Re: [Pals] [mpls] FW: New Version Notification for draft-jags-mpls-ps-mna-hdr-00.txt

[Greg Mirsky <[e-mail]>]

Hi Tony,
thank you for your response. Please find my notes below, tagged GIM2>>.

Regards,
Greg

On Tue, Apr 25, 2023 at 12:34 PM Tony Li <[e-mail]> wrote:

>
> Hi Greg,
>
> Sorry for the delay.
>
>
> If a transit node needs to find PSD, it can scan the entire label stack
>> for the bottom of stack bit.  I can’t think of a good reason to ever do
>> this, given that it’s a transit node and should be payload indifferent.
>>
> GIM>> I got confused. I am too not seeing the case when an LSR starts
> looking for PSD unless it has found a proper NAS.
>
>
>
> I’m told that some legacy transit nodes choose to look at the payload to
> extract ECMP.  A legacy implementation that does this and finds any MNA PSD
> is going to be mightily confused and would never see MNA ISD.  An MNA
> specific nibble here would be helpful in ensuring that it realizes that
> it’s about to make a mistake.
>
GIM2>> The PALS (PWE3 at that time) WG was faced with the result of
allowing PW CW to be optional in Ethernet PW in the scenario you've
described.  Because of the allocation of MACs with the first nibble 0x4 or
0x6, the payload was misinterpreted by a transit node as IPv4 or IPv6. And
thus, the WG published RFC 8469
<https://www.rfc-editor.org/rfc/rfc8469.html>. But I only know of cases
when a transit node mistakes the payload for IPv4 or IPv6. In principle, a
transit node must not draw a conclusion about the payload type based on the
value of the first nibble after the BoS LSE. If we agree on that, then
using any value other than 0x4 and 0x6, e.g., 0x0, should be satisfactory.
WDYT?

>
>
> > - we will only include the MNA first nibble if there is an MPLS Label in
>> the packet. right? Existing (legacy) non-MMA capable implementations will
>> not recognize the MNA first nibble as "MNA follows, only understand that
>> this is something unrecognised. What does legacy implementations do with
>> unrecognized first nibbles today?
>>
>>
>> There is not much point in having an MNA first nibble unless there is
>> also MNA PSD.  And there is not much point in MNA PSD unless there is a
>> label stack involved. Therefore, yes, there should be a label in the label
>> stack.
>>
> GIM>> More confusion in my head. Why it must be "an MNA first nibble" but
> not any non-IPv4 and non-IPv6?
>
>
>
> Just for clear semantics. I am a very big fan of avoiding overloading.
>
GIM2>> Considering the limited number of unused values, re-using what we
can seems like a reasonable trade-off.

>
> Tony
>
>
>