Re: [Idr] draft-keyupate-idr-bgp-prefix-sid-00

[Robert Raszuk <[e-mail]>]

Hi Jakob,

> Those labels are NOT advertised by RFC3107. The label index is used to derive
them.

It is not about what get's advertised with 3107 .. it is about using
distance vector (with policy and heavy ECMP) to build your data plane paths
which worries me.

- - -

> As far as SRGB is concerned, why would a data center operator
> not configure all routers with the same SRGB?

That is not the question to be asked.

The question is why would DC operator configure any SRGB at all.
For substantial arguments pls see:
http://www.ietf.org/mail-archive/web/spring/current/msg01188.html

Thx,
R.


On Sun, Nov 23, 2014 at 12:52 AM, Jakob Heitz (jheitz) <[e-mail]>
wrote:

> If only one segment is required to forward a packet, then
> we use regular 3107 and only L1 is needed. L2 is redundant.
>
> However...
>
> when we use multiple segments, for which the whole concept of
> segment routing is useful, the ingress router must know the labels
> used by subsequent segment routers in the segment list. Those
> labels are NOT advertised by RFC3107. The label index is used to
> derive them.
>
> L1 can be used to get to the nexthop at the end of the first
> segment, but to derive the subsequent labels in the segment list,
> the ingress router needs the label index and the SRGB of all the
> routers in the list. [Thanks to Albert Tian for pointing this
> out to me. Albert, correct me if I misinterpreted]
>
> As far as SRGB is concerned, why would a data center operator
> not configure all routers with the same SRGB?
>
> --Jakob
>
>
> > -----Original Message-----
> > From: Idr [mailto:[e-mail]] On Behalf Of Eric C Rosen
> > Sent: Friday, November 14, 2014 6:55 AM
> >
> > As Robert suggests, the draft is not very explicit about the
> > significance of the label that is encoded into the NLRI.  Suppose
> > one receives an update where the NLRI prefix is X, the label encoded
> > into the NLRI is L1, the next hop is N, and the label computed from
> > the combination of Label Index attribute and N's SRGB label base is
> > L2.  If one is sending a packet to X via N, the draft seems to leave
> > one with the option of using either L1 or L2 as the label.  This
> > means that N has to have both labels programmed for prefix X.
> >
> > Is this really the intention of the authors?  If so, then unless L1
> > is required to be the same as L2, this doubles the number of labels
> > that are needed.  On the other hand, one could require that L1 be
> > the same as L2.  But in this case, there doesn't seem to be any
> > point in having the Label Index attribute at all.
>
>