-attribute-ro-01 seems to indicate that the HOP
>>>> Attributes would be placed after all the type subobjects which
>>>> doesn't seem right.  I would have thought that the hop attributes
>>>> needed to be interleaved between the type subobjects - and I think
>>>> that is what the existing text in the lock draft is trying to say.
>>>>
>>>> So I think the draft-ietf-teas-lsp-attribute-ro-01 needs to be
>>>> clarified
>>>> - a specific section on Presence Rules as in RFC 5420 would help.
>>>>
>>>> The text about checking that the L bit in the lock draft is
>>>> confusing, because the L bit is *required* by
>>>> draft-ietf-teas-lsp-attribute-ro-01
>>>> to be 0 in *all* Hop Attributes subobjects whether they are
>>>> associated with loose or strict hops and need not be mentioned.  The
>>>> L bit is only set in the type subobject if it is a loose hop so the check 
>>>> should
>> be
>>>> that the type subobject has its L bit clear.    It should also be made
>>>> clearer in the lock draft that there can be multiple Hop Attribute
>>>> subobjects after the type subobject.
>>>>
>>>> I hope this helps.
>>> You are correct, the L bit in the ERO hop attributes subobject must be 0, 
>>> thus
>> we would only specify that the L bit in the prior ERO subobject MUST be set 
>> to 0
>> in the new revision.
>> OK.
>>>
>>> As for the rules about multiple Hop Attribute subobjects in ERO, IMO maybe 
>>> it
>> is more suitable to be specified in draft-ietf-teas-lsp-attribute-ro, but we 
>> can
>> also add such descriptions if you think they belongs to this draft.
>> I will negotiate with the authors of draft-ietf-teas-lsp-attribute-ro to see 
>> if we
>> can agree a revision.
>>>
>>>>>> s3.2, para 3:
>>>>>>> Currently, the type value MUST be verified to be
>>>>>>>      less than 32, and for type values 1 and 2 the prefix length MUST be
>>>>>>>      32 and 128 respectively.
>>>>>> It would be better to use the names of the types (1 = IPv4, 2 =
>>>>>> IPv6) that would make it clearer why the lengths are as they are 
>>>>>> specified.
>>>>>> Is it possible to explain why the type has to be less than this
>>>>>> apparently arbitrary limit of 32?  And hence what allocation
>>>>>> constraints ought to be applied in the future - this might need a
>>>>>> change in the IANA allocation rules that should be specified in this
>> document.
>>>>> The limit of 32 was to ensure the subobject type can identify a
>>>>> specific node
>>>> or interface. Maybe we can specify the required types (node,
>>>> interface) explicitly to make it clear.
>>>> That would help.
>>>> It has just occurred to me why you want to say that the prefix
>>>> lengths are
>>>> 32/128 bits - so that the IP address refers to exactly one node.
>>>> However this is possibly confusing because RFC 3209 *requires* the
>>>> values to be 32/128 respectively and the subobjects do carry just a
>>>> host address and not a domain prefix as might be implied by lower values.
>>> As specified in 4.3.3 of RFC 3209, for the ERO subobjects, the prefix 
>>> lengths
>> are not required to be just 32/128. While as specified in 4.4.1 of RFC 3209, 
>> for
>> the RRO subobjects, the prefix lengths are required to be 32/128.
>> You are right:  I missed that RRO and ERO were different.  So specifying the
>> prefix lengths is indeed needed as you say.
>>>
>>>> I noted having looked again at the registry [1], and it appears that
>>>> nobody has yet defined an ERO type for 4 octet AS numbers which have
>>>> now
>>>> (finally) been standardized.  This might mean that 32 was not the
>>>> right boundary.  Perhaps this means that you need to define an update
>>>> to the allocation rules for the registry so that values below (say)
>>>> 32 are required to identify exactly one physical node and not an
>>>> abstract node with potentially many physical nodes.
>>> Thanks for your suggestion. Since this draft just want to restrict the type 
>>> of
>> ERO subobjects that can appear prior to the ERO hop attributes subobject,
>> making such update to IANA was not our original intention, also requires
>> discussion with ADs and TEAS chairs.
>>>
>>> What about we specify the allowed ERO subobject types explicitly, similar to
>> 6.1.1 of RFC 4990? The specification could be something like:
>>>
>>> The prior subobject MUST be able to identify a specific node or interface,
>> specifically it MUST be one of the following: type "IPv4 prefix" with prefix
>> length 32, type "IPv6 prefix" with prefix length 128, or type "Unnumbered
>> Interface ID".
>> That would certainly cover the situation as it is at the moment.  I can't
>> immediately think of any new types that might be added and would be relevant
>> as loopback points, but you might have some ideas on this.
>> I was trying to think of an easy way to factor this into the IANA 
>> registration
>> process but I can't see anything very simple. I think a few 'weasel words' 
>> will do
>> for an unlikely event.  How about:
>> NEW mk 3:
>> The prior subobject MUST be able to identify a specific node or interface,
>> specifically, at the time of writing, it can be one of the
>> following:
>>    - Type "IPv4 prefix" with prefix length 32,
>>    - Type "IPv6 prefix" with prefix length 128, or
>>    - Type "Unnumbered Interface ID".
>> If any new subobject types are defined in future and they identify a specific
>> node or interface, the specification of the new type can extend this list.
>>
>>
>>>> [1]
>>>> http://www.iana.org/assignments/rsvp-parameters/rsvp-parameters.xml#r
>>>> svp-
>>>> parameters-25
> 

_______________________________________________
Gen-art mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/gen-art

Reply via email to