The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2001-Oct> msg00428



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Simplification for UNNUM draft.

  • From: "Kullberg, Alan" <akullber@netplane.com>
  • Date: Mon, 29 Oct 2001 17:05:48 -0500
  • Cc: mpls@UU.NET

Yakov & George,

I don't like this suggested change.  See reasoning below.

Alan

> -----Original Message-----
> From: Yakov Rekhter [mailto:yakov@juniper.net]
> Sent: Monday, October 29, 2001 11:36 AM
> To: George Swallow
> Cc: mpls@UU.NET
> Subject: Re: Simplification for UNNUM draft. 
> 
> 
> George,
> 
> > draft-ietf-mpls-rsvp-unnum-02.txt allows both the normal HOP and the
> > IF_ID Hop object to be used on an unnumbered link.  This leads to
> > unnecessary cases and extra checking at the incoming interface.

George's reasoning for making the following change is the above sentence.

> > I suggest the following modifications:
> > 
> > a) mandate that one always use the IF_ID HOP object on an unnumbered
> >    link.
> > 
> > b) pop the unnumbered interface subobject off of the ERO before
> >    forwarding the Path message over the unnumbered link (the
> >    information is now in the PHOP
> 
> May I also suggest that we change the semantics of the Unnumbered
> Interface ID subobject in the ERO from:
> 
>    The Interface ID is the outgoing interface identifier with 
> respect to
>    the LSR specified by the router ID.
> 
> to
> 
>    The Interface ID is the incoming interface identifier with 
> respect to
>    the LSR specified by the router ID.
> 
> Yakov.

> George replies:
>
> That would be fine.  In fact in that case, the upstream router would
> not have to pop off the subobject, since the next node would receive a
> an ero which has a valid first object.
>
> ...George

By not popping the subobject off, the receiver of the PATH must do some
"extra checking at the incoming interface" because the IF_ID must be
validated to refer to the same interface.  In this case, the IF_ID is
from the sender's perspective and the UI.interface_id is from the
receiver's perspective, which I think is worse than before for cross-
checking.

In addition, Yakov's change affects Explicit Label Control at the Egress
node.  Let's say the egress node is node B, and there is an unnumbered
interface from B to a node C.  OSPF is not running between B and C,
C doesn't have a Router ID, and B doesn't know what C's unnumbered ID
is.  This makes it impossible to use the ERO to specify the Explicit
Labels between B and C when that interface is unnumbered.