Re: [mpls] Concerns about ISD

[Robert Raszuk <[e-mail]>]

Hi Stewart,

Yes - exactly. And I fail to see a reason by MNA/NA/MIAD can not be an
overlay too from transport point of view.

Cheers,
R.



On Sat, 16 Apr 2022 at 17:47, Stewart Bryant <[e-mail]>
wrote:

>
>
> On 16 Apr 2022, at 14:38, Robert Raszuk <[e-mail]> wrote:
>
> Hi Tianran,
>
> I do not think DETNET(including bounded delay) need neither. There is
>> already specifications for MPLS.
>>
>> We are not going to review them all here.
>>
>
> That is a very interesting and important point. I just looked at RFC8964
> and surprise ... no need to change MPLS architecture. Just use a service
> label concept and carry service info outside of the data plane.
>
> What I called Forwarding Behaviour ID they call service label (analogy to
> other types of service labels already carried in MPLS). The
> difference between RFC8964 and FBI approach seems to be about types of
> services (aka actions) to execute on the packets. Also FBI concept works
> with domain wide labels while service labels are pretty much local to a
> given node.
>
> So a bit curious what the above architectures are missing which would
> really justify to define, develop and deploy a new MPLS data plane ?
>
> Kind regards,
> Robert
>
> PS. For those who are claiming that use a post stack data PSD should be
> the only option let me observe that if this solution is supposed to work
> with SR-MPLS the label stack of pure transport may be pretty large. So
> number of nodes may not be capable to fetch PSD (we invented EL just
> because transit LSRs were not even able to read IP header). That to me is a
> practical blocker for this approach if MNA will continue.
>
>
> Detnet is really an overlay system.
>
> The delivery label takes the packet to the next hop in the Detnet layer.
> This is popped and the service label exposed. The service label is then
> processed. Then if the packet is to continue a new service label is pushed
> followed by a new delivery label.
>
> In a PHPed network where every node is a service label the service label
> is always ToS. However the operation in not really swap as in MSPW, the
> service label and the CW are examined and processed.
>
> Note that Detnet does not support many of the features being asked for in
> NA/MIAD.
>
> - Stewart
>
>