Re: [mpls] The case for PSD in MPLS

[Rakesh Gandhi <[e-mail]>]

Thanks Stewart for summarizing the use-cases for PSD.
Agree with you that a PSD solution is needed in MPLS.

FYI:
The IOAM use-case with PSD is described in the following draft using the
MNA solution defined in [draft-jags-mpls-ps-mna-hdr] .

https://datatracker.ietf.org/doc/html/draft-gandhi-mpls-ioam-10

Thanks,
Rakesh


On Tue, Jul 18, 2023 at 4:54 AM Stewart Bryant <[e-mail]>
wrote:

> In recent discussions at the MNA Interims there have been two propositions
> that need deeper consideration by the WG.
>
> The first concerns the conflict between mutable data in the label field of
> an MNA ISD LDA and the ECMP behaviour of legacy routers. Whilst this could
> be addressed by requiring an ELI/EL pair there is concern that this
> increases the stack size of a packet carrying MNA by at least two LSEs, and
> stack size is likely to stress a number of platforms due to the presence of
> the MNA LSE set. This has led to the proposal that mutable data is only
> carried in the upper 11 bits of an MNA LSE since these bit are not normally
> used for ECMP calculation.
>
> The second assertion is that PSD is not needed as everything needed can be
> carried as ISD in the MHA header.
>
> I can think of at least three cases where PSD is needed given the
> constraints of ISD, particularly the recent proposition that variable data
> can only be carried outside the label in an LSE.
>
> 1) Time transfer over MPLS, where it is necessary to transmit the
> accumulation of the storage time as accurately as possible. Given the
> emerging major security concerns  regarding GNSS systems in an increasingly
> hostile world and further given the fundamental importance of time to many
> PHY layers, there is going to have to be work to provide a satisfactory
> alternate to GNSS, and the development of a wide area system of delivering
> precision time. If precision time is to be delivered over packet then there
> needs to be a more precise measurement and transmission of time of flight.
> It is not clear that this can be achieved in the right hand 11 bits of an
> LSE in any sensible way.
>
> 2) MPLS security is weak relying on the eggshell model. We know (Snowden)
> that state actors are capable of penetrating MPLS networks so we can
> foresee the need to provide a security shim inside MPLS, that would take a
> lot of 11 bit entries
>
> 3) We know that IOAM is trying to develop methods of minimising their
> in-packet overhead, but this is a world that is always pushing for more
> detailed information in the packet. It seems impractical to send this as a
> series of 11 bit data elements within a number of LSE entries.
>
> I am sure that if we think about this some more other use cases will
> emerge.
>
> I thus think that it would be unwise to preclude the use of PSD from our
> vision of MPLS.
>
> Best regards
>
> Stewart
>
> _______________________________________________
> mpls mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/mpls
>