Re: Next steps on advancing core IPv6 specifications to full Internet standard

[Hemant Singh <[e-mail]>]

Tom,

Please see in-line below against "<hs>".


On Mon, Nov 21, 2016 at 12:44 PM, Tom Herbert <[e-mail]> wrote:

Rather or not encapsulation is used, the SR EH would be an extension of the
> outermost IP header. Intermediate routers would not be concerned with
> rather the packet is encapsulated except when they are the terminal point
> of the encapsulation (need to decapsulate). So the SR EH is processed the
> same by intermediate nodes whether encapsulation is used or not.
>

<hs> Agree.  I thought intermediate nodes in SR could add a new SR info and
then another layer of encapsulation is used.   It turns out, this is not
the case.


> Likewise, Inband OAM has mixed router/switch vendor deployment.  One
>> vendor uses add/delete EH while another uses encap.  How are all
>> routers/switches instructed to use which method?
>>
>> This is inband OAM in an EH or within an encapsulation?
>


<hs> Inband OAM in an EH.  If encapsulation is used each node in the path
of the packet adds a new IPv6 header followed by a Hop-by-Hop (HBH) EH.
The EH includes OAM data.  Now if encapsulation is not used, each node adds
a its own EH - at least this is how the FD.io Inband was described to me.
Thus all nodes have to be instructed how to add the OAM data using
out-of-band mechanisms such as DHCPv6.


> Note that the suggestion to use encapsulation is not normative, it does
> not provide exactly the same behavior as inserting EH would (for instance
> ECMP handling would be very different). I think the interaction between EH
> and encapsulation needs to specified as part of the encapsulation, for
> instance the question of what EHs must be copied from the packet being
> encapsulated to the outer packet will need to be addressed.
>
> <hs>Hmm.  ECMP is used when multiple paths exists to a destination.   If
> each node in the packet path uses encapsulation, the new IPv6 header added
> by the node, still has the destination identical to the destination when
> the packet was sourced.  So why should ECMP be impacted whether
> encapsulation is used or not?
>

Thanks,

Hemant