Re: [mpls] NFFRR

[Kireeti Kompella <[e-mail]>]

Hi Michael,

Thanks for the link to the paper.  It is cool that both attack the problem
of loops due to multiple failures (real or perceived).

While the goals are similar, there are clearly differences in the approach
because of different assumptions.  ALDT attempts to minimize false
discards, at the expense of higher packet and processing overhead.  The
NFFRR approach is far more naive, but also far more efficient.  It would be
interesting to see how often NFFRR discards packets erroneously in
different topologies, as opposed to LDT or ALDT.  The trade-offs between
efficiency and correctness are harder to determine.

Sprinkled in the referenced paper in multiple places are limitations due to
OpenFlow.  P4 is a better base for prototyping.

One question: how is an LD label recognized as such?

Thanks too for the reference to the Topology Zoo.  The topologies I sampled
were pretty small, though.


On Thu, Sep 17, 2020 at 8:01 AM Michael Menth <[e-mail]>
wrote:

> Hi Kireeti, all,
>
> interesting to see this in MPLS. We've done a similar feature (Advanced
> Loop Detection and Termination, ALDT) for destination-based forwarding
> with LFAs and did some evaluations:
>
> D. Merling, W. Braun, and M. Menth: Efficient Data Plane Protection for
> SDN, in Proceedings of the 4th IEEE Conference on Network Softwarization
> (NetSoft), June 2018, Montreal, Canada.
> https://atlas.informatik.uni-tuebingen.de/~menth/papers/Menth18g.pdf
>
> We've implemented a prototype on the P4-based Tofino from barefoot, a
> paper is to come.
>
> Regards,
>
> Michael
>
> Am 17.09.2020 um 16:27 schrieb Kireeti Kompella:
> > Hi All,
> >
> > We have a prototype implementation of NFFRR in Junos.  The packet
> > forwarding code was easier to write using a well-known SPL rather than a
> > configured or signaled label.  This implementation is specifically for
> > the EVPN multihoming case -- to prevent loops if the CE is down rather
> > than a link to the CE, but the principle is the same for other NFFRR
> cases.
> >
> > Given that, and the fact that the problem is real, the solution
> > implemented and the draft reasonably comprehensive and stable, the
> > authors request WG adoption.
> >
> > --
> > Kireeti
> >
> > _______________________________________________
> > mpls mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/mpls
> >
>
> --
> Prof. Dr. habil. Michael Menth
> University of Tuebingen
> Faculty of Science
> Department of Computer Science
> Chair of Communication Networks
> Sand 13, 72076 Tuebingen, Germany
> phone: [phone]
> fax: [phone]
> mailto:[e-mail]
> http://kn.inf.uni-tuebingen.de
>


-- 
Kireeti