Re: WGLC for BFD Multipoint documents (ending July 14, 2017)

[Greg Mirsky <[e-mail]>]

Hi Santosh, et. al,
would like us to look closer into the statement in Goals section, the one
which stresses non-goal of the BFD for multipoint networks protocol:

>
>>    - Goals
>>       - the last statement of non-goal seems little unclear. In fact, if
>>       there's only one tail, the BFD for multipoint network does verify
>>       connectivity, though unidirectional, between the head and the tail. I think
>>       that the intention is to stress that p2p bi-directional connectivity
>>       verification is not supported by this document.
>>
>> [SPK] It only says that this document does not support verification of
> unicast path between head and tail. I can clarify a bit on this. Please let
> me know if you have a suggestion for this.
>
GIM>> I'd suggest to use unicast in place of point-to-point. Using my
earlier example, in case when there's only one tail multipoint becomes
point-to-point.

I think we've missed MPLS network scenario. What would be the difference
between p2p LSP and p2mp LSP? Both use labels though with different
context. So the only difference is that BFD for multipoint does not monitor
bi-directional continuity, i.e. viability of a path, between root and the
leaf. I believe such statement already exists in the document and doesn't
seem there's need to repeat it again. Perhaps we can remove that last
statement altogether?

Regards,
Greg

On Fri, Jul 7, 2017 at 6:16 PM, Greg Mirsky <[e-mail]> wrote:

> Hi Santosh, et. al,
> another note, question on IP/UDP encapsulation of BFD for multipoint
> networks. The document says nothing about values that may be used for
> Source UDP port number. Even though MultipointHead will not receive BFD
> packets from MultipointTail on the UDP port listed, should we recommend to
> use numbers from dynamic range, i.e. 49152 to 65535? I think that the
> multipoint document should state, as in RFC 5881:
>
> The source port MUST be in the range 49152 through 65535.
>
> What do you think?
>
> Regards,
> Greg
>
> On Thu, Jul 6, 2017 at 11:00 AM, Greg Mirsky <[e-mail]>
> wrote:
>
>> Hi Santosh,
>> many thanks for your thoughtful consideration of my comments. Please find
>> my answers and couple more notes in-line and tagged GIM>>.
>>
>> Regards,
>> Greg
>>
>> On Thu, Jul 6, 2017 at 10:27 AM, Santosh P K <
>> [e-mail]> wrote:
>>
>>> Hello Greg,
>>>    Thanks for your comments. Please see my reply inline tagged[ SPK].
>>>
>>> Thanks
>>> Santosh P K
>>>
>>> On Tue, Jul 4, 2017 at 1:02 AM, Greg Mirsky <[e-mail]>
>>> wrote:
>>>
>>>> Dear Authors, WG chairs, et. al,
>>>> please kindly consider my comments to the latest version of the draft
>>>> as part of WGLC:
>>>>
>>>>    - Very general
>>>>       - I suggest to add note clarifying that all terms that include
>>>>       "connectivity" in the document are being used as alternative to
>>>>       "continuity", i.e. existence of a path between the sender and the receiver,
>>>>       and should not be interpreted as "connectivity verification in terms of
>>>>       transport network".
>>>>    - Introduction
>>>>       - I find that characterization of BFD and unidirectional
>>>>       continuity verification as "natural fit" bit of stretching. Perhaps
>>>>       "capable of supporting this use case" is acceptable;
>>>>
>>>> [SPK] Will take care.
>>>
>> GIM>> Thanks
>>
>>>
>>>>    - Goals
>>>>       - the last statement of non-goal seems little unclear. In fact,
>>>>       if there's only one tail, the BFD for multipoint network does verify
>>>>       connectivity, though unidirectional, between the head and the tail. I think
>>>>       that the intention is to stress that p2p bi-directional connectivity
>>>>       verification is not supported by this document.
>>>>
>>>> [SPK] It only says that this document does not support verification of
>>> unicast path between head and tail. I can clarify a bit on this. Please let
>>> me know if you have a suggestion for this.
>>>
>> GIM>> I'd suggest to use unicast in place of point-to-point. Using my
>> earlier example, in case when there's only one tail multipoint becomes
>> point-to-point.
>>
>>>
>>>
>>>>
>>>>    - Section 4.7
>>>>       - the last paragraph notes that the discriminator value MUST NOT
>>>>       be changed. Since Your Discriminator MUST be set to 0 this refers to My
>>>>       Discriminator only. I think that explicit reference will make the statement
>>>>       more clear. Thus suggest s/the discriminator values/the My Discriminator
>>>>       value/
>>>>
>>>> [SPK] Will take care of this.
>>>
>> GIM>> Thanks.
>>
>>>
>>>>    - Section 4.8
>>>>       - I believe that requiring use of IP/UDP encapsulation for BFD
>>>>       in multipoint networks over MPSL LSP is too restrictive. I suggest changing
>>>>       text as following:
>>>>
>>>> OLD:
>>>>
>>>> For multipoint LSP, MultipointTail MUST use destination UDP port "3784" and IP "127.0.0.0/8" range.
>>>>
>>>>
>>>> NEW
>>>>
>>>> If IP/UDP encapsulation used by MultipointHead for multipoint LSP, MultipointTail MUST use IP/UDP encapsulation of BFD destination UDP port "3784" and IP "127.0.0.0/8" range.
>>>>
>>>> Use of other types of encapsulation for multipoint LSP is outside the scope of this document.
>>>>
>>>>
>>> [SPK] Thanks. I think this make sense for non MPLS tunnels.
>>>
>> GIM>> Thanks. As I've looked at the text, I've realized that it misses
>> IPv6 case. Please consider the following as my new proposed change (not
>> sure but I think that quote marks are not required):
>> NEW
>> If IP/UDP encapsulation used by MultipointHead for multipoint LSP,
>> MultipointTail MUST use IP/UDP encapsulation of BFD destination UDP port
>> 3784, and the 127/8 range for IPv4, and the 0:0:0:0:0:FFFF:7F00/104
>> range for IPv6.
>> Use of other types of encapsulation for multipoint LSP is outside the
>> scope of this document.
>>
>>
>>>>    - Section 4.10
>>>>       - I cannot say what bfd.DetectMult packet is. It has not been defined in RFC 5880, nor in this document. What is the scenario described in the second paragraph? Is it when MultipointHead reduces Desired Min TX  Interval thus making defect detection more aggressive?
>>>>    -
>>>>
>>>> [SPK] This section talks about how to handle Poll sequence. In case of
>>> Multipoint head we cant afford to send POLL and expect all tail to reply
>>> with F bit set. Keeping track and building state at headend will be costly.
>>>
>>>
>> GIM>> Perhaps I wasn't clear in my question.  It was to the opening of
>> this sentence:
>>
>>    The MultipointHead MUST send bfd.DetectMult packets with P bit set at
>>    the old transmit interval before using the higher value in order to
>>    avoid false detection timeouts at the tails.
>>
>> I couldn't find reference to "bfd.DetectMult packet" in any document related to BFD.
>>
>>
>>
>>>>    - Section 7
>>>>       - I think it should be plural in the first paragraph, i.e. s/MultipointTail session/MultipointTail sessions/
>>>>       - I think that we can add another consideration to improve, strengthen security of BFD for multipoint network by suggesting that MultipointTail sessions created only for known combination of MultipointHead and My Discriminator. Such information MAY be learned from out-of-band and mechanisms are outside the scope of this document.
>>>>
>>>>
>>>> Regards,
>>>>
>>>> Greg
>>>>
>>>>
>>>>
>>>> On Mon, Jun 19, 2017 at 12:39 PM, Jeffrey Haas <[e-mail]> wrote:
>>>>
>>>>> Working Group,
>>>>>
>>>>> https://tools.ietf.org/html/draft-ietf-bfd-multipoint-10
>>>>> https://tools.ietf.org/html/draft-ietf-bfd-multipoint-active-tail-04
>>>>>
>>>>>
>>>>> The BFD Multipoint documents have been stable for some time.  Prior
>>>>> discussion at meetings has suggested we have an implementation for the
>>>>> main
>>>>> protocol component.  Also per prior discussions, we split the
>>>>> active-tail
>>>>> component of the original multipoint document to permit implementors
>>>>> to not
>>>>> have to worry about implementing active-tail procedures if they weren't
>>>>> interested in that feature.
>>>>>
>>>>> We are starting an extended last call on these documents.  The WGLC
>>>>> will
>>>>> conclude on July 14.  This provides ample time for list discussion.  If
>>>>> necessary, the IETF-99 meeting may provide for opportunities to close
>>>>> any
>>>>> contentious technical points.  (BFD is not currently scheduled to
>>>>> meet.)
>>>>>
>>>>> One item I would like to kick off is the document status of the
>>>>> active-tail
>>>>> mechanism.  At this time, no one has implemented it that I am aware of.
>>>>> Discussion with our AD suggests that publishing the document with
>>>>> Experimental status may be reasonable to preserve the work that went
>>>>> into
>>>>> the proposal.
>>>>>
>>>>> -- Jeff
>>>>>
>>>>>
>>>>
>>>
>>
>